Merge branch 'main' into fix/vector-db-mandatory-provider-id

.github/actions/run-and-record-tests/action.yml

@@ -2,26 +2,28 @@ name: 'Run and Record Tests'
 description: 'Run integration tests and handle recording/artifact upload'
 
 inputs:
-  test-subdirs:
-    description: 'Comma-separated list of test subdirectories to run'
-    required: true
-  test-pattern:
-    description: 'Regex pattern to pass to pytest -k'
-    required: false
-    default: ''
   stack-config:
     description: 'Stack configuration to use'
     required: true
-  provider:
+  setup:
-    description: 'Provider to use for tests'
+    description: 'Setup to use for tests (e.g., ollama, gpt, vllm)'
-    required: true
+    required: false
+    default: ''
   inference-mode:
     description: 'Inference mode (record or replay)'
     required: true
-  run-vision-tests:
+  suite:
-    description: 'Whether to run vision tests'
+    description: 'Test suite to use: base, responses, vision, etc.'
     required: false
-    default: 'false'
+    default: ''
+  subdirs:
+    description: 'Comma-separated list of test subdirectories to run; overrides suite'
+    required: false
+    default: ''
+  pattern:
+    description: 'Regex pattern to pass to pytest -k'
+    required: false
+    default: ''
 
 runs:
   using: 'composite'
@@ -36,14 +38,23 @@ runs:
     - name: Run Integration Tests
       shell: bash
       run: |
-        uv run --no-sync ./scripts/integration-tests.sh \
+        SCRIPT_ARGS="--stack-config ${{ inputs.stack-config }} --inference-mode ${{ inputs.inference-mode }}"
-          --stack-config '${{ inputs.stack-config }}' \
+
-          --provider '${{ inputs.provider }}' \
+        # Add optional arguments only if they are provided
-          --test-subdirs '${{ inputs.test-subdirs }}' \
+        if [ -n '${{ inputs.setup }}' ]; then
-          --test-pattern '${{ inputs.test-pattern }}' \
+          SCRIPT_ARGS="$SCRIPT_ARGS --setup ${{ inputs.setup }}"
-          --inference-mode '${{ inputs.inference-mode }}' \
+        fi
-          ${{ inputs.run-vision-tests == 'true' && '--run-vision-tests' || '' }} \
+        if [ -n '${{ inputs.suite }}' ]; then
-          | tee pytest-${{ inputs.inference-mode }}.log
+          SCRIPT_ARGS="$SCRIPT_ARGS --suite ${{ inputs.suite }}"
+        fi
+        if [ -n '${{ inputs.subdirs }}' ]; then
+          SCRIPT_ARGS="$SCRIPT_ARGS --subdirs ${{ inputs.subdirs }}"
+        fi
+        if [ -n '${{ inputs.pattern }}' ]; then
+          SCRIPT_ARGS="$SCRIPT_ARGS --pattern ${{ inputs.pattern }}"
+        fi
+
+        uv run --no-sync ./scripts/integration-tests.sh $SCRIPT_ARGS | tee pytest-${{ inputs.inference-mode }}.log
 
 
     - name: Commit and push recordings
@@ -57,12 +68,7 @@ runs:
           echo "New recordings detected, committing and pushing"
           git add tests/integration/recordings/
 
-          if [ "${{ inputs.run-vision-tests }}" == "true" ]; then
+          git commit -m "Recordings update from CI (suite: ${{ inputs.suite }})"
-            git commit -m "Recordings update from CI (vision)"
-          else
-            git commit -m "Recordings update from CI"
-          fi
-
           git fetch origin ${{ github.ref_name }}
           git rebase origin/${{ github.ref_name }}
           echo "Rebased successfully"

.github/actions/setup-ollama/action.yml

@@ -1,17 +1,17 @@
 name: Setup Ollama
 description: Start Ollama
 inputs:
-  run-vision-tests:
+  suite:
-    description: 'Run vision tests: "true" or "false"'
+    description: 'Test suite to use: base, responses, vision, etc.'
     required: false
-    default: 'false'
+    default: ''
 runs:
   using: "composite"
   steps:
     - name: Start Ollama
       shell: bash
       run: |
-        if [ "${{ inputs.run-vision-tests }}" == "true" ]; then
+        if [ "${{ inputs.suite }}" == "vision" ]; then
           image="ollama-with-vision-model"
         else
           image="ollama-with-models"

.github/actions/setup-test-environment/action.yml

@@ -8,14 +8,14 @@ inputs:
   client-version:
     description: 'Client version (latest or published)'
     required: true
-  provider:
+  setup:
-    description: 'Provider to setup (ollama or vllm)'
+    description: 'Setup to configure (ollama, vllm, gpt, etc.)'
-    required: true
-    default: 'ollama'
-  run-vision-tests:
-    description: 'Whether to setup provider for vision tests'
     required: false
-    default: 'false'
+    default: 'ollama'
+  suite:
+    description: 'Test suite to use: base, responses, vision, etc.'
+    required: false
+    default: ''
   inference-mode:
     description: 'Inference mode (record or replay)'
     required: true
@@ -30,13 +30,13 @@ runs:
         client-version: ${{ inputs.client-version }}
 
     - name: Setup ollama
-      if: ${{ inputs.provider == 'ollama' && inputs.inference-mode == 'record' }}
+      if: ${{ (inputs.setup == 'ollama' || inputs.setup == 'ollama-vision') && inputs.inference-mode == 'record' }}
       uses: ./.github/actions/setup-ollama
       with:
-        run-vision-tests: ${{ inputs.run-vision-tests }}
+        suite: ${{ inputs.suite }}
 
     - name: Setup vllm
-      if: ${{ inputs.provider == 'vllm' && inputs.inference-mode == 'record' }}
+      if: ${{ inputs.setup == 'vllm' && inputs.inference-mode == 'record' }}
       uses: ./.github/actions/setup-vllm
 
     - name: Build Llama Stack

.github/workflows/README.md

@@ -5,10 +5,11 @@ Llama Stack uses GitHub Actions for Continuous Integration (CI). Below is a tabl
 | Name | File | Purpose |
 | ---- | ---- | ------- |
 | Update Changelog | [changelog.yml](changelog.yml) | Creates PR for updating the CHANGELOG.md |
+| API Conformance Tests | [conformance.yml](conformance.yml) | Run the API Conformance test suite on the changes. |
 | Installer CI | [install-script-ci.yml](install-script-ci.yml) | Test the installation script |
 | Integration Auth Tests | [integration-auth-tests.yml](integration-auth-tests.yml) | Run the integration test suite with Kubernetes authentication |
 | SqlStore Integration Tests | [integration-sql-store-tests.yml](integration-sql-store-tests.yml) | Run the integration test suite with SqlStore |
-| Integration Tests (Replay) | [integration-tests.yml](integration-tests.yml) | Run the integration test suite from tests/integration in replay mode |
+| Integration Tests (Replay) | [integration-tests.yml](integration-tests.yml) | Run the integration test suites from tests/integration in replay mode |
 | Vector IO Integration Tests | [integration-vector-io-tests.yml](integration-vector-io-tests.yml) | Run the integration test suite with various VectorIO providers |
 | Pre-commit | [pre-commit.yml](pre-commit.yml) | Run pre-commit checks |
 | Test Llama Stack Build | [providers-build.yml](providers-build.yml) | Test llama stack build |

.github/workflows/conformance.yml

@@ -0,0 +1,57 @@
+# API Conformance Tests
+# This workflow ensures that API changes maintain backward compatibility and don't break existing integrations
+# It runs schema validation and OpenAPI diff checks to catch breaking changes early
+
+name: API Conformance Tests
+
+run-name: Run the API Conformance test suite on the changes.
+
+on:
+  push:
+    branches: [ main ]
+  pull_request:
+    branches: [ main ]
+    types: [opened, synchronize, reopened]
+    paths:
+      - 'llama_stack/**'
+      - '!llama_stack/ui/**'
+      - 'tests/**'
+      - 'uv.lock'
+      - 'pyproject.toml'
+      - '.github/workflows/conformance.yml' # This workflow itself
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref == 'refs/heads/main' && github.run_id || github.ref }}
+  # Cancel in-progress runs when new commits are pushed to avoid wasting CI resources
+  cancel-in-progress: true
+
+jobs:
+  # Job to check if API schema changes maintain backward compatibility
+  check-schema-compatibility:
+    runs-on: ubuntu-latest
+    steps:
+      # Using specific version 4.1.7 because 5.0.0 fails when trying to run this locally using `act`
+      # This ensures consistent behavior between local testing and CI
+      - name: Checkout PR Code
+        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+
+      # Checkout the base branch to compare against (usually main)
+      # This allows us to diff the current changes against the previous state
+      - name: Checkout Base Branch
+        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        with:
+          ref: ${{ github.event.pull_request.base.ref }}
+          path: 'base'
+
+      # Install oasdiff: https://github.com/oasdiff/oasdiff, a tool for detecting breaking changes in OpenAPI specs.
+      - name: Install oasdiff
+        run: |
+          curl -fsSL https://raw.githubusercontent.com/oasdiff/oasdiff/main/install.sh | sh
+
+      # Run oasdiff to detect breaking changes in the API specification
+      # This step will fail if incompatible changes are detected, preventing breaking changes from being merged
+      - name: Run OpenAPI Breaking Change Diff
+        run: |
+          oasdiff breaking --fail-on ERR base/docs/_static/llama-stack-spec.yaml docs/_static/llama-stack-spec.yaml --match-path '^/v1/openai/v1' \
+          --match-path '^/v1/vector-io' \
+          --match-path '^/v1/vector-dbs'

.github/workflows/integration-tests.yml

@@ -1,6 +1,6 @@
 name: Integration Tests (Replay)
 
-run-name: Run the integration test suite from tests/integration in replay mode
+run-name: Run the integration test suites from tests/integration in replay mode
 
 on:
   push:
@@ -28,18 +28,10 @@ on:
         description: 'Test against both the latest and published versions'
         type: boolean
         default: false
-      test-provider:
+      test-setup:
-        description: 'Test against a specific provider'
+        description: 'Test against a specific setup'
         type: string
         default: 'ollama'
-      test-subdirs:
-        description: 'Comma-separated list of test subdirectories to run'
-        type: string
-        default: ''
-      test-pattern:
-        description: 'Regex pattern to pass to pytest -k'
-        type: string
-        default: ''
 
 concurrency:
   # Skip concurrency for pushes to main - each commit should be tested independently
@@ -50,18 +42,18 @@ jobs:
 
   run-replay-mode-tests:
     runs-on: ubuntu-latest
-    name: ${{ format('Integration Tests ({0}, {1}, {2}, client={3}, vision={4})', matrix.client-type, matrix.provider, matrix.python-version, matrix.client-version, matrix.run-vision-tests) }}
+    name: ${{ format('Integration Tests ({0}, {1}, {2}, client={3}, {4})', matrix.client-type, matrix.setup, matrix.python-version, matrix.client-version, matrix.suite) }}
 
     strategy:
       fail-fast: false
       matrix:
         client-type: [library, server]
-        # Use vllm on weekly schedule, otherwise use test-provider input (defaults to ollama)
+        # Use vllm on weekly schedule, otherwise use test-setup input (defaults to ollama)
-        provider: ${{ (github.event.schedule == '1 0 * * 0') && fromJSON('["vllm"]') || fromJSON(format('["{0}"]', github.event.inputs.test-provider || 'ollama')) }}
+        setup: ${{ (github.event.schedule == '1 0 * * 0') && fromJSON('["vllm"]') || fromJSON(format('["{0}"]', github.event.inputs.test-setup || 'ollama')) }}
         # Use Python 3.13 only on nightly schedule (daily latest client test), otherwise use 3.12
         python-version: ${{ github.event.schedule == '0 0 * * *' && fromJSON('["3.12", "3.13"]') || fromJSON('["3.12"]') }}
         client-version: ${{ (github.event.schedule == '0 0 * * *' || github.event.inputs.test-all-client-versions == 'true') && fromJSON('["published", "latest"]') || fromJSON('["latest"]') }}
-        run-vision-tests: [true, false]
+        suite: [base, vision]
 
     steps:
       - name: Checkout repository
@@ -72,16 +64,14 @@ jobs:
         with:
           python-version: ${{ matrix.python-version }}
           client-version: ${{ matrix.client-version }}
-          provider: ${{ matrix.provider }}
+          setup: ${{ matrix.setup }}
-          run-vision-tests: ${{ matrix.run-vision-tests }}
+          suite: ${{ matrix.suite }}
           inference-mode: 'replay'
 
       - name: Run tests
         uses: ./.github/actions/run-and-record-tests
         with:
-          test-subdirs: ${{ inputs.test-subdirs }}
-          test-pattern: ${{ inputs.test-pattern }}
           stack-config: ${{ matrix.client-type == 'library' && 'ci-tests' || 'server:ci-tests' }}
-          provider: ${{ matrix.provider }}
+          setup: ${{ matrix.setup }}
           inference-mode: 'replay'
-          run-vision-tests: ${{ matrix.run-vision-tests }}
+          suite: ${{ matrix.suite }}

.github/workflows/pre-commit.yml

@@ -28,7 +28,7 @@ jobs:
           fetch-depth: ${{ github.actor == 'dependabot[bot]' && 0 || 1 }}
 
       - name: Set up Python
-        uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
+        uses: actions/setup-python@e797f83bcb11b83ae66e0230d6156d7c80228e7c # v6.0.0
         with:
           python-version: '3.12'
           cache: pip
@@ -37,7 +37,7 @@ jobs:
             .pre-commit-config.yaml
 
       - name: Set up Node.js
-        uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
+        uses: actions/setup-node@a0853c24544627f65ddf259abe73b1d18a591444 # v5.0.0
         with:
           node-version: '20'
           cache: 'npm'
@@ -48,7 +48,6 @@ jobs:
         working-directory: llama_stack/ui
 
       - uses: pre-commit/action@2c7b3805fd2a0fd8c1884dcaebf91fc102a13ecd # v3.0.1
-        continue-on-error: true
         env:
           SKIP: no-commit-to-branch
           RUFF_OUTPUT_FORMAT: github

.github/workflows/python-build-test.yml

@@ -24,7 +24,7 @@ jobs:
       uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
 
     - name: Install uv
-      uses: astral-sh/setup-uv@4959332f0f014c5280e7eac8b70c90cb574c9f9b # v6.6.0
+      uses: astral-sh/setup-uv@557e51de59eb14aaaba2ed9621916900a91d50c6 # v6.6.1
       with:
         python-version: ${{ matrix.python-version }}
         activate-environment: true

.github/workflows/record-integration-tests.yml

@@ -10,19 +10,19 @@ run-name: Run the integration test suite from tests/integration
 on:
   workflow_dispatch:
     inputs:
-      test-subdirs:
+      test-setup:
-        description: 'Comma-separated list of test subdirectories to run'
+        description: 'Test against a specific setup'
-        type: string
-        default: ''
-      test-provider:
-        description: 'Test against a specific provider'
         type: string
         default: 'ollama'
-      run-vision-tests:
+      suite:
-        description: 'Whether to run vision tests'
+        description: 'Test suite to use: base, responses, vision, etc.'
-        type: boolean
+        type: string
-        default: false
+        default: ''
-      test-pattern:
+      subdirs:
+        description: 'Comma-separated list of test subdirectories to run; overrides suite'
+        type: string
+        default: ''
+      pattern:
         description: 'Regex pattern to pass to pytest -k'
         type: string
         default: ''
@@ -38,11 +38,11 @@ jobs:
       - name: Echo workflow inputs
         run: |
           echo "::group::Workflow Inputs"
-          echo "test-subdirs: ${{ inputs.test-subdirs }}"
-          echo "test-provider: ${{ inputs.test-provider }}"
-          echo "run-vision-tests: ${{ inputs.run-vision-tests }}"
-          echo "test-pattern: ${{ inputs.test-pattern }}"
           echo "branch: ${{ github.ref_name }}"
+          echo "test-setup: ${{ inputs.test-setup }}"
+          echo "suite: ${{ inputs.suite }}"
+          echo "subdirs: ${{ inputs.subdirs }}"
+          echo "pattern: ${{ inputs.pattern }}"
           echo "::endgroup::"
 
       - name: Checkout repository
@@ -55,16 +55,16 @@ jobs:
         with:
           python-version: "3.12"  # Use single Python version for recording
           client-version: "latest"
-          provider: ${{ inputs.test-provider || 'ollama' }}
+          setup: ${{ inputs.test-setup || 'ollama' }}
-          run-vision-tests: ${{ inputs.run-vision-tests }}
+          suite: ${{ inputs.suite }}
           inference-mode: 'record'
 
       - name: Run and record tests
         uses: ./.github/actions/run-and-record-tests
         with:
-          test-pattern: ${{ inputs.test-pattern }}
-          test-subdirs: ${{ inputs.test-subdirs }}
           stack-config: 'server:ci-tests'  # recording must be done with server since more tests are run
-          provider: ${{ inputs.test-provider || 'ollama' }}
+          setup: ${{ inputs.test-setup || 'ollama' }}
           inference-mode: 'record'
-          run-vision-tests: ${{ inputs.run-vision-tests }}
+          suite: ${{ inputs.suite }}
+          subdirs: ${{ inputs.subdirs }}
+          pattern: ${{ inputs.pattern }}

.github/workflows/stale_bot.yml

@@ -24,7 +24,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Stale Action
-        uses: actions/stale@5bef64f19d7facfb25b37b414482c7164d639639 # v9.1.0
+        uses: actions/stale@3a9db7e6a41a89f618792c92c0e97cc736e1b13f # v10.0.0
         with:
           stale-issue-label: 'stale'
           stale-issue-message: >

.github/workflows/ui-unit-tests.yml

@@ -29,7 +29,7 @@ jobs:
         uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
 
       - name: Setup Node.js
-        uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
+        uses: actions/setup-node@a0853c24544627f65ddf259abe73b1d18a591444 # v5.0.0
         with:
           node-version: ${{ matrix.node-version }}
           cache: 'npm'

.gitignore

@@ -26,5 +26,7 @@ venv/
 pytest-report.xml
 .coverage
 .python-version
+AGENTS.md
+server.log
 CLAUDE.md
 .claude/

.pre-commit-config.yaml

@@ -86,7 +86,7 @@ repos:
         language: python
         pass_filenames: false
         require_serial: true
-        files: ^llama_stack/templates/.*$|^llama_stack/providers/.*/inference/.*/models\.py$
+        files: ^llama_stack/distributions/.*$|^llama_stack/providers/.*/inference/.*/models\.py$
       - id: provider-codegen
         name: Provider Codegen
         additional_dependencies:

CHANGELOG.md

@@ -1,5 +1,103 @@
 # Changelog
 
+# v0.2.20
+Published on: 2025-08-29T22:25:32Z
+
+Here are some key changes that are coming as part of this release.
+
+### Build and Environment
+
+- Environment improvements: fixed env var replacement to preserve types.
+- Docker stability: fixed container startup failures for Fireworks AI provider.
+- Removed absolute paths in build for better portability.
+
+### Features
+
+- UI Enhancements: Implemented file upload and VectorDB creation/configuration directly in UI.
+- Vector Store Improvements: Added keyword, vector, and hybrid search inside vector store.
+- Added S3 authorization support for file providers.
+- SQL Store: Added inequality support to where clause.
+
+### Documentation
+
+- Fixed post-training docs.
+- Added Contributor Guidelines for creating Internal vs. External providers.
+
+### Fixes
+
+- Removed unsupported bfcl scoring function.
+- Multiple reliability and configuration fixes for providers and environment handling.
+
+### Engineering / Chores
+
+- Cleaner internal development setup with consistent paths.
+- Incremental improvements to provider integration and vector store behavior.
+
+
+### New Contributors
+- @omertuc made their first contribution in #3270
+- @r3v5 made their first contribution in vector store hybrid search
+
+---
+
+# v0.2.19
+Published on: 2025-08-26T22:06:55Z
+
+## Highlights
+* feat: Add CORS configuration support for server by @skamenan7 in https://github.com/llamastack/llama-stack/pull/3201
+* feat(api): introduce /rerank by @ehhuang in https://github.com/llamastack/llama-stack/pull/2940
+* feat: Add S3 Files Provider by @mattf in https://github.com/llamastack/llama-stack/pull/3202
+
+
+---
+
+# v0.2.18
+Published on: 2025-08-20T01:09:27Z
+
+## Highlights
+* Add moderations create API
+* Hybrid search in Milvus
+* Numerous Responses API improvements
+* Documentation updates
+
+
+---
+
+# v0.2.17
+Published on: 2025-08-05T01:51:14Z
+
+## Highlights
+
+* feat(tests): introduce inference record/replay to increase test reliability by @ashwinb in https://github.com/meta-llama/llama-stack/pull/2941
+* fix(library_client): improve initialization error handling and prevent AttributeError by @mattf in https://github.com/meta-llama/llama-stack/pull/2944
+* fix: use OLLAMA_URL to activate Ollama provider in starter by @ashwinb in https://github.com/meta-llama/llama-stack/pull/2963
+* feat(UI): adding MVP playground UI by @franciscojavierarceo in https://github.com/meta-llama/llama-stack/pull/2828
+* Standardization of errors (@nathan-weinberg)
+* feat: Enable DPO training with HuggingFace inline provider by @Nehanth in https://github.com/meta-llama/llama-stack/pull/2825
+* chore: rename templates to distributions by @ashwinb in https://github.com/meta-llama/llama-stack/pull/3035
+
+
+---
+
+# v0.2.16
+Published on: 2025-07-28T23:35:23Z
+
+## Highlights
+
+* Automatic model registration for self-hosted providers (ollama and vllm currently). No need for `INFERENCE_MODEL` environment variables which need to be updated, etc.
+* Much simplified starter distribution. Most `ENABLE_` env variables are now gone. When you set `VLLM_URL`, the `vllm` provider is auto-enabled. Similar for `MILVUS_URL`, `PGVECTOR_DB`, etc. Check the [run.yaml](https://github.com/meta-llama/llama-stack/blob/main/llama_stack/templates/starter/run.yaml) for more details.
+* All tests migrated to pytest now (thanks @Elbehery)
+* DPO implementation in the post-training provider (thanks @Nehanth)
+* (Huge!) Support for external APIs and providers thereof (thanks @leseb, @cdoern and others). This is a really big deal -- you can now add more APIs completely out of tree and experiment with them before (optionally) wanting to contribute back.
+* `inline::vllm` provider is gone thank you very much
+* several improvements to OpenAI inference implementations and LiteLLM backend (thanks @mattf)
+* Chroma now supports Vector Store API (thanks @franciscojavierarceo).
+* Authorization improvements: Vector Store/File APIs now supports access control (thanks @franciscojavierarceo); Telemetry read APIs are gated according to logged-in user's roles.
+
+
+
+---
+
 # v0.2.15
 Published on: 2025-07-16T03:30:01Z
 

benchmarking/k8s-benchmark/README.md

@@ -34,13 +34,12 @@ This data enables data-driven architectural decisions and performance optimizati
 
 **1. Deploy base k8s infrastructure:**
 ```bash
-cd ../k8s
+cd ../../docs/source/distributions/k8s
 ./apply.sh
 ```
 
 **2. Deploy benchmark components:**
 ```bash
-cd ../k8s-benchmark
 ./apply.sh
 ```
 
@@ -56,7 +55,6 @@ kubectl get pods
 
 **Benchmark Llama Stack (default):**
 ```bash
-cd docs/source/distributions/k8s-benchmark/
 ./run-benchmark.sh
 ```
 

benchmarking/k8s-benchmark/benchmark.py

@@ -14,7 +14,7 @@ import os
 import random
 import statistics
 import time
-from typing import Tuple
+
 import aiohttp
 
 
@@ -57,17 +57,9 @@ class BenchmarkStats:
         success_rate = (self.success_count / self.total_requests) * 100
 
         print(f"\n{'=' * 60}")
-        print(f"BENCHMARK RESULTS")
+        print("BENCHMARK RESULTS")
-        print(f"{'='*60}")
-        print(f"Total time: {total_time:.2f}s")
-        print(f"Concurrent users: {self.concurrent_users}")
-        print(f"Total requests: {self.total_requests}")
-        print(f"Successful requests: {self.success_count}")
-        print(f"Failed requests: {len(self.errors)}")
-        print(f"Success rate: {success_rate:.1f}%")
-        print(f"Requests per second: {self.success_count / total_time:.2f}")
 
-        print(f"\nResponse Time Statistics:")
+        print("\nResponse Time Statistics:")
         print(f"  Mean: {statistics.mean(self.response_times):.3f}s")
         print(f"  Median: {statistics.median(self.response_times):.3f}s")
         print(f"  Min: {min(self.response_times):.3f}s")
@@ -78,14 +70,14 @@ class BenchmarkStats:
 
         percentiles = [50, 90, 95, 99]
         sorted_times = sorted(self.response_times)
-        print(f"\nPercentiles:")
+        print("\nPercentiles:")
         for p in percentiles:
             idx = int(len(sorted_times) * p / 100) - 1
             idx = max(0, min(idx, len(sorted_times) - 1))
             print(f"  P{p}: {sorted_times[idx]:.3f}s")
 
         if self.ttft_times:
-            print(f"\nTime to First Token (TTFT) Statistics:")
+            print("\nTime to First Token (TTFT) Statistics:")
             print(f"  Mean: {statistics.mean(self.ttft_times):.3f}s")
             print(f"  Median: {statistics.median(self.ttft_times):.3f}s")
             print(f"  Min: {min(self.ttft_times):.3f}s")
@@ -95,26 +87,35 @@ class BenchmarkStats:
                 print(f"  Std Dev: {statistics.stdev(self.ttft_times):.3f}s")
 
             sorted_ttft = sorted(self.ttft_times)
-            print(f"\nTTFT Percentiles:")
+            print("\nTTFT Percentiles:")
             for p in percentiles:
                 idx = int(len(sorted_ttft) * p / 100) - 1
                 idx = max(0, min(idx, len(sorted_ttft) - 1))
                 print(f"  P{p}: {sorted_ttft[idx]:.3f}s")
 
         if self.chunks_received:
-            print(f"\nStreaming Statistics:")
+            print("\nStreaming Statistics:")
             print(f"  Mean chunks per response: {statistics.mean(self.chunks_received):.1f}")
             print(f"  Total chunks received: {sum(self.chunks_received)}")
 
+        print(f"{'=' * 60}")
+        print(f"Total time: {total_time:.2f}s")
+        print(f"Concurrent users: {self.concurrent_users}")
+        print(f"Total requests: {self.total_requests}")
+        print(f"Successful requests: {self.success_count}")
+        print(f"Failed requests: {len(self.errors)}")
+        print(f"Success rate: {success_rate:.1f}%")
+        print(f"Requests per second: {self.success_count / total_time:.2f}")
+
         if self.errors:
-            print(f"\nErrors (showing first 5):")
+            print("\nErrors (showing first 5):")
             for error in self.errors[:5]:
                 print(f"  {error}")
 
 
 class LlamaStackBenchmark:
     def __init__(self, base_url: str, model_id: str):
-        self.base_url = base_url.rstrip('/')
+        self.base_url = base_url.rstrip("/")
         self.model_id = model_id
         self.headers = {"Content-Type": "application/json"}
         self.test_messages = [
@@ -125,20 +126,14 @@ class LlamaStackBenchmark:
             [
                 {"role": "user", "content": "What is machine learning?"},
                 {"role": "assistant", "content": "Machine learning is a subset of AI..."},
-                {"role": "user", "content": "Can you give me a practical example?"}
+                {"role": "user", "content": "Can you give me a practical example?"},
-            ]
+            ],
         ]
 
-
+    async def make_async_streaming_request(self) -> tuple[float, int, float | None, str | None]:
-    async def make_async_streaming_request(self) -> Tuple[float, int, float | None, str | None]:
         """Make a single async streaming chat completion request."""
         messages = random.choice(self.test_messages)
-        payload = {
+        payload = {"model": self.model_id, "messages": messages, "stream": True, "max_tokens": 100}
-            "model": self.model_id,
-            "messages": messages,
-            "stream": True,
-            "max_tokens": 100
-        }
 
         start_time = time.time()
         chunks_received = 0
@@ -152,17 +147,17 @@ class LlamaStackBenchmark:
                 f"{self.base_url}/chat/completions",
                 headers=self.headers,
                 json=payload,
-                timeout=aiohttp.ClientTimeout(total=30)
+                timeout=aiohttp.ClientTimeout(total=30),
             ) as response:
                 if response.status == 200:
                     async for line in response.content:
                         if line:
-                            line_str = line.decode('utf-8').strip()
+                            line_str = line.decode("utf-8").strip()
-                            if line_str.startswith('data: '):
+                            if line_str.startswith("data: "):
                                 chunks_received += 1
                                 if ttft is None:
                                     ttft = time.time() - start_time
-                                if line_str == 'data: [DONE]':
+                                if line_str == "data: [DONE]":
                                     break
 
                     if chunks_received == 0:
@@ -179,7 +174,6 @@ class LlamaStackBenchmark:
         response_time = time.time() - start_time
         return response_time, chunks_received, ttft, error
 
-
     async def run_benchmark(self, duration: int, concurrent_users: int) -> BenchmarkStats:
         """Run benchmark using async requests for specified duration."""
         stats = BenchmarkStats()
@@ -191,7 +185,7 @@ class LlamaStackBenchmark:
         print(f"Model: {self.model_id}")
 
         connector = aiohttp.TCPConnector(limit=concurrent_users)
-        async with aiohttp.ClientSession(connector=connector) as session:
+        async with aiohttp.ClientSession(connector=connector):
 
             async def worker(worker_id: int):
                 """Worker that sends requests sequentially until canceled."""
@@ -215,7 +209,9 @@ class LlamaStackBenchmark:
                         await asyncio.sleep(1)  # Report every second
                         if time.time() >= last_report_time + 10:  # Report every 10 seconds
                             elapsed = time.time() - stats.start_time
-                            print(f"Completed: {stats.total_requests} requests in {elapsed:.1f}s")
+                            print(
+                                f"Completed: {stats.total_requests} requests in {elapsed:.1f}s, RPS: {stats.total_requests / elapsed:.1f}"
+                            )
                             last_report_time = time.time()
                     except asyncio.CancelledError:
                         break
@@ -240,14 +236,16 @@ class LlamaStackBenchmark:
 
 def main():
     parser = argparse.ArgumentParser(description="Llama Stack Benchmark Tool")
-    parser.add_argument("--base-url", default=os.getenv("BENCHMARK_BASE_URL", "http://localhost:8000/v1/openai/v1"),
+    parser.add_argument(
-                       help="Base URL for the API (default: http://localhost:8000/v1/openai/v1)")
+        "--base-url",
-    parser.add_argument("--model", default=os.getenv("INFERENCE_MODEL", "test-model"),
+        default=os.getenv("BENCHMARK_BASE_URL", "http://localhost:8000/v1/openai/v1"),
-                       help="Model ID to use for requests")
+        help="Base URL for the API (default: http://localhost:8000/v1/openai/v1)",
-    parser.add_argument("--duration", type=int, default=60,
+    )
-                       help="Duration in seconds to run benchmark (default: 60)")
+    parser.add_argument(
-    parser.add_argument("--concurrent", type=int, default=10,
+        "--model", default=os.getenv("INFERENCE_MODEL", "test-model"), help="Model ID to use for requests"
-                       help="Number of concurrent users (default: 10)")
+    )
+    parser.add_argument("--duration", type=int, default=60, help="Duration in seconds to run benchmark (default: 60)")
+    parser.add_argument("--concurrent", type=int, default=10, help="Number of concurrent users (default: 10)")
 
     args = parser.parse_args()
 

benchmarking/k8s-benchmark/openai-mock-server.py

@@ -11,16 +11,18 @@ OpenAI-compatible mock server that returns:
 - Valid OpenAI-formatted chat completion responses with dynamic content
 """
 
-from flask import Flask, request, jsonify, Response
-import time
-import random
-import uuid
-import json
 import argparse
+import json
 import os
+import random
+import time
+import uuid
+
+from flask import Flask, Response, jsonify, request
 
 app = Flask(__name__)
 
+
 # Models from environment variables
 def get_models():
     models_str = os.getenv("MOCK_MODELS", "meta-llama/Llama-3.2-3B-Instruct")
@@ -29,40 +31,72 @@ def get_models():
     return {
         "object": "list",
         "data": [
-            {
+            {"id": model_id, "object": "model", "created": 1234567890, "owned_by": "vllm"} for model_id in model_ids
-                "id": model_id,
+        ],
-                "object": "model",
-                "created": 1234567890,
-                "owned_by": "vllm"
-            }
-            for model_id in model_ids
-        ]
     }
 
+
 def generate_random_text(length=50):
     """Generate random but coherent text for responses."""
     words = [
-        "Hello", "there", "I'm", "an", "AI", "assistant", "ready", "to", "help", "you",
+        "Hello",
-        "with", "your", "questions", "and", "tasks", "today", "Let", "me","know", "what",
+        "there",
-        "you'd", "like", "to", "discuss", "or", "explore", "together", "I", "can", "assist",
+        "I'm",
-        "with", "various", "topics", "including", "coding", "writing", "analysis", "and", "more"
+        "an",
+        "AI",
+        "assistant",
+        "ready",
+        "to",
+        "help",
+        "you",
+        "with",
+        "your",
+        "questions",
+        "and",
+        "tasks",
+        "today",
+        "Let",
+        "me",
+        "know",
+        "what",
+        "you'd",
+        "like",
+        "to",
+        "discuss",
+        "or",
+        "explore",
+        "together",
+        "I",
+        "can",
+        "assist",
+        "with",
+        "various",
+        "topics",
+        "including",
+        "coding",
+        "writing",
+        "analysis",
+        "and",
+        "more",
     ]
     return " ".join(random.choices(words, k=length))
 
-@app.route('/v1/models', methods=['GET'])
+
+@app.route("/v1/models", methods=["GET"])
 def list_models():
     models = get_models()
     print(f"[MOCK] Returning models: {[m['id'] for m in models['data']]}")
     return jsonify(models)
 
-@app.route('/v1/chat/completions', methods=['POST'])
+
+@app.route("/v1/chat/completions", methods=["POST"])
 def chat_completions():
     """Return OpenAI-formatted chat completion responses."""
     data = request.get_json()
-    default_model = get_models()['data'][0]['id']
+    default_model = get_models()["data"][0]["id"]
-    model = data.get('model', default_model)
+    model = data.get("model", default_model)
-    messages = data.get('messages', [])
+    messages = data.get("messages", [])
-    stream = data.get('stream', False)
+    stream = data.get("stream", False)
 
     print(f"[MOCK] Chat completion request - model: {model}, stream: {stream}")
 
@@ -71,11 +105,12 @@ def chat_completions():
     else:
         return handle_non_streaming_completion(model, messages)
 
+
 def handle_non_streaming_completion(model, messages):
     response_text = generate_random_text(random.randint(20, 80))
 
     # Calculate realistic token counts
-    prompt_tokens = sum(len(str(msg.get('content', '')).split()) for msg in messages)
+    prompt_tokens = sum(len(str(msg.get("content", "")).split()) for msg in messages)
     completion_tokens = len(response_text.split())
 
     response = {
@@ -83,25 +118,17 @@ def handle_non_streaming_completion(model, messages):
         "object": "chat.completion",
         "created": int(time.time()),
         "model": model,
-        "choices": [
+        "choices": [{"index": 0, "message": {"role": "assistant", "content": response_text}, "finish_reason": "stop"}],
-            {
-                "index": 0,
-                "message": {
-                    "role": "assistant",
-                    "content": response_text
-                },
-                "finish_reason": "stop"
-            }
-        ],
         "usage": {
             "prompt_tokens": prompt_tokens,
             "completion_tokens": completion_tokens,
-            "total_tokens": prompt_tokens + completion_tokens
+            "total_tokens": prompt_tokens + completion_tokens,
-        }
+        },
     }
 
     return jsonify(response)
 
+
 def handle_streaming_completion(model, messages):
     def generate_stream():
         # Generate response text
@@ -114,12 +141,7 @@ def handle_streaming_completion(model, messages):
             "object": "chat.completion.chunk",
             "created": int(time.time()),
             "model": model,
-            "choices": [
+            "choices": [{"index": 0, "delta": {"role": "assistant", "content": ""}}],
-                {
-                    "index": 0,
-                    "delta": {"role": "assistant", "content": ""}
-                }
-            ]
         }
         yield f"data: {json.dumps(initial_chunk)}\n\n"
 
@@ -130,12 +152,7 @@ def handle_streaming_completion(model, messages):
                 "object": "chat.completion.chunk",
                 "created": int(time.time()),
                 "model": model,
-                "choices": [
+                "choices": [{"index": 0, "delta": {"content": f"{word} " if i < len(words) - 1 else word}}],
-                    {
-                        "index": 0,
-                        "delta": {"content": f"{word} " if i < len(words) - 1 else word}
-                    }
-                ]
             }
             yield f"data: {json.dumps(chunk)}\n\n"
             # Configurable delay to simulate realistic streaming
@@ -148,35 +165,30 @@ def handle_streaming_completion(model, messages):
             "object": "chat.completion.chunk",
             "created": int(time.time()),
             "model": model,
-            "choices": [
+            "choices": [{"index": 0, "delta": {"content": ""}, "finish_reason": "stop"}],
-                {
-                    "index": 0,
-                    "delta": {"content": ""},
-                    "finish_reason": "stop"
-                }
-            ]
         }
         yield f"data: {json.dumps(final_chunk)}\n\n"
         yield "data: [DONE]\n\n"
 
     return Response(
         generate_stream(),
-        mimetype='text/event-stream',
+        mimetype="text/event-stream",
         headers={
-            'Cache-Control': 'no-cache',
+            "Cache-Control": "no-cache",
-            'Connection': 'keep-alive',
+            "Connection": "keep-alive",
-            'Access-Control-Allow-Origin': '*',
+            "Access-Control-Allow-Origin": "*",
-        }
+        },
     )
 
-@app.route('/health', methods=['GET'])
+
+@app.route("/health", methods=["GET"])
 def health():
     return jsonify({"status": "healthy", "type": "openai-mock"})
 
-if __name__ == '__main__':
+
-    parser = argparse.ArgumentParser(description='OpenAI-compatible mock server')
+if __name__ == "__main__":
-    parser.add_argument('--port', type=int, default=8081, 
+    parser = argparse.ArgumentParser(description="OpenAI-compatible mock server")
-                       help='Port to run the server on (default: 8081)')
+    parser.add_argument("--port", type=int, default=8081, help="Port to run the server on (default: 8081)")
     args = parser.parse_args()
 
     port = args.port
@@ -187,4 +199,4 @@ if __name__ == '__main__':
     print("- OpenAI-formatted chat/completion responses with dynamic content")
     print("- Streaming support with valid SSE format")
     print(f"- Listening on: http://0.0.0.0:{port}")
-    app.run(host='0.0.0.0', port=port, debug=False)
+    app.run(host="0.0.0.0", port=port, debug=False)

benchmarking/k8s-benchmark/stack-configmap.yaml

@@ -6,6 +6,7 @@ data:
     apis:
     - agents
     - inference
+    - files
     - safety
     - telemetry
     - tool_runtime
@@ -19,13 +20,6 @@ data:
           max_tokens: ${env.VLLM_MAX_TOKENS:=4096}
           api_token: ${env.VLLM_API_TOKEN:=fake}
           tls_verify: ${env.VLLM_TLS_VERIFY:=true}
-      - provider_id: vllm-safety
-        provider_type: remote::vllm
-        config:
-          url: ${env.VLLM_SAFETY_URL:=http://localhost:8000/v1}
-          max_tokens: ${env.VLLM_MAX_TOKENS:=4096}
-          api_token: ${env.VLLM_API_TOKEN:=fake}
-          tls_verify: ${env.VLLM_TLS_VERIFY:=true}
       - provider_id: sentence-transformers
         provider_type: inline::sentence-transformers
         config: {}
@@ -41,6 +35,14 @@ data:
             db: ${env.POSTGRES_DB:=llamastack}
             user: ${env.POSTGRES_USER:=llamastack}
             password: ${env.POSTGRES_PASSWORD:=llamastack}
+      files:
+      - provider_id: meta-reference-files
+        provider_type: inline::localfs
+        config:
+          storage_dir: ${env.FILES_STORAGE_DIR:=~/.llama/distributions/starter/files}
+          metadata_store:
+            type: sqlite
+            db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/files_metadata.db
       safety:
       - provider_id: llama-guard
         provider_type: inline::llama-guard
@@ -111,9 +113,6 @@ data:
     - model_id: ${env.INFERENCE_MODEL}
       provider_id: vllm-inference
       model_type: llm
-    - model_id: ${env.SAFETY_MODEL}
-      provider_id: vllm-safety
-      model_type: llm
     shields:
     - shield_id: ${env.SAFETY_MODEL:=meta-llama/Llama-Guard-3-1B}
     vector_dbs: []

benchmarking/k8s-benchmark/stack_run_config.yaml

@@ -2,7 +2,10 @@ version: '2'
 image_name: kubernetes-benchmark-demo
 apis:
 - agents
+- files
 - inference
+- files
+- safety
 - telemetry
 - tool_runtime
 - vector_io
@@ -18,6 +21,14 @@ providers:
   - provider_id: sentence-transformers
     provider_type: inline::sentence-transformers
     config: {}
+  files:
+  - provider_id: meta-reference-files
+    provider_type: inline::localfs
+    config:
+      storage_dir: ${env.FILES_STORAGE_DIR:=~/.llama/distributions/starter/files}
+      metadata_store:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/files_metadata.db
   vector_io:
   - provider_id: ${env.ENABLE_CHROMADB:+chromadb}
     provider_type: remote::chromadb
@@ -30,6 +41,19 @@ providers:
         db: ${env.POSTGRES_DB:=llamastack}
         user: ${env.POSTGRES_USER:=llamastack}
         password: ${env.POSTGRES_PASSWORD:=llamastack}
+  files:
+  - provider_id: meta-reference-files
+    provider_type: inline::localfs
+    config:
+      storage_dir: ${env.FILES_STORAGE_DIR:=~/.llama/distributions/starter/files}
+      metadata_store:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/files_metadata.db
+  safety:
+  - provider_id: llama-guard
+    provider_type: inline::llama-guard
+    config:
+      excluded_categories: []
   agents:
   - provider_id: meta-reference
     provider_type: inline::meta-reference
@@ -95,6 +119,8 @@ models:
 - model_id: ${env.INFERENCE_MODEL}
   provider_id: vllm-inference
   model_type: llm
+shields:
+- shield_id: ${env.SAFETY_MODEL:=meta-llama/Llama-Guard-3-1B}
 vector_dbs: []
 datasets: []
 scoring_fns: []

docs/_static/css/my_theme.css

@@ -1,5 +1,106 @@
 @import url("theme.css");
 
+/* Horizontal Navigation Bar */
+.horizontal-nav {
+    background-color: #ffffff;
+    border-bottom: 1px solid #e5e5e5;
+    padding: 0;
+    position: fixed;
+    top: 0;
+    left: 0;
+    right: 0;
+    z-index: 1050;
+    height: 50px;
+    box-shadow: 0 2px 4px rgba(0,0,0,0.1);
+}
+
+[data-theme="dark"] .horizontal-nav {
+    background-color: #1a1a1a;
+    border-bottom: 1px solid #333;
+}
+
+.horizontal-nav .nav-container {
+    max-width: 1200px;
+    margin: 0 auto;
+    display: flex;
+    align-items: center;
+    justify-content: space-between;
+    padding: 0 20px;
+    height: 100%;
+}
+
+.horizontal-nav .nav-brand {
+    font-size: 18px;
+    font-weight: 600;
+    color: #333;
+    text-decoration: none;
+}
+
+[data-theme="dark"] .horizontal-nav .nav-brand {
+    color: #fff;
+}
+
+.horizontal-nav .nav-links {
+    display: flex;
+    align-items: center;
+    gap: 30px;
+    list-style: none;
+    margin: 0;
+    padding: 0;
+}
+
+.horizontal-nav .nav-links a {
+    color: #666;
+    text-decoration: none;
+    font-size: 14px;
+    font-weight: 500;
+    padding: 8px 12px;
+    border-radius: 6px;
+    transition: all 0.2s ease;
+}
+
+.horizontal-nav .nav-links a:hover,
+.horizontal-nav .nav-links a.active {
+    color: #333;
+    background-color: #f5f5f5;
+}
+
+.horizontal-nav .nav-links a.active {
+    font-weight: 600;
+}
+
+[data-theme="dark"] .horizontal-nav .nav-links a {
+    color: #ccc;
+}
+
+[data-theme="dark"] .horizontal-nav .nav-links a:hover,
+[data-theme="dark"] .horizontal-nav .nav-links a.active {
+    color: #fff;
+    background-color: #333;
+}
+
+.horizontal-nav .nav-links .github-link {
+    display: flex;
+    align-items: center;
+    gap: 6px;
+}
+
+.horizontal-nav .nav-links .github-icon {
+    width: 16px;
+    height: 16px;
+    fill: currentColor;
+}
+
+/* Adjust main content to account for fixed nav */
+.wy-nav-side {
+    top: 50px;
+    height: calc(100vh - 50px);
+}
+
+.wy-nav-content-wrap {
+    margin-top: 50px;
+}
+
 .wy-nav-content {
     max-width: 90%;
 }

docs/_static/js/horizontal_nav.js

@@ -0,0 +1,44 @@
+// Horizontal Navigation Bar for Llama Stack Documentation
+document.addEventListener('DOMContentLoaded', function() {
+    // Create the horizontal navigation HTML
+    const navHTML = `
+        <nav class="horizontal-nav">
+            <div class="nav-container">
+                <a href="/" class="nav-brand">Llama Stack</a>
+                <ul class="nav-links">
+                    <li><a href="/">Docs</a></li>
+                    <li><a href="/references/api_reference/">API Reference</a></li>
+                    <li><a href="https://github.com/meta-llama/llama-stack" target="_blank" class="github-link">
+                        <svg class="github-icon" viewBox="0 0 16 16" aria-hidden="true">
+                            <path d="M8 0C3.58 0 0 3.58 0 8c0 3.54 2.29 6.53 5.47 7.59.4.07.55-.17.55-.38 0-.19-.01-.82-.01-1.49-2.01.37-2.53-.49-2.69-.94-.09-.23-.48-.94-.82-1.13-.28-.15-.68-.52-.01-.53.63-.01 1.08.58 1.23.82.72 1.21 1.87.87 2.33.66.07-.52.28-.87.51-1.07-1.78-.2-3.64-.89-3.64-3.95 0-.87.31-1.59.82-2.15-.08-.2-.36-1.02.08-2.12 0 0 .67-.21 2.2.82.64-.18 1.32-.27 2-.27.68 0 1.36.09 2 .27 1.53-1.04 2.2-.82 2.2-.82.44 1.1.16 1.92.08 2.12.51.56.82 1.27.82 2.15 0 3.07-1.87 3.75-3.65 3.95.29.25.54.73.54 1.48 0 1.07-.01 1.93-.01 2.2 0 .21.15.46.55.38A8.013 8.013 0 0016 8c0-4.42-3.58-8-8-8z"/>
+                        </svg>
+                        GitHub
+                    </a></li>
+                </ul>
+            </div>
+        </nav>
+    `;
+
+    // Insert the navigation at the beginning of the body
+    document.body.insertAdjacentHTML('afterbegin', navHTML);
+
+    // Update navigation links based on current page
+    updateActiveNav();
+});
+
+function updateActiveNav() {
+    const currentPath = window.location.pathname;
+    const navLinks = document.querySelectorAll('.horizontal-nav .nav-links a');
+
+    navLinks.forEach(link => {
+        // Remove any existing active classes
+        link.classList.remove('active');
+
+        // Add active class based on current path
+        if (currentPath === '/' && link.getAttribute('href') === '/') {
+            link.classList.add('active');
+        } else if (currentPath.includes('/references/api_reference/') && link.getAttribute('href').includes('api_reference')) {
+            link.classList.add('active');
+        }
+    });
+}

docs/_static/llama-stack-spec.html

@@ -633,6 +633,80 @@
                 }
             }
         },
+        "/v1/prompts": {
+            "get": {
+                "responses": {
+                    "200": {
+                        "description": "A ListPromptsResponse containing all prompts.",
+                        "content": {
+                            "application/json": {
+                                "schema": {
+                                    "$ref": "#/components/schemas/ListPromptsResponse"
+                                }
+                            }
+                        }
+                    },
+                    "400": {
+                        "$ref": "#/components/responses/BadRequest400"
+                    },
+                    "429": {
+                        "$ref": "#/components/responses/TooManyRequests429"
+                    },
+                    "500": {
+                        "$ref": "#/components/responses/InternalServerError500"
+                    },
+                    "default": {
+                        "$ref": "#/components/responses/DefaultError"
+                    }
+                },
+                "tags": [
+                    "Prompts"
+                ],
+                "description": "List all prompts.",
+                "parameters": []
+            },
+            "post": {
+                "responses": {
+                    "200": {
+                        "description": "The created Prompt resource.",
+                        "content": {
+                            "application/json": {
+                                "schema": {
+                                    "$ref": "#/components/schemas/Prompt"
+                                }
+                            }
+                        }
+                    },
+                    "400": {
+                        "$ref": "#/components/responses/BadRequest400"
+                    },
+                    "429": {
+                        "$ref": "#/components/responses/TooManyRequests429"
+                    },
+                    "500": {
+                        "$ref": "#/components/responses/InternalServerError500"
+                    },
+                    "default": {
+                        "$ref": "#/components/responses/DefaultError"
+                    }
+                },
+                "tags": [
+                    "Prompts"
+                ],
+                "description": "Create a new prompt.",
+                "parameters": [],
+                "requestBody": {
+                    "content": {
+                        "application/json": {
+                            "schema": {
+                                "$ref": "#/components/schemas/CreatePromptRequest"
+                            }
+                        }
+                    },
+                    "required": true
+                }
+            }
+        },
         "/v1/agents/{agent_id}": {
             "get": {
                 "responses": {
@@ -901,6 +975,143 @@
                 ]
             }
         },
+        "/v1/prompts/{prompt_id}": {
+            "get": {
+                "responses": {
+                    "200": {
+                        "description": "A Prompt resource.",
+                        "content": {
+                            "application/json": {
+                                "schema": {
+                                    "$ref": "#/components/schemas/Prompt"
+                                }
+                            }
+                        }
+                    },
+                    "400": {
+                        "$ref": "#/components/responses/BadRequest400"
+                    },
+                    "429": {
+                        "$ref": "#/components/responses/TooManyRequests429"
+                    },
+                    "500": {
+                        "$ref": "#/components/responses/InternalServerError500"
+                    },
+                    "default": {
+                        "$ref": "#/components/responses/DefaultError"
+                    }
+                },
+                "tags": [
+                    "Prompts"
+                ],
+                "description": "Get a prompt by its identifier and optional version.",
+                "parameters": [
+                    {
+                        "name": "prompt_id",
+                        "in": "path",
+                        "description": "The identifier of the prompt to get.",
+                        "required": true,
+                        "schema": {
+                            "type": "string"
+                        }
+                    },
+                    {
+                        "name": "version",
+                        "in": "query",
+                        "description": "The version of the prompt to get (defaults to latest).",
+                        "required": false,
+                        "schema": {
+                            "type": "integer"
+                        }
+                    }
+                ]
+            },
+            "post": {
+                "responses": {
+                    "200": {
+                        "description": "The updated Prompt resource with incremented version.",
+                        "content": {
+                            "application/json": {
+                                "schema": {
+                                    "$ref": "#/components/schemas/Prompt"
+                                }
+                            }
+                        }
+                    },
+                    "400": {
+                        "$ref": "#/components/responses/BadRequest400"
+                    },
+                    "429": {
+                        "$ref": "#/components/responses/TooManyRequests429"
+                    },
+                    "500": {
+                        "$ref": "#/components/responses/InternalServerError500"
+                    },
+                    "default": {
+                        "$ref": "#/components/responses/DefaultError"
+                    }
+                },
+                "tags": [
+                    "Prompts"
+                ],
+                "description": "Update an existing prompt (increments version).",
+                "parameters": [
+                    {
+                        "name": "prompt_id",
+                        "in": "path",
+                        "description": "The identifier of the prompt to update.",
+                        "required": true,
+                        "schema": {
+                            "type": "string"
+                        }
+                    }
+                ],
+                "requestBody": {
+                    "content": {
+                        "application/json": {
+                            "schema": {
+                                "$ref": "#/components/schemas/UpdatePromptRequest"
+                            }
+                        }
+                    },
+                    "required": true
+                }
+            },
+            "delete": {
+                "responses": {
+                    "200": {
+                        "description": "OK"
+                    },
+                    "400": {
+                        "$ref": "#/components/responses/BadRequest400"
+                    },
+                    "429": {
+                        "$ref": "#/components/responses/TooManyRequests429"
+                    },
+                    "500": {
+                        "$ref": "#/components/responses/InternalServerError500"
+                    },
+                    "default": {
+                        "$ref": "#/components/responses/DefaultError"
+                    }
+                },
+                "tags": [
+                    "Prompts"
+                ],
+                "description": "Delete a prompt.",
+                "parameters": [
+                    {
+                        "name": "prompt_id",
+                        "in": "path",
+                        "description": "The identifier of the prompt to delete.",
+                        "required": true,
+                        "schema": {
+                            "type": "string"
+                        }
+                    }
+                ]
+            }
+        },
         "/v1/inference/embeddings": {
             "post": {
                 "responses": {
@@ -2836,6 +3047,49 @@
                 ]
             }
         },
+        "/v1/prompts/{prompt_id}/versions": {
+            "get": {
+                "responses": {
+                    "200": {
+                        "description": "A ListPromptsResponse containing all versions of the prompt.",
+                        "content": {
+                            "application/json": {
+                                "schema": {
+                                    "$ref": "#/components/schemas/ListPromptsResponse"
+                                }
+                            }
+                        }
+                    },
+                    "400": {
+                        "$ref": "#/components/responses/BadRequest400"
+                    },
+                    "429": {
+                        "$ref": "#/components/responses/TooManyRequests429"
+                    },
+                    "500": {
+                        "$ref": "#/components/responses/InternalServerError500"
+                    },
+                    "default": {
+                        "$ref": "#/components/responses/DefaultError"
+                    }
+                },
+                "tags": [
+                    "Prompts"
+                ],
+                "description": "List all versions of a specific prompt.",
+                "parameters": [
+                    {
+                        "name": "prompt_id",
+                        "in": "path",
+                        "description": "The identifier of the prompt to list versions for.",
+                        "required": true,
+                        "schema": {
+                            "type": "string"
+                        }
+                    }
+                ]
+            }
+        },
         "/v1/providers": {
             "get": {
                 "responses": {
@@ -5007,6 +5261,59 @@
                 }
             }
         },
+        "/v1/prompts/{prompt_id}/set-default-version": {
+            "post": {
+                "responses": {
+                    "200": {
+                        "description": "The prompt with the specified version now set as default.",
+                        "content": {
+                            "application/json": {
+                                "schema": {
+                                    "$ref": "#/components/schemas/Prompt"
+                                }
+                            }
+                        }
+                    },
+                    "400": {
+                        "$ref": "#/components/responses/BadRequest400"
+                    },
+                    "429": {
+                        "$ref": "#/components/responses/TooManyRequests429"
+                    },
+                    "500": {
+                        "$ref": "#/components/responses/InternalServerError500"
+                    },
+                    "default": {
+                        "$ref": "#/components/responses/DefaultError"
+                    }
+                },
+                "tags": [
+                    "Prompts"
+                ],
+                "description": "Set which version of a prompt should be the default in get_prompt (latest).",
+                "parameters": [
+                    {
+                        "name": "prompt_id",
+                        "in": "path",
+                        "description": "The identifier of the prompt.",
+                        "required": true,
+                        "schema": {
+                            "type": "string"
+                        }
+                    }
+                ],
+                "requestBody": {
+                    "content": {
+                        "application/json": {
+                            "schema": {
+                                "$ref": "#/components/schemas/SetDefaultVersionRequest"
+                            }
+                        }
+                    },
+                    "required": true
+                }
+            }
+        },
         "/v1/post-training/supervised-fine-tune": {
             "post": {
                 "responses": {
@@ -9670,6 +9977,65 @@
                 ],
                 "title": "OpenAIResponseObjectStreamResponseWebSearchCallSearching"
             },
+            "CreatePromptRequest": {
+                "type": "object",
+                "properties": {
+                    "prompt": {
+                        "type": "string",
+                        "description": "The prompt text content with variable placeholders."
+                    },
+                    "variables": {
+                        "type": "array",
+                        "items": {
+                            "type": "string"
+                        },
+                        "description": "List of variable names that can be used in the prompt template."
+                    }
+                },
+                "additionalProperties": false,
+                "required": [
+                    "prompt"
+                ],
+                "title": "CreatePromptRequest"
+            },
+            "Prompt": {
+                "type": "object",
+                "properties": {
+                    "prompt": {
+                        "type": "string",
+                        "description": "The system prompt text with variable placeholders. Variables are only supported when using the Responses API."
+                    },
+                    "version": {
+                        "type": "integer",
+                        "description": "Version (integer starting at 1, incremented on save)"
+                    },
+                    "prompt_id": {
+                        "type": "string",
+                        "description": "Unique identifier formatted as 'pmpt_<48-digit-hash>'"
+                    },
+                    "variables": {
+                        "type": "array",
+                        "items": {
+                            "type": "string"
+                        },
+                        "description": "List of prompt variable names that can be used in the prompt template"
+                    },
+                    "is_default": {
+                        "type": "boolean",
+                        "default": false,
+                        "description": "Boolean indicating whether this version is the default version for this prompt"
+                    }
+                },
+                "additionalProperties": false,
+                "required": [
+                    "version",
+                    "prompt_id",
+                    "variables",
+                    "is_default"
+                ],
+                "title": "Prompt",
+                "description": "A prompt resource representing a stored OpenAI Compatible prompt template in Llama Stack."
+            },
             "OpenAIDeleteResponseObject": {
                 "type": "object",
                 "properties": {
@@ -10296,7 +10662,8 @@
                             "scoring_function",
                             "benchmark",
                             "tool",
-                            "tool_group"
+                            "tool_group",
+                            "prompt"
                         ],
                         "const": "benchmark",
                         "default": "benchmark",
@@ -10923,7 +11290,8 @@
                             "scoring_function",
                             "benchmark",
                             "tool",
-                            "tool_group"
+                            "tool_group",
+                            "prompt"
                         ],
                         "const": "dataset",
                         "default": "dataset",
@@ -11073,7 +11441,8 @@
                             "scoring_function",
                             "benchmark",
                             "tool",
-                            "tool_group"
+                            "tool_group",
+                            "prompt"
                         ],
                         "const": "model",
                         "default": "model",
@@ -11338,7 +11707,8 @@
                             "scoring_function",
                             "benchmark",
                             "tool",
-                            "tool_group"
+                            "tool_group",
+                            "prompt"
                         ],
                         "const": "scoring_function",
                         "default": "scoring_function",
@@ -11446,7 +11816,8 @@
                             "scoring_function",
                             "benchmark",
                             "tool",
-                            "tool_group"
+                            "tool_group",
+                            "prompt"
                         ],
                         "const": "shield",
                         "default": "shield",
@@ -11691,7 +12062,8 @@
                             "scoring_function",
                             "benchmark",
                             "tool",
-                            "tool_group"
+                            "tool_group",
+                            "prompt"
                         ],
                         "const": "tool",
                         "default": "tool",
@@ -11773,7 +12145,8 @@
                             "scoring_function",
                             "benchmark",
                             "tool",
-                            "tool_group"
+                            "tool_group",
+                            "prompt"
                         ],
                         "const": "tool_group",
                         "default": "tool_group",
@@ -12067,7 +12440,8 @@
                             "scoring_function",
                             "benchmark",
                             "tool",
-                            "tool_group"
+                            "tool_group",
+                            "prompt"
                         ],
                         "const": "vector_db",
                         "default": "vector_db",
@@ -12882,6 +13256,23 @@
                 "title": "OpenAIResponseObjectWithInput",
                 "description": "OpenAI response object extended with input context information."
             },
+            "ListPromptsResponse": {
+                "type": "object",
+                "properties": {
+                    "data": {
+                        "type": "array",
+                        "items": {
+                            "$ref": "#/components/schemas/Prompt"
+                        }
+                    }
+                },
+                "additionalProperties": false,
+                "required": [
+                    "data"
+                ],
+                "title": "ListPromptsResponse",
+                "description": "Response model to list prompts."
+            },
             "ListProvidersResponse": {
                 "type": "object",
                 "properties": {
@@ -17129,6 +17520,20 @@
                 "title": "ScoreBatchResponse",
                 "description": "Response from batch scoring operations on datasets."
             },
+            "SetDefaultVersionRequest": {
+                "type": "object",
+                "properties": {
+                    "version": {
+                        "type": "integer",
+                        "description": "The version to set as default."
+                    }
+                },
+                "additionalProperties": false,
+                "required": [
+                    "version"
+                ],
+                "title": "SetDefaultVersionRequest"
+            },
             "AlgorithmConfig": {
                 "oneOf": [
                     {
@@ -17413,6 +17818,37 @@
                 "title": "SyntheticDataGenerationResponse",
                 "description": "Response from the synthetic data generation. Batch of (prompt, response, score) tuples that pass the threshold."
             },
+            "UpdatePromptRequest": {
+                "type": "object",
+                "properties": {
+                    "prompt": {
+                        "type": "string",
+                        "description": "The updated prompt text content."
+                    },
+                    "version": {
+                        "type": "integer",
+                        "description": "The current version of the prompt being updated."
+                    },
+                    "variables": {
+                        "type": "array",
+                        "items": {
+                            "type": "string"
+                        },
+                        "description": "Updated list of variable names that can be used in the prompt template."
+                    },
+                    "set_as_default": {
+                        "type": "boolean",
+                        "description": "Set the new version as the default (default=True)."
+                    }
+                },
+                "additionalProperties": false,
+                "required": [
+                    "prompt",
+                    "version",
+                    "set_as_default"
+                ],
+                "title": "UpdatePromptRequest"
+            },
             "VersionInfo": {
                 "type": "object",
                 "properties": {
@@ -17538,6 +17974,10 @@
         {
             "name": "PostTraining (Coming Soon)"
         },
+        {
+            "name": "Prompts",
+            "x-displayName": "Protocol for prompt management operations."
+        },
         {
             "name": "Providers",
             "x-displayName": "Providers API for inspecting, listing, and modifying providers and their configurations."
@@ -17588,6 +18028,7 @@
                 "Inspect",
                 "Models",
                 "PostTraining (Coming Soon)",
+                "Prompts",
                 "Providers",
                 "Safety",
                 "Scoring",

docs/_static/llama-stack-spec.yaml

@@ -427,6 +427,58 @@ paths:
             schema:
               $ref: '#/components/schemas/CreateOpenaiResponseRequest'
         required: true
+  /v1/prompts:
+    get:
+      responses:
+        '200':
+          description: >-
+            A ListPromptsResponse containing all prompts.
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/ListPromptsResponse'
+        '400':
+          $ref: '#/components/responses/BadRequest400'
+        '429':
+          $ref: >-
+            #/components/responses/TooManyRequests429
+        '500':
+          $ref: >-
+            #/components/responses/InternalServerError500
+        default:
+          $ref: '#/components/responses/DefaultError'
+      tags:
+        - Prompts
+      description: List all prompts.
+      parameters: []
+    post:
+      responses:
+        '200':
+          description: The created Prompt resource.
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/Prompt'
+        '400':
+          $ref: '#/components/responses/BadRequest400'
+        '429':
+          $ref: >-
+            #/components/responses/TooManyRequests429
+        '500':
+          $ref: >-
+            #/components/responses/InternalServerError500
+        default:
+          $ref: '#/components/responses/DefaultError'
+      tags:
+        - Prompts
+      description: Create a new prompt.
+      parameters: []
+      requestBody:
+        content:
+          application/json:
+            schema:
+              $ref: '#/components/schemas/CreatePromptRequest'
+        required: true
   /v1/agents/{agent_id}:
     get:
       responses:
@@ -616,6 +668,103 @@ paths:
           required: true
           schema:
             type: string
+  /v1/prompts/{prompt_id}:
+    get:
+      responses:
+        '200':
+          description: A Prompt resource.
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/Prompt'
+        '400':
+          $ref: '#/components/responses/BadRequest400'
+        '429':
+          $ref: >-
+            #/components/responses/TooManyRequests429
+        '500':
+          $ref: >-
+            #/components/responses/InternalServerError500
+        default:
+          $ref: '#/components/responses/DefaultError'
+      tags:
+        - Prompts
+      description: >-
+        Get a prompt by its identifier and optional version.
+      parameters:
+        - name: prompt_id
+          in: path
+          description: The identifier of the prompt to get.
+          required: true
+          schema:
+            type: string
+        - name: version
+          in: query
+          description: >-
+            The version of the prompt to get (defaults to latest).
+          required: false
+          schema:
+            type: integer
+    post:
+      responses:
+        '200':
+          description: >-
+            The updated Prompt resource with incremented version.
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/Prompt'
+        '400':
+          $ref: '#/components/responses/BadRequest400'
+        '429':
+          $ref: >-
+            #/components/responses/TooManyRequests429
+        '500':
+          $ref: >-
+            #/components/responses/InternalServerError500
+        default:
+          $ref: '#/components/responses/DefaultError'
+      tags:
+        - Prompts
+      description: >-
+        Update an existing prompt (increments version).
+      parameters:
+        - name: prompt_id
+          in: path
+          description: The identifier of the prompt to update.
+          required: true
+          schema:
+            type: string
+      requestBody:
+        content:
+          application/json:
+            schema:
+              $ref: '#/components/schemas/UpdatePromptRequest'
+        required: true
+    delete:
+      responses:
+        '200':
+          description: OK
+        '400':
+          $ref: '#/components/responses/BadRequest400'
+        '429':
+          $ref: >-
+            #/components/responses/TooManyRequests429
+        '500':
+          $ref: >-
+            #/components/responses/InternalServerError500
+        default:
+          $ref: '#/components/responses/DefaultError'
+      tags:
+        - Prompts
+      description: Delete a prompt.
+      parameters:
+        - name: prompt_id
+          in: path
+          description: The identifier of the prompt to delete.
+          required: true
+          schema:
+            type: string
   /v1/inference/embeddings:
     post:
       responses:
@@ -1983,6 +2132,37 @@ paths:
           required: false
           schema:
             $ref: '#/components/schemas/Order'
+  /v1/prompts/{prompt_id}/versions:
+    get:
+      responses:
+        '200':
+          description: >-
+            A ListPromptsResponse containing all versions of the prompt.
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/ListPromptsResponse'
+        '400':
+          $ref: '#/components/responses/BadRequest400'
+        '429':
+          $ref: >-
+            #/components/responses/TooManyRequests429
+        '500':
+          $ref: >-
+            #/components/responses/InternalServerError500
+        default:
+          $ref: '#/components/responses/DefaultError'
+      tags:
+        - Prompts
+      description: List all versions of a specific prompt.
+      parameters:
+        - name: prompt_id
+          in: path
+          description: >-
+            The identifier of the prompt to list versions for.
+          required: true
+          schema:
+            type: string
   /v1/providers:
     get:
       responses:
@@ -3546,6 +3726,43 @@ paths:
             schema:
               $ref: '#/components/schemas/ScoreBatchRequest'
         required: true
+  /v1/prompts/{prompt_id}/set-default-version:
+    post:
+      responses:
+        '200':
+          description: >-
+            The prompt with the specified version now set as default.
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/Prompt'
+        '400':
+          $ref: '#/components/responses/BadRequest400'
+        '429':
+          $ref: >-
+            #/components/responses/TooManyRequests429
+        '500':
+          $ref: >-
+            #/components/responses/InternalServerError500
+        default:
+          $ref: '#/components/responses/DefaultError'
+      tags:
+        - Prompts
+      description: >-
+        Set which version of a prompt should be the default in get_prompt (latest).
+      parameters:
+        - name: prompt_id
+          in: path
+          description: The identifier of the prompt.
+          required: true
+          schema:
+            type: string
+      requestBody:
+        content:
+          application/json:
+            schema:
+              $ref: '#/components/schemas/SetDefaultVersionRequest'
+        required: true
   /v1/post-training/supervised-fine-tune:
     post:
       responses:
@@ -7148,6 +7365,61 @@ components:
         - type
       title: >-
         OpenAIResponseObjectStreamResponseWebSearchCallSearching
+    CreatePromptRequest:
+      type: object
+      properties:
+        prompt:
+          type: string
+          description: >-
+            The prompt text content with variable placeholders.
+        variables:
+          type: array
+          items:
+            type: string
+          description: >-
+            List of variable names that can be used in the prompt template.
+      additionalProperties: false
+      required:
+        - prompt
+      title: CreatePromptRequest
+    Prompt:
+      type: object
+      properties:
+        prompt:
+          type: string
+          description: >-
+            The system prompt text with variable placeholders. Variables are only
+            supported when using the Responses API.
+        version:
+          type: integer
+          description: >-
+            Version (integer starting at 1, incremented on save)
+        prompt_id:
+          type: string
+          description: >-
+            Unique identifier formatted as 'pmpt_<48-digit-hash>'
+        variables:
+          type: array
+          items:
+            type: string
+          description: >-
+            List of prompt variable names that can be used in the prompt template
+        is_default:
+          type: boolean
+          default: false
+          description: >-
+            Boolean indicating whether this version is the default version for this
+            prompt
+      additionalProperties: false
+      required:
+        - version
+        - prompt_id
+        - variables
+        - is_default
+      title: Prompt
+      description: >-
+        A prompt resource representing a stored OpenAI Compatible prompt template
+        in Llama Stack.
     OpenAIDeleteResponseObject:
       type: object
       properties:
@@ -7621,6 +7893,7 @@ components:
             - benchmark
             - tool
             - tool_group
+            - prompt
           const: benchmark
           default: benchmark
           description: The resource type, always benchmark
@@ -8107,6 +8380,7 @@ components:
             - benchmark
             - tool
             - tool_group
+            - prompt
           const: dataset
           default: dataset
           description: >-
@@ -8219,6 +8493,7 @@ components:
             - benchmark
             - tool
             - tool_group
+            - prompt
           const: model
           default: model
           description: >-
@@ -8410,6 +8685,7 @@ components:
             - benchmark
             - tool
             - tool_group
+            - prompt
           const: scoring_function
           default: scoring_function
           description: >-
@@ -8486,6 +8762,7 @@ components:
             - benchmark
             - tool
             - tool_group
+            - prompt
           const: shield
           default: shield
           description: The resource type, always shield
@@ -8665,6 +8942,7 @@ components:
             - benchmark
             - tool
             - tool_group
+            - prompt
           const: tool
           default: tool
           description: Type of resource, always 'tool'
@@ -8723,6 +9001,7 @@ components:
             - benchmark
             - tool
             - tool_group
+            - prompt
           const: tool_group
           default: tool_group
           description: Type of resource, always 'tool_group'
@@ -8951,6 +9230,7 @@ components:
             - benchmark
             - tool
             - tool_group
+            - prompt
           const: vector_db
           default: vector_db
           description: >-
@@ -9577,6 +9857,18 @@ components:
       title: OpenAIResponseObjectWithInput
       description: >-
         OpenAI response object extended with input context information.
+    ListPromptsResponse:
+      type: object
+      properties:
+        data:
+          type: array
+          items:
+            $ref: '#/components/schemas/Prompt'
+      additionalProperties: false
+      required:
+        - data
+      title: ListPromptsResponse
+      description: Response model to list prompts.
     ListProvidersResponse:
       type: object
       properties:
@@ -12723,6 +13015,16 @@ components:
       title: ScoreBatchResponse
       description: >-
         Response from batch scoring operations on datasets.
+    SetDefaultVersionRequest:
+      type: object
+      properties:
+        version:
+          type: integer
+          description: The version to set as default.
+      additionalProperties: false
+      required:
+        - version
+      title: SetDefaultVersionRequest
     AlgorithmConfig:
       oneOf:
         - $ref: '#/components/schemas/LoraFinetuningConfig'
@@ -12919,6 +13221,32 @@ components:
       description: >-
         Response from the synthetic data generation. Batch of (prompt, response, score)
         tuples that pass the threshold.
+    UpdatePromptRequest:
+      type: object
+      properties:
+        prompt:
+          type: string
+          description: The updated prompt text content.
+        version:
+          type: integer
+          description: >-
+            The current version of the prompt being updated.
+        variables:
+          type: array
+          items:
+            type: string
+          description: >-
+            Updated list of variable names that can be used in the prompt template.
+        set_as_default:
+          type: boolean
+          description: >-
+            Set the new version as the default (default=True).
+      additionalProperties: false
+      required:
+        - prompt
+        - version
+        - set_as_default
+      title: UpdatePromptRequest
     VersionInfo:
       type: object
       properties:
@@ -13030,6 +13358,9 @@ tags:
   - name: Inspect
   - name: Models
   - name: PostTraining (Coming Soon)
+  - name: Prompts
+    x-displayName: >-
+      Protocol for prompt management operations.
   - name: Providers
     x-displayName: >-
       Providers API for inspecting, listing, and modifying providers and their configurations.
@@ -13057,6 +13388,7 @@ x-tagGroups:
       - Inspect
       - Models
       - PostTraining (Coming Soon)
+      - Prompts
       - Providers
       - Safety
       - Scoring

docs/source/conf.py

@@ -131,6 +131,7 @@ html_static_path = ["../_static"]
 def setup(app):
     app.add_css_file("css/my_theme.css")
     app.add_js_file("js/detect_theme.js")
+    app.add_js_file("js/horizontal_nav.js")
     app.add_js_file("js/keyboard_shortcuts.js")
 
     def dockerhub_role(name, rawtext, text, lineno, inliner, options={}, content=[]):

docs/source/contributing/index.md

@@ -35,5 +35,5 @@ testing/record-replay
 
 ### Benchmarking
 
-```{include} ../../../docs/source/distributions/k8s-benchmark/README.md
+```{include} ../../../benchmarking/k8s-benchmark/README.md
 ```

docs/source/contributing/testing/record-replay.md

@@ -40,18 +40,15 @@ The system patches OpenAI and Ollama client methods to intercept calls before th
 
 ### Storage Architecture
 
-Recordings use a two-tier storage system optimized for both speed and debuggability:
+Recordings are stored as JSON files in the recording directory. They are looked up by their request hash.
 
 ```
 recordings/
-├── index.sqlite          # Fast lookup by request hash
 └── responses/
     ├── abc123def456.json  # Individual response files
     └── def789ghi012.json
 ```
 
-**SQLite index** enables O(log n) hash lookups and metadata queries without loading response bodies.
-
 **JSON files** store complete request/response pairs in human-readable format for debugging.
 
 ## Recording Modes
@@ -166,8 +163,8 @@ This preserves type safety - when replayed, you get the same Pydantic objects wi
 Control recording behavior globally:
 
 ```bash
-export LLAMA_STACK_TEST_INFERENCE_MODE=replay
+export LLAMA_STACK_TEST_INFERENCE_MODE=replay   # this is the default
-export LLAMA_STACK_TEST_RECORDING_DIR=/path/to/recordings
+export LLAMA_STACK_TEST_RECORDING_DIR=/path/to/recordings   # default is tests/integration/recordings
 pytest tests/integration/
 ```
 

docs/source/distributions/configuration.md

@@ -354,6 +354,47 @@ You can easily validate a request by running:
 curl -s -L -H "Authorization: Bearer $(cat llama-stack-auth-token)" http://127.0.0.1:8321/v1/providers
 ```
 
+#### Kubernetes Authentication Provider
+
+The server can be configured to use Kubernetes SelfSubjectReview API to validate tokens directly against the Kubernetes API server:
+
+```yaml
+server:
+  auth:
+    provider_config:
+      type: "kubernetes"
+      api_server_url: "https://kubernetes.default.svc"
+      claims_mapping:
+        username: "roles"
+        groups: "roles"
+        uid: "uid_attr"
+      verify_tls: true
+      tls_cafile: "/path/to/ca.crt"
+```
+
+Configuration options:
+- `api_server_url`: The Kubernetes API server URL (e.g., https://kubernetes.default.svc:6443)
+- `verify_tls`: Whether to verify TLS certificates (default: true)
+- `tls_cafile`: Path to CA certificate file for TLS verification
+- `claims_mapping`: Mapping of Kubernetes user claims to access attributes
+
+The provider validates tokens by sending a SelfSubjectReview request to the Kubernetes API server at `/apis/authentication.k8s.io/v1/selfsubjectreviews`. The provider extracts user information from the response:
+- Username from the `userInfo.username` field
+- Groups from the `userInfo.groups` field
+- UID from the `userInfo.uid` field
+
+To obtain a token for testing:
+```bash
+kubectl create namespace llama-stack
+kubectl create serviceaccount llama-stack-auth -n llama-stack
+kubectl create token llama-stack-auth -n llama-stack > llama-stack-auth-token
+```
+
+You can validate a request by running:
+```bash
+curl -s -L -H "Authorization: Bearer $(cat llama-stack-auth-token)" http://127.0.0.1:8321/v1/providers
+```
+
 #### GitHub Token Provider
 Validates GitHub personal access tokens or OAuth tokens directly:
 ```yaml

docs/source/distributions/k8s/stack-configmap.yaml

@@ -1,137 +1,55 @@
 apiVersion: v1
 data:
-  stack_run_config.yaml: |
+  stack_run_config.yaml: "version: '2'\nimage_name: kubernetes-demo\napis:\n- agents\n-
-    version: '2'
+    inference\n- files\n- safety\n- telemetry\n- tool_runtime\n- vector_io\nproviders:\n
-    image_name: kubernetes-demo
+    \ inference:\n  - provider_id: vllm-inference\n    provider_type: remote::vllm\n
-    apis:
+    \   config:\n      url: ${env.VLLM_URL:=http://localhost:8000/v1}\n      max_tokens:
-    - agents
+    ${env.VLLM_MAX_TOKENS:=4096}\n      api_token: ${env.VLLM_API_TOKEN:=fake}\n      tls_verify:
-    - inference
+    ${env.VLLM_TLS_VERIFY:=true}\n  - provider_id: vllm-safety\n    provider_type:
-    - safety
+    remote::vllm\n    config:\n      url: ${env.VLLM_SAFETY_URL:=http://localhost:8000/v1}\n
-    - telemetry
+    \     max_tokens: ${env.VLLM_MAX_TOKENS:=4096}\n      api_token: ${env.VLLM_API_TOKEN:=fake}\n
-    - tool_runtime
+    \     tls_verify: ${env.VLLM_TLS_VERIFY:=true}\n  - provider_id: sentence-transformers\n
-    - vector_io
+    \   provider_type: inline::sentence-transformers\n    config: {}\n  vector_io:\n
-    providers:
+    \ - provider_id: ${env.ENABLE_CHROMADB:+chromadb}\n    provider_type: remote::chromadb\n
-      inference:
+    \   config:\n      url: ${env.CHROMADB_URL:=}\n      kvstore:\n        type: postgres\n
-      - provider_id: vllm-inference
+    \       host: ${env.POSTGRES_HOST:=localhost}\n        port: ${env.POSTGRES_PORT:=5432}\n
-        provider_type: remote::vllm
+    \       db: ${env.POSTGRES_DB:=llamastack}\n        user: ${env.POSTGRES_USER:=llamastack}\n
-        config:
+    \       password: ${env.POSTGRES_PASSWORD:=llamastack}\n  files:\n  - provider_id:
-          url: ${env.VLLM_URL:=http://localhost:8000/v1}
+    meta-reference-files\n    provider_type: inline::localfs\n    config:\n      storage_dir:
-          max_tokens: ${env.VLLM_MAX_TOKENS:=4096}
+    ${env.FILES_STORAGE_DIR:=~/.llama/distributions/starter/files}\n      metadata_store:\n
-          api_token: ${env.VLLM_API_TOKEN:=fake}
+    \       type: sqlite\n        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/files_metadata.db
-          tls_verify: ${env.VLLM_TLS_VERIFY:=true}
+    \ \n  safety:\n  - provider_id: llama-guard\n    provider_type: inline::llama-guard\n
-      - provider_id: vllm-safety
+    \   config:\n      excluded_categories: []\n  agents:\n  - provider_id: meta-reference\n
-        provider_type: remote::vllm
+    \   provider_type: inline::meta-reference\n    config:\n      persistence_store:\n
-        config:
+    \       type: postgres\n        host: ${env.POSTGRES_HOST:=localhost}\n        port:
-          url: ${env.VLLM_SAFETY_URL:=http://localhost:8000/v1}
+    ${env.POSTGRES_PORT:=5432}\n        db: ${env.POSTGRES_DB:=llamastack}\n        user:
-          max_tokens: ${env.VLLM_MAX_TOKENS:=4096}
+    ${env.POSTGRES_USER:=llamastack}\n        password: ${env.POSTGRES_PASSWORD:=llamastack}\n
-          api_token: ${env.VLLM_API_TOKEN:=fake}
+    \     responses_store:\n        type: postgres\n        host: ${env.POSTGRES_HOST:=localhost}\n
-          tls_verify: ${env.VLLM_TLS_VERIFY:=true}
+    \       port: ${env.POSTGRES_PORT:=5432}\n        db: ${env.POSTGRES_DB:=llamastack}\n
-      - provider_id: sentence-transformers
+    \       user: ${env.POSTGRES_USER:=llamastack}\n        password: ${env.POSTGRES_PASSWORD:=llamastack}\n
-        provider_type: inline::sentence-transformers
+    \ telemetry:\n  - provider_id: meta-reference\n    provider_type: inline::meta-reference\n
-        config: {}
+    \   config:\n      service_name: \"${env.OTEL_SERVICE_NAME:=\\u200B}\"\n      sinks:
-      vector_io:
+    ${env.TELEMETRY_SINKS:=console}\n  tool_runtime:\n  - provider_id: brave-search\n
-      - provider_id: ${env.ENABLE_CHROMADB:+chromadb}
+    \   provider_type: remote::brave-search\n    config:\n      api_key: ${env.BRAVE_SEARCH_API_KEY:+}\n
-        provider_type: remote::chromadb
+    \     max_results: 3\n  - provider_id: tavily-search\n    provider_type: remote::tavily-search\n
-        config:
+    \   config:\n      api_key: ${env.TAVILY_SEARCH_API_KEY:+}\n      max_results:
-          url: ${env.CHROMADB_URL:=}
+    3\n  - provider_id: rag-runtime\n    provider_type: inline::rag-runtime\n    config:
-          kvstore:
+    {}\n  - provider_id: model-context-protocol\n    provider_type: remote::model-context-protocol\n
-            type: postgres
+    \   config: {}\nmetadata_store:\n  type: postgres\n  host: ${env.POSTGRES_HOST:=localhost}\n
-            host: ${env.POSTGRES_HOST:=localhost}
+    \ port: ${env.POSTGRES_PORT:=5432}\n  db: ${env.POSTGRES_DB:=llamastack}\n  user:
-            port: ${env.POSTGRES_PORT:=5432}
+    ${env.POSTGRES_USER:=llamastack}\n  password: ${env.POSTGRES_PASSWORD:=llamastack}\n
-            db: ${env.POSTGRES_DB:=llamastack}
+    \ table_name: llamastack_kvstore\ninference_store:\n  type: postgres\n  host:
-            user: ${env.POSTGRES_USER:=llamastack}
+    ${env.POSTGRES_HOST:=localhost}\n  port: ${env.POSTGRES_PORT:=5432}\n  db: ${env.POSTGRES_DB:=llamastack}\n
-            password: ${env.POSTGRES_PASSWORD:=llamastack}
+    \ user: ${env.POSTGRES_USER:=llamastack}\n  password: ${env.POSTGRES_PASSWORD:=llamastack}\nmodels:\n-
-      safety:
+    metadata:\n    embedding_dimension: 384\n  model_id: all-MiniLM-L6-v2\n  provider_id:
-      - provider_id: llama-guard
+    sentence-transformers\n  model_type: embedding\n- metadata: {}\n  model_id: ${env.INFERENCE_MODEL}\n
-        provider_type: inline::llama-guard
+    \ provider_id: vllm-inference\n  model_type: llm\n- metadata: {}\n  model_id:
-        config:
+    ${env.SAFETY_MODEL:=meta-llama/Llama-Guard-3-1B}\n  provider_id: vllm-safety\n
-          excluded_categories: []
+    \ model_type: llm\nshields:\n- shield_id: ${env.SAFETY_MODEL:=meta-llama/Llama-Guard-3-1B}\nvector_dbs:
-      agents:
+    []\ndatasets: []\nscoring_fns: []\nbenchmarks: []\ntool_groups:\n- toolgroup_id:
-      - provider_id: meta-reference
+    builtin::websearch\n  provider_id: tavily-search\n- toolgroup_id: builtin::rag\n
-        provider_type: inline::meta-reference
+    \ provider_id: rag-runtime\nserver:\n  port: 8321\n  auth:\n    provider_config:\n
-        config:
+    \     type: github_token\n"
-          persistence_store:
-            type: postgres
-            host: ${env.POSTGRES_HOST:=localhost}
-            port: ${env.POSTGRES_PORT:=5432}
-            db: ${env.POSTGRES_DB:=llamastack}
-            user: ${env.POSTGRES_USER:=llamastack}
-            password: ${env.POSTGRES_PASSWORD:=llamastack}
-          responses_store:
-            type: postgres
-            host: ${env.POSTGRES_HOST:=localhost}
-            port: ${env.POSTGRES_PORT:=5432}
-            db: ${env.POSTGRES_DB:=llamastack}
-            user: ${env.POSTGRES_USER:=llamastack}
-            password: ${env.POSTGRES_PASSWORD:=llamastack}
-      telemetry:
-      - provider_id: meta-reference
-        provider_type: inline::meta-reference
-        config:
-          service_name: "${env.OTEL_SERVICE_NAME:=\u200B}"
-          sinks: ${env.TELEMETRY_SINKS:=console}
-      tool_runtime:
-      - provider_id: brave-search
-        provider_type: remote::brave-search
-        config:
-          api_key: ${env.BRAVE_SEARCH_API_KEY:+}
-          max_results: 3
-      - provider_id: tavily-search
-        provider_type: remote::tavily-search
-        config:
-          api_key: ${env.TAVILY_SEARCH_API_KEY:+}
-          max_results: 3
-      - provider_id: rag-runtime
-        provider_type: inline::rag-runtime
-        config: {}
-      - provider_id: model-context-protocol
-        provider_type: remote::model-context-protocol
-        config: {}
-    metadata_store:
-      type: postgres
-      host: ${env.POSTGRES_HOST:=localhost}
-      port: ${env.POSTGRES_PORT:=5432}
-      db: ${env.POSTGRES_DB:=llamastack}
-      user: ${env.POSTGRES_USER:=llamastack}
-      password: ${env.POSTGRES_PASSWORD:=llamastack}
-      table_name: llamastack_kvstore
-    inference_store:
-      type: postgres
-      host: ${env.POSTGRES_HOST:=localhost}
-      port: ${env.POSTGRES_PORT:=5432}
-      db: ${env.POSTGRES_DB:=llamastack}
-      user: ${env.POSTGRES_USER:=llamastack}
-      password: ${env.POSTGRES_PASSWORD:=llamastack}
-    models:
-    - metadata:
-        embedding_dimension: 384
-      model_id: all-MiniLM-L6-v2
-      provider_id: sentence-transformers
-      model_type: embedding
-    - metadata: {}
-      model_id: ${env.INFERENCE_MODEL}
-      provider_id: vllm-inference
-      model_type: llm
-    - metadata: {}
-      model_id: ${env.SAFETY_MODEL:=meta-llama/Llama-Guard-3-1B}
-      provider_id: vllm-safety
-      model_type: llm
-    shields:
-    - shield_id: ${env.SAFETY_MODEL:=meta-llama/Llama-Guard-3-1B}
-    vector_dbs: []
-    datasets: []
-    scoring_fns: []
-    benchmarks: []
-    tool_groups:
-    - toolgroup_id: builtin::websearch
-      provider_id: tavily-search
-    - toolgroup_id: builtin::rag
-      provider_id: rag-runtime
-    server:
-      port: 8321
-      auth:
-        provider_config:
-          type: github_token
 kind: ConfigMap
 metadata:
   creationTimestamp: null

docs/source/distributions/k8s/stack_run_config.yaml

@@ -3,6 +3,7 @@ image_name: kubernetes-demo
 apis:
 - agents
 - inference
+- files
 - safety
 - telemetry
 - tool_runtime
@@ -38,6 +39,14 @@ providers:
         db: ${env.POSTGRES_DB:=llamastack}
         user: ${env.POSTGRES_USER:=llamastack}
         password: ${env.POSTGRES_PASSWORD:=llamastack}
+  files:
+  - provider_id: meta-reference-files
+    provider_type: inline::localfs
+    config:
+      storage_dir: ${env.FILES_STORAGE_DIR:=~/.llama/distributions/starter/files}
+      metadata_store:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/files_metadata.db
   safety:
   - provider_id: llama-guard
     provider_type: inline::llama-guard

docs/source/getting_started/demo_script.py

@@ -18,12 +18,13 @@ embedding_model_id = (
 ).identifier
 embedding_dimension = em.metadata["embedding_dimension"]
 
-_ = client.vector_dbs.register(
+vector_db = client.vector_dbs.register(
     vector_db_id=vector_db_id,
     embedding_model=embedding_model_id,
     embedding_dimension=embedding_dimension,
     provider_id="faiss",
 )
+vector_db_id = vector_db.identifier
 source = "https://www.paulgraham.com/greatwork.html"
 print("rag_tool> Ingesting document:", source)
 document = RAGDocument(
@@ -35,7 +36,7 @@ document = RAGDocument(
 client.tool_runtime.rag_tool.insert(
     documents=[document],
     vector_db_id=vector_db_id,
-    chunk_size_in_tokens=50,
+    chunk_size_in_tokens=100,
 )
 agent = Agent(
     client,

docs/source/providers/external/external-providers-list.md

@@ -8,3 +8,4 @@ Here's a list of known external providers that you can use with Llama Stack:
 | KubeFlow Pipelines | Train models with KubeFlow Pipelines | Post Training | Inline **and** Remote | [llama-stack-provider-kfp-trainer](https://github.com/opendatahub-io/llama-stack-provider-kfp-trainer) |
 | RamaLama | Inference models with RamaLama | Inference | Remote | [ramalama-stack](https://github.com/containers/ramalama-stack) |
 | TrustyAI LM-Eval | Evaluate models with TrustyAI LM-Eval | Eval | Remote | [llama-stack-provider-lmeval](https://github.com/trustyai-explainability/llama-stack-provider-lmeval) |
+| MongoDB | VectorIO with MongoDB | Vector_IO | Remote | [mongodb-llama-stack](https://github.com/mongodb-partners/mongodb-llama-stack) |

docs/source/providers/inference/remote_bedrock.md

@@ -15,8 +15,8 @@ AWS Bedrock inference provider for accessing various AI models through AWS's man
 | `profile_name` | `str \| None` | No |  | The profile name that contains credentials to use.Default use environment variable: AWS_PROFILE |
 | `total_max_attempts` | `int \| None` | No |  | An integer representing the maximum number of attempts that will be made for a single request, including the initial attempt. Default use environment variable: AWS_MAX_ATTEMPTS |
 | `retry_mode` | `str \| None` | No |  | A string representing the type of retries Boto3 will perform.Default use environment variable: AWS_RETRY_MODE |
-| `connect_timeout` | `float \| None` | No | 60 | The time in seconds till a timeout exception is thrown when attempting to make a connection. The default is 60 seconds. |
+| `connect_timeout` | `float \| None` | No | 60.0 | The time in seconds till a timeout exception is thrown when attempting to make a connection. The default is 60 seconds. |
-| `read_timeout` | `float \| None` | No | 60 | The time in seconds till a timeout exception is thrown when attempting to read from a connection.The default is 60 seconds. |
+| `read_timeout` | `float \| None` | No | 60.0 | The time in seconds till a timeout exception is thrown when attempting to read from a connection.The default is 60 seconds. |
 | `session_ttl` | `int \| None` | No | 3600 | The time in seconds till a session expires. The default is 3600 seconds (1 hour). |
 
 ## Sample Configuration

docs/source/providers/safety/remote_bedrock.md

@@ -15,8 +15,8 @@ AWS Bedrock safety provider for content moderation using AWS's safety services.
 | `profile_name` | `str \| None` | No |  | The profile name that contains credentials to use.Default use environment variable: AWS_PROFILE |
 | `total_max_attempts` | `int \| None` | No |  | An integer representing the maximum number of attempts that will be made for a single request, including the initial attempt. Default use environment variable: AWS_MAX_ATTEMPTS |
 | `retry_mode` | `str \| None` | No |  | A string representing the type of retries Boto3 will perform.Default use environment variable: AWS_RETRY_MODE |
-| `connect_timeout` | `float \| None` | No | 60 | The time in seconds till a timeout exception is thrown when attempting to make a connection. The default is 60 seconds. |
+| `connect_timeout` | `float \| None` | No | 60.0 | The time in seconds till a timeout exception is thrown when attempting to make a connection. The default is 60 seconds. |
-| `read_timeout` | `float \| None` | No | 60 | The time in seconds till a timeout exception is thrown when attempting to read from a connection.The default is 60 seconds. |
+| `read_timeout` | `float \| None` | No | 60.0 | The time in seconds till a timeout exception is thrown when attempting to read from a connection.The default is 60 seconds. |
 | `session_ttl` | `int \| None` | No | 3600 | The time in seconds till a session expires. The default is 3600 seconds (1 hour). |
 
 ## Sample Configuration

llama_stack/apis/common/errors.py

@@ -79,3 +79,10 @@ class ConflictError(ValueError):
 
     def __init__(self, message: str) -> None:
         super().__init__(message)
+
+
+class TokenValidationError(ValueError):
+    """raised when token validation fails during authentication"""
+
+    def __init__(self, message: str) -> None:
+        super().__init__(message)

llama_stack/apis/datatypes.py

@@ -102,6 +102,7 @@ class Api(Enum, metaclass=DynamicApiMeta):
     :cvar benchmarks: Benchmark suite management
     :cvar tool_groups: Tool group organization
     :cvar files: File storage and management
+    :cvar prompts: Prompt versions and management
     :cvar inspect: Built-in system inspection and introspection
     """
 
@@ -127,6 +128,7 @@ class Api(Enum, metaclass=DynamicApiMeta):
     benchmarks = "benchmarks"
     tool_groups = "tool_groups"
     files = "files"
+    prompts = "prompts"
 
     # built-in API
     inspect = "inspect"

llama_stack/apis/prompts/__init__.py

@@ -0,0 +1,9 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+from .prompts import ListPromptsResponse, Prompt, Prompts
+
+__all__ = ["Prompt", "Prompts", "ListPromptsResponse"]

llama_stack/apis/prompts/prompts.py

@@ -0,0 +1,189 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+import re
+import secrets
+from typing import Protocol, runtime_checkable
+
+from pydantic import BaseModel, Field, field_validator, model_validator
+
+from llama_stack.providers.utils.telemetry.trace_protocol import trace_protocol
+from llama_stack.schema_utils import json_schema_type, webmethod
+
+
+@json_schema_type
+class Prompt(BaseModel):
+    """A prompt resource representing a stored OpenAI Compatible prompt template in Llama Stack.
+
+    :param prompt: The system prompt text with variable placeholders. Variables are only supported when using the Responses API.
+    :param version: Version (integer starting at 1, incremented on save)
+    :param prompt_id: Unique identifier formatted as 'pmpt_<48-digit-hash>'
+    :param variables: List of prompt variable names that can be used in the prompt template
+    :param is_default: Boolean indicating whether this version is the default version for this prompt
+    """
+
+    prompt: str | None = Field(default=None, description="The system prompt with variable placeholders")
+    version: int = Field(description="Version (integer starting at 1, incremented on save)", ge=1)
+    prompt_id: str = Field(description="Unique identifier in format 'pmpt_<48-digit-hash>'")
+    variables: list[str] = Field(
+        default_factory=list, description="List of variable names that can be used in the prompt template"
+    )
+    is_default: bool = Field(
+        default=False, description="Boolean indicating whether this version is the default version"
+    )
+
+    @field_validator("prompt_id")
+    @classmethod
+    def validate_prompt_id(cls, prompt_id: str) -> str:
+        if not isinstance(prompt_id, str):
+            raise TypeError("prompt_id must be a string in format 'pmpt_<48-digit-hash>'")
+
+        if not prompt_id.startswith("pmpt_"):
+            raise ValueError("prompt_id must start with 'pmpt_' prefix")
+
+        hex_part = prompt_id[5:]
+        if len(hex_part) != 48:
+            raise ValueError("prompt_id must be in format 'pmpt_<48-digit-hash>' (48 lowercase hex chars)")
+
+        for char in hex_part:
+            if char not in "0123456789abcdef":
+                raise ValueError("prompt_id hex part must contain only lowercase hex characters [0-9a-f]")
+
+        return prompt_id
+
+    @field_validator("version")
+    @classmethod
+    def validate_version(cls, prompt_version: int) -> int:
+        if prompt_version < 1:
+            raise ValueError("version must be >= 1")
+        return prompt_version
+
+    @model_validator(mode="after")
+    def validate_prompt_variables(self):
+        """Validate that all variables used in the prompt are declared in the variables list."""
+        if not self.prompt:
+            return self
+
+        prompt_variables = set(re.findall(r"{{\s*(\w+)\s*}}", self.prompt))
+        declared_variables = set(self.variables)
+
+        undeclared = prompt_variables - declared_variables
+        if undeclared:
+            raise ValueError(f"Prompt contains undeclared variables: {sorted(undeclared)}")
+
+        return self
+
+    @classmethod
+    def generate_prompt_id(cls) -> str:
+        # Generate 48 hex characters (24 bytes)
+        random_bytes = secrets.token_bytes(24)
+        hex_string = random_bytes.hex()
+        return f"pmpt_{hex_string}"
+
+
+class ListPromptsResponse(BaseModel):
+    """Response model to list prompts."""
+
+    data: list[Prompt]
+
+
+@runtime_checkable
+@trace_protocol
+class Prompts(Protocol):
+    """Protocol for prompt management operations."""
+
+    @webmethod(route="/prompts", method="GET")
+    async def list_prompts(self) -> ListPromptsResponse:
+        """List all prompts.
+
+        :returns: A ListPromptsResponse containing all prompts.
+        """
+        ...
+
+    @webmethod(route="/prompts/{prompt_id}/versions", method="GET")
+    async def list_prompt_versions(
+        self,
+        prompt_id: str,
+    ) -> ListPromptsResponse:
+        """List all versions of a specific prompt.
+
+        :param prompt_id: The identifier of the prompt to list versions for.
+        :returns: A ListPromptsResponse containing all versions of the prompt.
+        """
+        ...
+
+    @webmethod(route="/prompts/{prompt_id}", method="GET")
+    async def get_prompt(
+        self,
+        prompt_id: str,
+        version: int | None = None,
+    ) -> Prompt:
+        """Get a prompt by its identifier and optional version.
+
+        :param prompt_id: The identifier of the prompt to get.
+        :param version: The version of the prompt to get (defaults to latest).
+        :returns: A Prompt resource.
+        """
+        ...
+
+    @webmethod(route="/prompts", method="POST")
+    async def create_prompt(
+        self,
+        prompt: str,
+        variables: list[str] | None = None,
+    ) -> Prompt:
+        """Create a new prompt.
+
+        :param prompt: The prompt text content with variable placeholders.
+        :param variables: List of variable names that can be used in the prompt template.
+        :returns: The created Prompt resource.
+        """
+        ...
+
+    @webmethod(route="/prompts/{prompt_id}", method="PUT")
+    async def update_prompt(
+        self,
+        prompt_id: str,
+        prompt: str,
+        version: int,
+        variables: list[str] | None = None,
+        set_as_default: bool = True,
+    ) -> Prompt:
+        """Update an existing prompt (increments version).
+
+        :param prompt_id: The identifier of the prompt to update.
+        :param prompt: The updated prompt text content.
+        :param version: The current version of the prompt being updated.
+        :param variables: Updated list of variable names that can be used in the prompt template.
+        :param set_as_default: Set the new version as the default (default=True).
+        :returns: The updated Prompt resource with incremented version.
+        """
+        ...
+
+    @webmethod(route="/prompts/{prompt_id}", method="DELETE")
+    async def delete_prompt(
+        self,
+        prompt_id: str,
+    ) -> None:
+        """Delete a prompt.
+
+        :param prompt_id: The identifier of the prompt to delete.
+        """
+        ...
+
+    @webmethod(route="/prompts/{prompt_id}/set-default-version", method="PUT")
+    async def set_default_version(
+        self,
+        prompt_id: str,
+        version: int,
+    ) -> Prompt:
+        """Set which version of a prompt should be the default in get_prompt (latest).
+
+        :param prompt_id: The identifier of the prompt.
+        :param version: The version to set as default.
+        :returns: The prompt with the specified version now set as default.
+        """
+        ...

llama_stack/apis/resource.py

@@ -19,6 +19,7 @@ class ResourceType(StrEnum):
     benchmark = "benchmark"
     tool = "tool"
     tool_group = "tool_group"
+    prompt = "prompt"
 
 
 class Resource(BaseModel):

llama_stack/cli/stack/_build.py

@@ -45,6 +45,7 @@ from llama_stack.core.utils.dynamic import instantiate_class_type
 from llama_stack.core.utils.exec import formulate_run_args, run_command
 from llama_stack.core.utils.image_types import LlamaStackImageType
 from llama_stack.providers.datatypes import Api
+from llama_stack.providers.utils.sqlstore.sqlstore import SqliteSqlStoreConfig
 
 DISTRIBS_PATH = Path(__file__).parent.parent.parent / "distributions"
 
@@ -294,6 +295,12 @@ def _generate_run_config(
         if build_config.external_providers_dir
         else EXTERNAL_PROVIDERS_DIR,
     )
+    if not run_config.inference_store:
+        run_config.inference_store = SqliteSqlStoreConfig(
+            **SqliteSqlStoreConfig.sample_run_config(
+                __distro_dir__=(DISTRIBS_BASE_DIR / image_name).as_posix(), db_name="inference_store.db"
+            )
+        )
     # build providers dict
     provider_registry = get_provider_registry(build_config)
     for api in apis:

llama_stack/core/datatypes.py

@@ -7,6 +7,7 @@
 from enum import StrEnum
 from pathlib import Path
 from typing import Annotated, Any, Literal, Self
+from urllib.parse import urlparse
 
 from pydantic import BaseModel, Field, field_validator, model_validator
 
@@ -212,6 +213,7 @@ class AuthProviderType(StrEnum):
     OAUTH2_TOKEN = "oauth2_token"
     GITHUB_TOKEN = "github_token"
     CUSTOM = "custom"
+    KUBERNETES = "kubernetes"
 
 
 class OAuth2TokenAuthConfig(BaseModel):
@@ -282,8 +284,45 @@ class GitHubTokenAuthConfig(BaseModel):
     )
 
 
+class KubernetesAuthProviderConfig(BaseModel):
+    """Configuration for Kubernetes authentication provider."""
+
+    type: Literal[AuthProviderType.KUBERNETES] = AuthProviderType.KUBERNETES
+    api_server_url: str = Field(
+        default="https://kubernetes.default.svc",
+        description="Kubernetes API server URL (e.g., https://api.cluster.domain:6443)",
+    )
+    verify_tls: bool = Field(default=True, description="Whether to verify TLS certificates")
+    tls_cafile: Path | None = Field(default=None, description="Path to CA certificate file for TLS verification")
+    claims_mapping: dict[str, str] = Field(
+        default_factory=lambda: {
+            "username": "roles",
+            "groups": "roles",
+        },
+        description="Mapping of Kubernetes user claims to access attributes",
+    )
+
+    @field_validator("api_server_url")
+    @classmethod
+    def validate_api_server_url(cls, v):
+        parsed = urlparse(v)
+        if not parsed.scheme or not parsed.netloc:
+            raise ValueError(f"api_server_url must be a valid URL with scheme and host: {v}")
+        if parsed.scheme not in ["http", "https"]:
+            raise ValueError(f"api_server_url scheme must be http or https: {v}")
+        return v
+
+    @field_validator("claims_mapping")
+    @classmethod
+    def validate_claims_mapping(cls, v):
+        for key, value in v.items():
+            if not value:
+                raise ValueError(f"claims_mapping value cannot be empty: {key}")
+        return v
+
+
 AuthProviderConfig = Annotated[
-    OAuth2TokenAuthConfig | GitHubTokenAuthConfig | CustomAuthConfig,
+    OAuth2TokenAuthConfig | GitHubTokenAuthConfig | CustomAuthConfig | KubernetesAuthProviderConfig,
     Field(discriminator="type"),
 ]
 
@@ -392,6 +431,12 @@ class ServerConfig(BaseModel):
     )
 
 
+class InferenceStoreConfig(BaseModel):
+    sql_store_config: SqlStoreConfig
+    max_write_queue_size: int = Field(default=10000, description="Max queued writes for inference store")
+    num_writers: int = Field(default=4, description="Number of concurrent background writers")
+
+
 class StackRunConfig(BaseModel):
     version: int = LLAMA_STACK_RUN_CONFIG_VERSION
 
@@ -425,11 +470,12 @@ Configuration for the persistence store used by the distribution registry. If no
 a default SQLite store will be used.""",
     )
 
-    inference_store: SqlStoreConfig | None = Field(
+    inference_store: InferenceStoreConfig | SqlStoreConfig | None = Field(
         default=None,
         description="""
-Configuration for the persistence store used by the inference API. If not specified,
+Configuration for the persistence store used by the inference API. Can be either a
-a default SQLite store will be used.""",
+InferenceStoreConfig (with queue tuning parameters) or a SqlStoreConfig (deprecated).
+If not specified, a default SQLite store will be used.""",
     )
 
     # registry of "resources" in the distribution

llama_stack/core/library_client.py

@@ -10,7 +10,6 @@ import json
 import logging  # allow-direct-logging
 import os
 import sys
-from concurrent.futures import ThreadPoolExecutor
 from enum import Enum
 from io import BytesIO
 from pathlib import Path
@@ -148,7 +147,6 @@ class LlamaStackAsLibraryClient(LlamaStackClient):
         self.async_client = AsyncLlamaStackAsLibraryClient(
             config_path_or_distro_name, custom_provider_registry, provider_data, skip_logger_removal
         )
-        self.pool_executor = ThreadPoolExecutor(max_workers=4)
         self.provider_data = provider_data
 
         self.loop = asyncio.new_event_loop()

llama_stack/core/prompts/prompts.py

@@ -0,0 +1,233 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+import json
+from typing import Any
+
+from pydantic import BaseModel
+
+from llama_stack.apis.prompts import ListPromptsResponse, Prompt, Prompts
+from llama_stack.core.datatypes import StackRunConfig
+from llama_stack.core.utils.config_dirs import DISTRIBS_BASE_DIR
+from llama_stack.providers.utils.kvstore import KVStore, kvstore_impl
+from llama_stack.providers.utils.kvstore.config import SqliteKVStoreConfig
+
+
+class PromptServiceConfig(BaseModel):
+    """Configuration for the built-in prompt service.
+
+    :param run_config: Stack run configuration containing distribution info
+    """
+
+    run_config: StackRunConfig
+
+
+async def get_provider_impl(config: PromptServiceConfig, deps: dict[Any, Any]):
+    """Get the prompt service implementation."""
+    impl = PromptServiceImpl(config, deps)
+    await impl.initialize()
+    return impl
+
+
+class PromptServiceImpl(Prompts):
+    """Built-in prompt service implementation using KVStore."""
+
+    def __init__(self, config: PromptServiceConfig, deps: dict[Any, Any]):
+        self.config = config
+        self.deps = deps
+        self.kvstore: KVStore
+
+    async def initialize(self) -> None:
+        kvstore_config = SqliteKVStoreConfig(
+            db_path=(DISTRIBS_BASE_DIR / self.config.run_config.image_name / "prompts.db").as_posix()
+        )
+        self.kvstore = await kvstore_impl(kvstore_config)
+
+    def _get_default_key(self, prompt_id: str) -> str:
+        """Get the KVStore key that stores the default version number."""
+        return f"prompts:v1:{prompt_id}:default"
+
+    async def _get_prompt_key(self, prompt_id: str, version: int | None = None) -> str:
+        """Get the KVStore key for prompt data, returning default version if applicable."""
+        if version:
+            return self._get_version_key(prompt_id, str(version))
+
+        default_key = self._get_default_key(prompt_id)
+        resolved_version = await self.kvstore.get(default_key)
+        if resolved_version is None:
+            raise ValueError(f"Prompt {prompt_id}:default not found")
+        return self._get_version_key(prompt_id, resolved_version)
+
+    def _get_version_key(self, prompt_id: str, version: str) -> str:
+        """Get the KVStore key for a specific prompt version."""
+        return f"prompts:v1:{prompt_id}:{version}"
+
+    def _get_list_key_prefix(self) -> str:
+        """Get the key prefix for listing prompts."""
+        return "prompts:v1:"
+
+    def _serialize_prompt(self, prompt: Prompt) -> str:
+        """Serialize a prompt to JSON string for storage."""
+        return json.dumps(
+            {
+                "prompt_id": prompt.prompt_id,
+                "prompt": prompt.prompt,
+                "version": prompt.version,
+                "variables": prompt.variables or [],
+                "is_default": prompt.is_default,
+            }
+        )
+
+    def _deserialize_prompt(self, data: str) -> Prompt:
+        """Deserialize a prompt from JSON string."""
+        obj = json.loads(data)
+        return Prompt(
+            prompt_id=obj["prompt_id"],
+            prompt=obj["prompt"],
+            version=obj["version"],
+            variables=obj.get("variables", []),
+            is_default=obj.get("is_default", False),
+        )
+
+    async def list_prompts(self) -> ListPromptsResponse:
+        """List all prompts (default versions only)."""
+        prefix = self._get_list_key_prefix()
+        keys = await self.kvstore.keys_in_range(prefix, prefix + "\xff")
+
+        prompts = []
+        for key in keys:
+            if key.endswith(":default"):
+                try:
+                    default_version = await self.kvstore.get(key)
+                    if default_version:
+                        prompt_id = key.replace(prefix, "").replace(":default", "")
+                        version_key = self._get_version_key(prompt_id, default_version)
+                        data = await self.kvstore.get(version_key)
+                        if data:
+                            prompt = self._deserialize_prompt(data)
+                            prompts.append(prompt)
+                except (json.JSONDecodeError, KeyError):
+                    continue
+
+        prompts.sort(key=lambda p: p.prompt_id or "", reverse=True)
+        return ListPromptsResponse(data=prompts)
+
+    async def get_prompt(self, prompt_id: str, version: int | None = None) -> Prompt:
+        """Get a prompt by its identifier and optional version."""
+        key = await self._get_prompt_key(prompt_id, version)
+        data = await self.kvstore.get(key)
+        if data is None:
+            raise ValueError(f"Prompt {prompt_id}:{version if version else 'default'} not found")
+        return self._deserialize_prompt(data)
+
+    async def create_prompt(
+        self,
+        prompt: str,
+        variables: list[str] | None = None,
+    ) -> Prompt:
+        """Create a new prompt."""
+        if variables is None:
+            variables = []
+
+        prompt_obj = Prompt(
+            prompt_id=Prompt.generate_prompt_id(),
+            prompt=prompt,
+            version=1,
+            variables=variables,
+        )
+
+        version_key = self._get_version_key(prompt_obj.prompt_id, str(prompt_obj.version))
+        data = self._serialize_prompt(prompt_obj)
+        await self.kvstore.set(version_key, data)
+
+        default_key = self._get_default_key(prompt_obj.prompt_id)
+        await self.kvstore.set(default_key, str(prompt_obj.version))
+
+        return prompt_obj
+
+    async def update_prompt(
+        self,
+        prompt_id: str,
+        prompt: str,
+        version: int,
+        variables: list[str] | None = None,
+        set_as_default: bool = True,
+    ) -> Prompt:
+        """Update an existing prompt (increments version)."""
+        if version < 1:
+            raise ValueError("Version must be >= 1")
+        if variables is None:
+            variables = []
+
+        prompt_versions = await self.list_prompt_versions(prompt_id)
+        latest_prompt = max(prompt_versions.data, key=lambda x: int(x.version))
+
+        if version and latest_prompt.version != version:
+            raise ValueError(
+                f"'{version}' is not the latest prompt version for prompt_id='{prompt_id}'. Use the latest version '{latest_prompt.version}' in request."
+            )
+
+        current_version = latest_prompt.version if version is None else version
+        new_version = current_version + 1
+
+        updated_prompt = Prompt(prompt_id=prompt_id, prompt=prompt, version=new_version, variables=variables)
+
+        version_key = self._get_version_key(prompt_id, str(new_version))
+        data = self._serialize_prompt(updated_prompt)
+        await self.kvstore.set(version_key, data)
+
+        if set_as_default:
+            await self.set_default_version(prompt_id, new_version)
+
+        return updated_prompt
+
+    async def delete_prompt(self, prompt_id: str) -> None:
+        """Delete a prompt and all its versions."""
+        await self.get_prompt(prompt_id)
+
+        prefix = f"prompts:v1:{prompt_id}:"
+        keys = await self.kvstore.keys_in_range(prefix, prefix + "\xff")
+
+        for key in keys:
+            await self.kvstore.delete(key)
+
+    async def list_prompt_versions(self, prompt_id: str) -> ListPromptsResponse:
+        """List all versions of a specific prompt."""
+        prefix = f"prompts:v1:{prompt_id}:"
+        keys = await self.kvstore.keys_in_range(prefix, prefix + "\xff")
+
+        default_version = None
+        prompts = []
+
+        for key in keys:
+            data = await self.kvstore.get(key)
+            if key.endswith(":default"):
+                default_version = data
+            else:
+                if data:
+                    prompt_obj = self._deserialize_prompt(data)
+                    prompts.append(prompt_obj)
+
+        if not prompts:
+            raise ValueError(f"Prompt {prompt_id} not found")
+
+        for prompt in prompts:
+            prompt.is_default = str(prompt.version) == default_version
+
+        prompts.sort(key=lambda x: x.version)
+        return ListPromptsResponse(data=prompts)
+
+    async def set_default_version(self, prompt_id: str, version: int) -> Prompt:
+        """Set which version of a prompt should be the default, If not set. the default is the latest."""
+        version_key = self._get_version_key(prompt_id, str(version))
+        data = await self.kvstore.get(version_key)
+        if data is None:
+            raise ValueError(f"Prompt {prompt_id} version {version} not found")
+
+        default_key = self._get_default_key(prompt_id)
+        await self.kvstore.set(default_key, str(version))
+
+        return self._deserialize_prompt(data)

llama_stack/core/resolver.py

@@ -19,6 +19,7 @@ from llama_stack.apis.inference import Inference, InferenceProvider
 from llama_stack.apis.inspect import Inspect
 from llama_stack.apis.models import Models
 from llama_stack.apis.post_training import PostTraining
+from llama_stack.apis.prompts import Prompts
 from llama_stack.apis.providers import Providers as ProvidersAPI
 from llama_stack.apis.safety import Safety
 from llama_stack.apis.scoring import Scoring
@@ -93,6 +94,7 @@ def api_protocol_map(external_apis: dict[Api, ExternalApiSpec] | None = None) ->
         Api.tool_groups: ToolGroups,
         Api.tool_runtime: ToolRuntime,
         Api.files: Files,
+        Api.prompts: Prompts,
     }
 
     if external_apis:
@@ -284,7 +286,15 @@ async def instantiate_providers(
         if provider.provider_id is None:
             continue
 
+        try:
             deps = {a: impls[a] for a in provider.spec.api_dependencies}
+        except KeyError as e:
+            missing_api = e.args[0]
+            raise RuntimeError(
+                f"Failed to resolve '{provider.spec.api.value}' provider '{provider.provider_id}' of type '{provider.spec.provider_type}': "
+                f"required dependency '{missing_api.value}' is not available. "
+                f"Please add a '{missing_api.value}' provider to your configuration or check if the provider is properly configured."
+            ) from e
         for a in provider.spec.optional_api_dependencies:
             if a in impls:
                 deps[a] = impls[a]

llama_stack/core/routers/__init__.py

@@ -78,7 +78,10 @@ async def get_auto_router_impl(
 
     # TODO: move pass configs to routers instead
     if api == Api.inference and run_config.inference_store:
-        inference_store = InferenceStore(run_config.inference_store, policy)
+        inference_store = InferenceStore(
+            config=run_config.inference_store,
+            policy=policy,
+        )
         await inference_store.initialize()
         api_to_dep_impl["store"] = inference_store
 

llama_stack/core/routers/inference.py

@@ -63,7 +63,7 @@ from llama_stack.models.llama.llama3.chat_format import ChatFormat
 from llama_stack.models.llama.llama3.tokenizer import Tokenizer
 from llama_stack.providers.datatypes import HealthResponse, HealthStatus, RoutingTable
 from llama_stack.providers.utils.inference.inference_store import InferenceStore
-from llama_stack.providers.utils.telemetry.tracing import get_current_span
+from llama_stack.providers.utils.telemetry.tracing import enqueue_event, get_current_span
 
 logger = get_logger(name=__name__, category="core::routers")
 
@@ -90,6 +90,11 @@ class InferenceRouter(Inference):
 
     async def shutdown(self) -> None:
         logger.debug("InferenceRouter.shutdown")
+        if self.store:
+            try:
+                await self.store.shutdown()
+            except Exception as e:
+                logger.warning(f"Error during InferenceStore shutdown: {e}")
 
     async def register_model(
         self,
@@ -160,7 +165,7 @@ class InferenceRouter(Inference):
         metrics = self._construct_metrics(prompt_tokens, completion_tokens, total_tokens, model)
         if self.telemetry:
             for metric in metrics:
-                await self.telemetry.log_event(metric)
+                enqueue_event(metric)
         return [MetricInResponse(metric=metric.metric, value=metric.value) for metric in metrics]
 
     async def _count_tokens(
@@ -431,7 +436,7 @@ class InferenceRouter(Inference):
                 model=model_obj,
             )
             for metric in metrics:
-                await self.telemetry.log_event(metric)
+                enqueue_event(metric)
 
             # these metrics will show up in the client response.
             response.metrics = (
@@ -527,7 +532,7 @@ class InferenceRouter(Inference):
 
         # Store the response with the ID that will be returned to the client
         if self.store:
-            await self.store.store_chat_completion(response, messages)
+            asyncio.create_task(self.store.store_chat_completion(response, messages))
 
         if self.telemetry:
             metrics = self._construct_metrics(
@@ -537,7 +542,7 @@ class InferenceRouter(Inference):
                 model=model_obj,
             )
             for metric in metrics:
-                await self.telemetry.log_event(metric)
+                enqueue_event(metric)
             # these metrics will show up in the client response.
             response.metrics = (
                 metrics if not hasattr(response, "metrics") or response.metrics is None else response.metrics + metrics
@@ -664,7 +669,7 @@ class InferenceRouter(Inference):
                             "completion_tokens",
                             "total_tokens",
                         ]:  # Only log completion and total tokens
-                            await self.telemetry.log_event(metric)
+                            enqueue_event(metric)
 
                         # Return metrics in response
                         async_metrics = [
@@ -710,7 +715,7 @@ class InferenceRouter(Inference):
             )
             for metric in completion_metrics:
                 if metric.metric in ["completion_tokens", "total_tokens"]:  # Only log completion and total tokens
-                    await self.telemetry.log_event(metric)
+                    enqueue_event(metric)
 
             # Return metrics in response
             return [MetricInResponse(metric=metric.metric, value=metric.value) for metric in completion_metrics]
@@ -755,7 +760,7 @@ class InferenceRouter(Inference):
                             choices_data[idx] = {
                                 "content_parts": [],
                                 "tool_calls_builder": {},
-                                "finish_reason": None,
+                                "finish_reason": "stop",
                                 "logprobs_content_parts": [],
                             }
                         current_choice_data = choices_data[idx]
@@ -806,7 +811,7 @@ class InferenceRouter(Inference):
                             model=model,
                         )
                         for metric in metrics:
-                            await self.telemetry.log_event(metric)
+                            enqueue_event(metric)
 
                 yield chunk
         finally:
@@ -855,4 +860,4 @@ class InferenceRouter(Inference):
                     object="chat.completion",
                 )
                 logger.debug(f"InferenceRouter.completion_response: {final_response}")
-                await self.store.store_chat_completion(final_response, messages)
+                asyncio.create_task(self.store.store_chat_completion(final_response, messages))

llama_stack/core/routing_tables/vector_dbs.py

@@ -53,6 +53,7 @@ class VectorDBsRoutingTable(CommonRoutingTableImpl, VectorDBs):
         vector_db_name: str | None = None,
     ) -> VectorDB:
         provider_vector_db_id = provider_vector_db_id or vector_db_id
+
         model = await lookup_model(self, embedding_model)
         if model is None:
             raise ModelNotFoundError(embedding_model)
@@ -60,14 +61,33 @@ class VectorDBsRoutingTable(CommonRoutingTableImpl, VectorDBs):
             raise ModelTypeError(embedding_model, model.model_type, ModelType.embedding)
         if "embedding_dimension" not in model.metadata:
             raise ValueError(f"Model {embedding_model} does not have an embedding dimension")
+
+        provider = self.impls_by_provider_id[provider_id]
+        logger.warning(
+            "VectorDB is being deprecated in future releases in favor of VectorStore. Please migrate your usage accordingly."
+        )
+        vector_store = await provider.openai_create_vector_store(
+            name=vector_db_name or vector_db_id,
+            embedding_model=embedding_model,
+            embedding_dimension=model.metadata["embedding_dimension"],
+            provider_id=provider_id,
+            provider_vector_db_id=provider_vector_db_id,
+        )
+
+        vector_store_id = vector_store.id
+        actual_provider_vector_db_id = provider_vector_db_id or vector_store_id
+        logger.warning(
+            f"Ignoring vector_db_id {vector_db_id} and using vector_store_id {vector_store_id} instead. Setting VectorDB {vector_db_id} to VectorDB.vector_db_name"
+        )
+
         vector_db_data = {
-            "identifier": vector_db_id,
+            "identifier": vector_store_id,
             "type": ResourceType.vector_db.value,
             "provider_id": provider_id,
-            "provider_resource_id": provider_vector_db_id,
+            "provider_resource_id": actual_provider_vector_db_id,
             "embedding_model": embedding_model,
             "embedding_dimension": model.metadata["embedding_dimension"],
-            "vector_db_name": vector_db_name,
+            "vector_db_name": vector_store.name,
         }
         vector_db = TypeAdapter(VectorDBWithOwner).validate_python(vector_db_data)
         await self.register_object(vector_db)

llama_stack/core/server/auth_providers.py

@@ -8,16 +8,18 @@ import ssl
 import time
 from abc import ABC, abstractmethod
 from asyncio import Lock
-from urllib.parse import parse_qs, urlparse
+from urllib.parse import parse_qs, urljoin, urlparse
 
 import httpx
 from jose import jwt
 from pydantic import BaseModel, Field
 
+from llama_stack.apis.common.errors import TokenValidationError
 from llama_stack.core.datatypes import (
     AuthenticationConfig,
     CustomAuthConfig,
     GitHubTokenAuthConfig,
+    KubernetesAuthProviderConfig,
     OAuth2TokenAuthConfig,
     User,
 )
@@ -162,7 +164,7 @@ class OAuth2TokenAuthProvider(AuthProvider):
                     auth=auth,
                     timeout=10.0,  # Add a reasonable timeout
                 )
-                if response.status_code != 200:
+                if response.status_code != httpx.codes.OK:
                     logger.warning(f"Token introspection failed with status code: {response.status_code}")
                     raise ValueError(f"Token introspection failed: {response.status_code}")
 
@@ -272,7 +274,7 @@ class CustomAuthProvider(AuthProvider):
                     json=auth_request.model_dump(),
                     timeout=10.0,  # Add a reasonable timeout
                 )
-                if response.status_code != 200:
+                if response.status_code != httpx.codes.OK:
                     logger.warning(f"Authentication failed with status code: {response.status_code}")
                     raise ValueError(f"Authentication failed: {response.status_code}")
 
@@ -374,6 +376,89 @@ async def _get_github_user_info(access_token: str, github_api_base_url: str) ->
         }
 
 
+class KubernetesAuthProvider(AuthProvider):
+    """
+    Kubernetes authentication provider that validates tokens using the Kubernetes SelfSubjectReview API.
+    This provider integrates with Kubernetes API server by using the
+    /apis/authentication.k8s.io/v1/selfsubjectreviews endpoint to validate tokens and extract user information.
+    """
+
+    def __init__(self, config: KubernetesAuthProviderConfig):
+        self.config = config
+
+    def _httpx_verify_value(self) -> bool | str:
+        """
+        Build the value for httpx's `verify` parameter.
+        - False disables verification.
+        - Path string points to a CA bundle.
+        - True uses system defaults.
+        """
+        if not self.config.verify_tls:
+            return False
+        if self.config.tls_cafile:
+            return self.config.tls_cafile.as_posix()
+        return True
+
+    async def validate_token(self, token: str, scope: dict | None = None) -> User:
+        """Validate a token using Kubernetes SelfSubjectReview API endpoint."""
+        # Build the Kubernetes SelfSubjectReview API endpoint URL
+        review_api_url = urljoin(self.config.api_server_url, "/apis/authentication.k8s.io/v1/selfsubjectreviews")
+
+        # Create SelfSubjectReview request body
+        review_request = {"apiVersion": "authentication.k8s.io/v1", "kind": "SelfSubjectReview"}
+        verify = self._httpx_verify_value()
+
+        try:
+            async with httpx.AsyncClient(verify=verify, timeout=10.0) as client:
+                response = await client.post(
+                    review_api_url,
+                    json=review_request,
+                    headers={
+                        "Authorization": f"Bearer {token}",
+                        "Content-Type": "application/json",
+                    },
+                )
+
+                if response.status_code == httpx.codes.UNAUTHORIZED:
+                    raise TokenValidationError("Invalid token")
+                if response.status_code != httpx.codes.CREATED:
+                    logger.warning(f"Kubernetes SelfSubjectReview API failed with status code: {response.status_code}")
+                    raise TokenValidationError(f"Token validation failed: {response.status_code}")
+
+                review_response = response.json()
+                # Extract user information from SelfSubjectReview response
+                status = review_response.get("status", {})
+                if not status:
+                    raise ValueError("No status found in SelfSubjectReview response")
+
+                user_info = status.get("userInfo", {})
+                if not user_info:
+                    raise ValueError("No userInfo found in SelfSubjectReview response")
+
+                username = user_info.get("username")
+                if not username:
+                    raise ValueError("No username found in SelfSubjectReview response")
+
+                # Build user attributes from Kubernetes user info
+                user_attributes = get_attributes_from_claims(user_info, self.config.claims_mapping)
+
+                return User(
+                    principal=username,
+                    attributes=user_attributes,
+                )
+
+        except httpx.TimeoutException:
+            logger.warning("Kubernetes SelfSubjectReview API request timed out")
+            raise ValueError("Token validation timeout") from None
+        except Exception as e:
+            logger.warning(f"Error during token validation: {str(e)}")
+            raise ValueError(f"Token validation error: {str(e)}") from e
+
+    async def close(self):
+        """Close any resources."""
+        pass
+
+
 def create_auth_provider(config: AuthenticationConfig) -> AuthProvider:
     """Factory function to create the appropriate auth provider."""
     provider_config = config.provider_config
@@ -384,5 +469,7 @@ def create_auth_provider(config: AuthenticationConfig) -> AuthProvider:
         return OAuth2TokenAuthProvider(provider_config)
     elif isinstance(provider_config, GitHubTokenAuthConfig):
         return GitHubTokenAuthProvider(provider_config)
+    elif isinstance(provider_config, KubernetesAuthProviderConfig):
+        return KubernetesAuthProvider(provider_config)
     else:
         raise ValueError(f"Unknown authentication provider config type: {type(provider_config)}")

llama_stack/core/server/server.py

@@ -132,15 +132,17 @@ def translate_exception(exc: Exception) -> HTTPException | RequestValidationErro
             },
         )
     elif isinstance(exc, ConflictError):
-        return HTTPException(status_code=409, detail=str(exc))
+        return HTTPException(status_code=httpx.codes.CONFLICT, detail=str(exc))
     elif isinstance(exc, ResourceNotFoundError):
-        return HTTPException(status_code=404, detail=str(exc))
+        return HTTPException(status_code=httpx.codes.NOT_FOUND, detail=str(exc))
     elif isinstance(exc, ValueError):
         return HTTPException(status_code=httpx.codes.BAD_REQUEST, detail=f"Invalid value: {str(exc)}")
     elif isinstance(exc, BadRequestError):
         return HTTPException(status_code=httpx.codes.BAD_REQUEST, detail=str(exc))
     elif isinstance(exc, PermissionError | AccessDeniedError):
         return HTTPException(status_code=httpx.codes.FORBIDDEN, detail=f"Permission denied: {str(exc)}")
+    elif isinstance(exc, ConnectionError | httpx.ConnectError):
+        return HTTPException(status_code=httpx.codes.BAD_GATEWAY, detail=str(exc))
     elif isinstance(exc, asyncio.TimeoutError | TimeoutError):
         return HTTPException(status_code=httpx.codes.GATEWAY_TIMEOUT, detail=f"Operation timed out: {str(exc)}")
     elif isinstance(exc, NotImplementedError):
@@ -513,6 +515,7 @@ def main(args: argparse.Namespace | None = None):
 
     apis_to_serve.add("inspect")
     apis_to_serve.add("providers")
+    apis_to_serve.add("prompts")
     for api_str in apis_to_serve:
         api = Api(api_str)
 

llama_stack/core/stack.py

@@ -24,6 +24,7 @@ from llama_stack.apis.inference import Inference
 from llama_stack.apis.inspect import Inspect
 from llama_stack.apis.models import Models
 from llama_stack.apis.post_training import PostTraining
+from llama_stack.apis.prompts import Prompts
 from llama_stack.apis.providers import Providers
 from llama_stack.apis.safety import Safety
 from llama_stack.apis.scoring import Scoring
@@ -37,6 +38,7 @@ from llama_stack.apis.vector_io import VectorIO
 from llama_stack.core.datatypes import Provider, StackRunConfig
 from llama_stack.core.distribution import get_provider_registry
 from llama_stack.core.inspect import DistributionInspectConfig, DistributionInspectImpl
+from llama_stack.core.prompts.prompts import PromptServiceConfig, PromptServiceImpl
 from llama_stack.core.providers import ProviderImpl, ProviderImplConfig
 from llama_stack.core.resolver import ProviderRegistry, resolve_impls
 from llama_stack.core.routing_tables.common import CommonRoutingTableImpl
@@ -72,6 +74,7 @@ class LlamaStack(
     ToolRuntime,
     RAGToolRuntime,
     Files,
+    Prompts,
 ):
     pass
 
@@ -305,6 +308,12 @@ def add_internal_implementations(impls: dict[Api, Any], run_config: StackRunConf
     )
     impls[Api.providers] = providers_impl
 
+    prompts_impl = PromptServiceImpl(
+        PromptServiceConfig(run_config=run_config),
+        deps=impls,
+    )
+    impls[Api.prompts] = prompts_impl
+
 
 # Produces a stack of providers for the given run config. Not all APIs may be
 # asked for in the run config.
@@ -329,6 +338,9 @@ async def construct_stack(
     # Add internal implementations after all other providers are resolved
     add_internal_implementations(impls, run_config)
 
+    if Api.prompts in impls:
+        await impls[Api.prompts].initialize()
+
     await register_resources(run_config, impls)
 
     await refresh_registry_once(impls)

llama_stack/distributions/ci-tests/ci_tests.py

@@ -11,9 +11,7 @@ from ..starter.starter import get_distribution_template as get_starter_distribut
 
 
 def get_distribution_template() -> DistributionTemplate:
-    template = get_starter_distribution_template()
+    template = get_starter_distribution_template(name="ci-tests")
-    name = "ci-tests"
-    template.name = name
     template.description = "CI tests for Llama Stack"
 
     return template

llama_stack/distributions/ci-tests/run.yaml

@@ -89,28 +89,28 @@ providers:
     config:
       kvstore:
         type: sqlite
-        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/faiss_store.db
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/ci-tests}/faiss_store.db
   - provider_id: sqlite-vec
     provider_type: inline::sqlite-vec
     config:
-      db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/sqlite_vec.db
+      db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/ci-tests}/sqlite_vec.db
       kvstore:
         type: sqlite
-        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/sqlite_vec_registry.db
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/ci-tests}/sqlite_vec_registry.db
   - provider_id: ${env.MILVUS_URL:+milvus}
     provider_type: inline::milvus
     config:
-      db_path: ${env.MILVUS_DB_PATH:=~/.llama/distributions/starter}/milvus.db
+      db_path: ${env.MILVUS_DB_PATH:=~/.llama/distributions/ci-tests}/milvus.db
       kvstore:
         type: sqlite
-        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/milvus_registry.db
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/ci-tests}/milvus_registry.db
   - provider_id: ${env.CHROMADB_URL:+chromadb}
     provider_type: remote::chromadb
     config:
       url: ${env.CHROMADB_URL:=}
       kvstore:
         type: sqlite
-        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter/}/chroma_remote_registry.db
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/ci-tests/}/chroma_remote_registry.db
   - provider_id: ${env.PGVECTOR_DB:+pgvector}
     provider_type: remote::pgvector
     config:
@@ -121,15 +121,15 @@ providers:
       password: ${env.PGVECTOR_PASSWORD:=}
       kvstore:
         type: sqlite
-        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/pgvector_registry.db
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/ci-tests}/pgvector_registry.db
   files:
   - provider_id: meta-reference-files
     provider_type: inline::localfs
     config:
-      storage_dir: ${env.FILES_STORAGE_DIR:=~/.llama/distributions/starter/files}
+      storage_dir: ${env.FILES_STORAGE_DIR:=~/.llama/distributions/ci-tests/files}
       metadata_store:
         type: sqlite
-        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/files_metadata.db
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/ci-tests}/files_metadata.db
   safety:
   - provider_id: llama-guard
     provider_type: inline::llama-guard

llama_stack/distributions/starter-gpu/run.yaml

@@ -89,28 +89,28 @@ providers:
     config:
       kvstore:
         type: sqlite
-        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/faiss_store.db
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu}/faiss_store.db
   - provider_id: sqlite-vec
     provider_type: inline::sqlite-vec
     config:
-      db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/sqlite_vec.db
+      db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu}/sqlite_vec.db
       kvstore:
         type: sqlite
-        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/sqlite_vec_registry.db
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu}/sqlite_vec_registry.db
   - provider_id: ${env.MILVUS_URL:+milvus}
     provider_type: inline::milvus
     config:
-      db_path: ${env.MILVUS_DB_PATH:=~/.llama/distributions/starter}/milvus.db
+      db_path: ${env.MILVUS_DB_PATH:=~/.llama/distributions/starter-gpu}/milvus.db
       kvstore:
         type: sqlite
-        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/milvus_registry.db
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu}/milvus_registry.db
   - provider_id: ${env.CHROMADB_URL:+chromadb}
     provider_type: remote::chromadb
     config:
       url: ${env.CHROMADB_URL:=}
       kvstore:
         type: sqlite
-        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter/}/chroma_remote_registry.db
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu/}/chroma_remote_registry.db
   - provider_id: ${env.PGVECTOR_DB:+pgvector}
     provider_type: remote::pgvector
     config:
@@ -121,15 +121,15 @@ providers:
       password: ${env.PGVECTOR_PASSWORD:=}
       kvstore:
         type: sqlite
-        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/pgvector_registry.db
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu}/pgvector_registry.db
   files:
   - provider_id: meta-reference-files
     provider_type: inline::localfs
     config:
-      storage_dir: ${env.FILES_STORAGE_DIR:=~/.llama/distributions/starter/files}
+      storage_dir: ${env.FILES_STORAGE_DIR:=~/.llama/distributions/starter-gpu/files}
       metadata_store:
         type: sqlite
-        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/files_metadata.db
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu}/files_metadata.db
   safety:
   - provider_id: llama-guard
     provider_type: inline::llama-guard

llama_stack/distributions/starter-gpu/starter_gpu.py

@@ -11,9 +11,7 @@ from ..starter.starter import get_distribution_template as get_starter_distribut
 
 
 def get_distribution_template() -> DistributionTemplate:
-    template = get_starter_distribution_template()
+    template = get_starter_distribution_template(name="starter-gpu")
-    name = "starter-gpu"
-    template.name = name
     template.description = "Quick start template for running Llama Stack with several popular providers. This distribution is intended for GPU-enabled environments."
 
     template.providers["post_training"] = [

llama_stack/distributions/starter/starter.py

@@ -99,9 +99,8 @@ def get_remote_inference_providers() -> list[Provider]:
     return inference_providers
 
 
-def get_distribution_template() -> DistributionTemplate:
+def get_distribution_template(name: str = "starter") -> DistributionTemplate:
     remote_inference_providers = get_remote_inference_providers()
-    name = "starter"
 
     providers = {
         "inference": [BuildProvider(provider_type=p.provider_type, module=p.module) for p in remote_inference_providers]

llama_stack/providers/inline/batches/reference/batches.py

@@ -178,9 +178,9 @@ class ReferenceBatchesImpl(Batches):
 
         # TODO: set expiration time for garbage collection
 
-        if endpoint not in ["/v1/chat/completions"]:
+        if endpoint not in ["/v1/chat/completions", "/v1/completions"]:
             raise ValueError(
-                f"Invalid endpoint: {endpoint}. Supported values: /v1/chat/completions. Code: invalid_value. Param: endpoint",
+                f"Invalid endpoint: {endpoint}. Supported values: /v1/chat/completions, /v1/completions. Code: invalid_value. Param: endpoint",
             )
 
         if completion_window != "24h":
@@ -424,13 +424,21 @@ class ReferenceBatchesImpl(Batches):
                             )
                             valid = False
 
-                        for param, expected_type, type_string in [
+                        if batch.endpoint == "/v1/chat/completions":
+                            required_params = [
                                 ("model", str, "a string"),
                                 # messages is specific to /v1/chat/completions
                                 # we could skip validating messages here and let inference fail. however,
                                 # that would be a very expensive way to find out messages is wrong.
                                 ("messages", list, "an array"),  # TODO: allow messages to be a string?
-                        ]:
+                            ]
+                        else:  # /v1/completions
+                            required_params = [
+                                ("model", str, "a string"),
+                                ("prompt", str, "a string"),  # TODO: allow prompt to be a list of strings??
+                            ]
+
+                        for param, expected_type, type_string in required_params:
                             if param not in body:
                                 errors.append(
                                     BatchError(
@@ -591,6 +599,7 @@ class ReferenceBatchesImpl(Batches):
 
         try:
             # TODO(SECURITY): review body for security issues
+            if request.url == "/v1/chat/completions":
                 request.body["messages"] = [convert_to_openai_message_param(msg) for msg in request.body["messages"]]
                 chat_response = await self.inference_api.openai_chat_completion(**request.body)
 
@@ -605,6 +614,22 @@ class ReferenceBatchesImpl(Batches):
                         "body": chat_response.model_dump_json(),
                     },
                 }
+            else:  # /v1/completions
+                completion_response = await self.inference_api.openai_completion(**request.body)
+
+                # this is for mypy, we don't allow streaming so we'll get the right type
+                assert hasattr(completion_response, "model_dump_json"), (
+                    "Completion response must have model_dump_json method"
+                )
+                return {
+                    "id": request_id,
+                    "custom_id": request.custom_id,
+                    "response": {
+                        "status_code": 200,
+                        "request_id": request_id,
+                        "body": completion_response.model_dump_json(),
+                    },
+                }
         except Exception as e:
             logger.info(f"Error processing request {request.custom_id} in batch {batch_id}: {e}")
             return {

llama_stack/providers/inline/tool_runtime/rag/__init__.py

@@ -14,6 +14,6 @@ from .config import RagToolRuntimeConfig
 async def get_provider_impl(config: RagToolRuntimeConfig, deps: dict[Api, Any]):
     from .memory import MemoryToolRuntimeImpl
 
-    impl = MemoryToolRuntimeImpl(config, deps[Api.vector_io], deps[Api.inference])
+    impl = MemoryToolRuntimeImpl(config, deps[Api.vector_io], deps[Api.inference], deps[Api.files])
     await impl.initialize()
     return impl

llama_stack/providers/inline/tool_runtime/rag/memory.py

@@ -5,10 +5,15 @@
 # the root directory of this source tree.
 
 import asyncio
+import base64
+import io
+import mimetypes
 import secrets
 import string
 from typing import Any
 
+import httpx
+from fastapi import UploadFile
 from pydantic import TypeAdapter
 
 from llama_stack.apis.common.content_types import (
@@ -17,6 +22,7 @@ from llama_stack.apis.common.content_types import (
     InterleavedContentItem,
     TextContentItem,
 )
+from llama_stack.apis.files import Files, OpenAIFilePurpose
 from llama_stack.apis.inference import Inference
 from llama_stack.apis.tools import (
     ListToolDefsResponse,
@@ -30,13 +36,18 @@ from llama_stack.apis.tools import (
     ToolParameter,
     ToolRuntime,
 )
-from llama_stack.apis.vector_io import QueryChunksResponse, VectorIO
+from llama_stack.apis.vector_io import (
+    QueryChunksResponse,
+    VectorIO,
+    VectorStoreChunkingStrategyStatic,
+    VectorStoreChunkingStrategyStaticConfig,
+)
 from llama_stack.log import get_logger
 from llama_stack.providers.datatypes import ToolGroupsProtocolPrivate
 from llama_stack.providers.utils.inference.prompt_adapter import interleaved_content_as_str
 from llama_stack.providers.utils.memory.vector_store import (
     content_from_doc,
-    make_overlapped_chunks,
+    parse_data_url,
 )
 
 from .config import RagToolRuntimeConfig
@@ -55,10 +66,12 @@ class MemoryToolRuntimeImpl(ToolGroupsProtocolPrivate, ToolRuntime, RAGToolRunti
         config: RagToolRuntimeConfig,
         vector_io_api: VectorIO,
         inference_api: Inference,
+        files_api: Files,
     ):
         self.config = config
         self.vector_io_api = vector_io_api
         self.inference_api = inference_api
+        self.files_api = files_api
 
     async def initialize(self):
         pass
@@ -78,26 +91,49 @@ class MemoryToolRuntimeImpl(ToolGroupsProtocolPrivate, ToolRuntime, RAGToolRunti
         vector_db_id: str,
         chunk_size_in_tokens: int = 512,
     ) -> None:
-        chunks = []
+        if not documents:
-        for doc in documents:
-            content = await content_from_doc(doc)
-            # TODO: we should add enrichment here as URLs won't be added to the metadata by default
-            chunks.extend(
-                make_overlapped_chunks(
-                    doc.document_id,
-                    content,
-                    chunk_size_in_tokens,
-                    chunk_size_in_tokens // 4,
-                    doc.metadata,
-                )
-            )
-
-        if not chunks:
             return
 
-        await self.vector_io_api.insert_chunks(
+        for doc in documents:
-            chunks=chunks,
+            if isinstance(doc.content, URL):
-            vector_db_id=vector_db_id,
+                if doc.content.uri.startswith("data:"):
+                    parts = parse_data_url(doc.content.uri)
+                    file_data = base64.b64decode(parts["data"]) if parts["is_base64"] else parts["data"].encode()
+                    mime_type = parts["mimetype"]
+                else:
+                    async with httpx.AsyncClient() as client:
+                        response = await client.get(doc.content.uri)
+                        file_data = response.content
+                        mime_type = doc.mime_type or response.headers.get("content-type", "application/octet-stream")
+            else:
+                content_str = await content_from_doc(doc)
+                file_data = content_str.encode("utf-8")
+                mime_type = doc.mime_type or "text/plain"
+
+            file_extension = mimetypes.guess_extension(mime_type) or ".txt"
+            filename = doc.metadata.get("filename", f"{doc.document_id}{file_extension}")
+
+            file_obj = io.BytesIO(file_data)
+            file_obj.name = filename
+
+            upload_file = UploadFile(file=file_obj, filename=filename)
+
+            created_file = await self.files_api.openai_upload_file(
+                file=upload_file, purpose=OpenAIFilePurpose.ASSISTANTS
+            )
+
+            chunking_strategy = VectorStoreChunkingStrategyStatic(
+                static=VectorStoreChunkingStrategyStaticConfig(
+                    max_chunk_size_tokens=chunk_size_in_tokens,
+                    chunk_overlap_tokens=chunk_size_in_tokens // 4,
+                )
+            )
+
+            await self.vector_io_api.openai_attach_file_to_vector_store(
+                vector_store_id=vector_db_id,
+                file_id=created_file.id,
+                attributes=doc.metadata,
+                chunking_strategy=chunking_strategy,
             )
 
     async def query(
@@ -131,8 +167,18 @@ class MemoryToolRuntimeImpl(ToolGroupsProtocolPrivate, ToolRuntime, RAGToolRunti
             for vector_db_id in vector_db_ids
         ]
         results: list[QueryChunksResponse] = await asyncio.gather(*tasks)
-        chunks = [c for r in results for c in r.chunks]
+
-        scores = [s for r in results for s in r.scores]
+        chunks = []
+        scores = []
+
+        for vector_db_id, result in zip(vector_db_ids, results, strict=False):
+            for chunk, score in zip(result.chunks, result.scores, strict=False):
+                if not hasattr(chunk, "metadata") or chunk.metadata is None:
+                    chunk.metadata = {}
+                chunk.metadata["vector_db_id"] = vector_db_id
+
+                chunks.append(chunk)
+                scores.append(score)
 
         if not chunks:
             return RAGQueryResult(content=None)
@@ -167,6 +213,7 @@ class MemoryToolRuntimeImpl(ToolGroupsProtocolPrivate, ToolRuntime, RAGToolRunti
             metadata_keys_to_exclude_from_context = [
                 "token_count",
                 "metadata_token_count",
+                "vector_db_id",
             ]
             metadata_for_context = {}
             for k in chunk_metadata_keys_to_include_from_context:
@@ -191,6 +238,7 @@ class MemoryToolRuntimeImpl(ToolGroupsProtocolPrivate, ToolRuntime, RAGToolRunti
                 "document_ids": [c.metadata["document_id"] for c in chunks[: len(picked)]],
                 "chunks": [c.content for c in chunks[: len(picked)]],
                 "scores": scores[: len(picked)],
+                "vector_db_ids": [c.metadata["vector_db_id"] for c in chunks[: len(picked)]],
             },
         )
 

llama_stack/providers/inline/vector_io/sqlite_vec/sqlite_vec.py

@@ -30,11 +30,11 @@ from llama_stack.providers.utils.kvstore.api import KVStore
 from llama_stack.providers.utils.memory.openai_vector_store_mixin import OpenAIVectorStoreMixin
 from llama_stack.providers.utils.memory.vector_store import (
     RERANKER_TYPE_RRF,
-    RERANKER_TYPE_WEIGHTED,
     ChunkForDeletion,
     EmbeddingIndex,
     VectorDBWithIndex,
 )
+from llama_stack.providers.utils.vector_io.vector_utils import WeightedInMemoryAggregator
 
 logger = get_logger(name=__name__, category="vector_io")
 
@@ -66,59 +66,6 @@ def _create_sqlite_connection(db_path):
     return connection
 
 
-def _normalize_scores(scores: dict[str, float]) -> dict[str, float]:
-    """Normalize scores to [0,1] range using min-max normalization."""
-    if not scores:
-        return {}
-    min_score = min(scores.values())
-    max_score = max(scores.values())
-    score_range = max_score - min_score
-    if score_range > 0:
-        return {doc_id: (score - min_score) / score_range for doc_id, score in scores.items()}
-    return dict.fromkeys(scores, 1.0)
-
-
-def _weighted_rerank(
-    vector_scores: dict[str, float],
-    keyword_scores: dict[str, float],
-    alpha: float = 0.5,
-) -> dict[str, float]:
-    """ReRanker that uses weighted average of scores."""
-    all_ids = set(vector_scores.keys()) | set(keyword_scores.keys())
-    normalized_vector_scores = _normalize_scores(vector_scores)
-    normalized_keyword_scores = _normalize_scores(keyword_scores)
-
-    return {
-        doc_id: (alpha * normalized_keyword_scores.get(doc_id, 0.0))
-        + ((1 - alpha) * normalized_vector_scores.get(doc_id, 0.0))
-        for doc_id in all_ids
-    }
-
-
-def _rrf_rerank(
-    vector_scores: dict[str, float],
-    keyword_scores: dict[str, float],
-    impact_factor: float = 60.0,
-) -> dict[str, float]:
-    """ReRanker that uses Reciprocal Rank Fusion."""
-    # Convert scores to ranks
-    vector_ranks = {
-        doc_id: i + 1 for i, (doc_id, _) in enumerate(sorted(vector_scores.items(), key=lambda x: x[1], reverse=True))
-    }
-    keyword_ranks = {
-        doc_id: i + 1 for i, (doc_id, _) in enumerate(sorted(keyword_scores.items(), key=lambda x: x[1], reverse=True))
-    }
-
-    all_ids = set(vector_scores.keys()) | set(keyword_scores.keys())
-    rrf_scores = {}
-    for doc_id in all_ids:
-        vector_rank = vector_ranks.get(doc_id, float("inf"))
-        keyword_rank = keyword_ranks.get(doc_id, float("inf"))
-        # RRF formula: score = 1/(k + r) where k is impact_factor and r is the rank
-        rrf_scores[doc_id] = (1.0 / (impact_factor + vector_rank)) + (1.0 / (impact_factor + keyword_rank))
-    return rrf_scores
-
-
 def _make_sql_identifier(name: str) -> str:
     return re.sub(r"[^a-zA-Z0-9_]", "_", name)
 
@@ -398,14 +345,10 @@ class SQLiteVecIndex(EmbeddingIndex):
             for chunk, score in zip(keyword_response.chunks, keyword_response.scores, strict=False)
         }
 
-        # Combine scores using the specified reranker
+        # Combine scores using the reranking utility
-        if reranker_type == RERANKER_TYPE_WEIGHTED:
+        combined_scores = WeightedInMemoryAggregator.combine_search_results(
-            alpha = reranker_params.get("alpha", 0.5)
+            vector_scores, keyword_scores, reranker_type, reranker_params
-            combined_scores = _weighted_rerank(vector_scores, keyword_scores, alpha)
+        )
-        else:
-            # Default to RRF for None, RRF, or any unknown types
-            impact_factor = reranker_params.get("impact_factor", 60.0)
-            combined_scores = _rrf_rerank(vector_scores, keyword_scores, impact_factor)
 
         # Sort by combined score and get top k results
         sorted_items = sorted(combined_scores.items(), key=lambda x: x[1], reverse=True)

llama_stack/providers/registry/batches.py

@@ -13,7 +13,7 @@ def available_providers() -> list[ProviderSpec]:
         InlineProviderSpec(
             api=Api.batches,
             provider_type="inline::reference",
-            pip_packages=["openai"],
+            pip_packages=[],
             module="llama_stack.providers.inline.batches.reference",
             config_class="llama_stack.providers.inline.batches.reference.config.ReferenceBatchesImplConfig",
             api_dependencies=[

llama_stack/providers/registry/datasetio.py

@@ -30,7 +30,7 @@ def available_providers() -> list[ProviderSpec]:
             adapter=AdapterSpec(
                 adapter_type="huggingface",
                 pip_packages=[
-                    "datasets",
+                    "datasets>=4.0.0",
                 ],
                 module="llama_stack.providers.remote.datasetio.huggingface",
                 config_class="llama_stack.providers.remote.datasetio.huggingface.HuggingfaceDatasetIOConfig",
@@ -42,7 +42,7 @@ def available_providers() -> list[ProviderSpec]:
             adapter=AdapterSpec(
                 adapter_type="nvidia",
                 pip_packages=[
-                    "datasets",
+                    "datasets>=4.0.0",
                 ],
                 module="llama_stack.providers.remote.datasetio.nvidia",
                 config_class="llama_stack.providers.remote.datasetio.nvidia.NvidiaDatasetIOConfig",

llama_stack/providers/registry/inference.py

@@ -75,7 +75,7 @@ def available_providers() -> list[ProviderSpec]:
             api=Api.inference,
             adapter=AdapterSpec(
                 adapter_type="vllm",
-                pip_packages=["openai"],
+                pip_packages=[],
                 module="llama_stack.providers.remote.inference.vllm",
                 config_class="llama_stack.providers.remote.inference.vllm.VLLMInferenceAdapterConfig",
                 description="Remote vLLM inference provider for connecting to vLLM servers.",
@@ -116,7 +116,7 @@ def available_providers() -> list[ProviderSpec]:
             adapter=AdapterSpec(
                 adapter_type="fireworks",
                 pip_packages=[
-                    "fireworks-ai<=0.18.0",
+                    "fireworks-ai<=0.17.16",
                 ],
                 module="llama_stack.providers.remote.inference.fireworks",
                 config_class="llama_stack.providers.remote.inference.fireworks.FireworksImplConfig",
@@ -151,9 +151,7 @@ def available_providers() -> list[ProviderSpec]:
             api=Api.inference,
             adapter=AdapterSpec(
                 adapter_type="databricks",
-                pip_packages=[
+                pip_packages=[],
-                    "openai",
-                ],
                 module="llama_stack.providers.remote.inference.databricks",
                 config_class="llama_stack.providers.remote.inference.databricks.DatabricksImplConfig",
                 description="Databricks inference provider for running models on Databricks' unified analytics platform.",
@@ -163,9 +161,7 @@ def available_providers() -> list[ProviderSpec]:
             api=Api.inference,
             adapter=AdapterSpec(
                 adapter_type="nvidia",
-                pip_packages=[
+                pip_packages=[],
-                    "openai",
-                ],
                 module="llama_stack.providers.remote.inference.nvidia",
                 config_class="llama_stack.providers.remote.inference.nvidia.NVIDIAConfig",
                 description="NVIDIA inference provider for accessing NVIDIA NIM models and AI services.",
@@ -175,7 +171,7 @@ def available_providers() -> list[ProviderSpec]:
             api=Api.inference,
             adapter=AdapterSpec(
                 adapter_type="runpod",
-                pip_packages=["openai"],
+                pip_packages=[],
                 module="llama_stack.providers.remote.inference.runpod",
                 config_class="llama_stack.providers.remote.inference.runpod.RunpodImplConfig",
                 description="RunPod inference provider for running models on RunPod's cloud GPU platform.",
@@ -292,7 +288,7 @@ Available Models:
             api=Api.inference,
             adapter=AdapterSpec(
                 adapter_type="watsonx",
-                pip_packages=["ibm_watson_machine_learning"],
+                pip_packages=["ibm_watsonx_ai"],
                 module="llama_stack.providers.remote.inference.watsonx",
                 config_class="llama_stack.providers.remote.inference.watsonx.WatsonXConfig",
                 provider_data_validator="llama_stack.providers.remote.inference.watsonx.WatsonXProviderDataValidator",

llama_stack/providers/registry/post_training.py

@@ -48,7 +48,7 @@ def available_providers() -> list[ProviderSpec]:
         InlineProviderSpec(
             api=Api.post_training,
             provider_type="inline::huggingface-gpu",
-            pip_packages=["trl", "transformers", "peft", "datasets", "torch"],
+            pip_packages=["trl", "transformers", "peft", "datasets>=4.0.0", "torch"],
             module="llama_stack.providers.inline.post_training.huggingface",
             config_class="llama_stack.providers.inline.post_training.huggingface.HuggingFacePostTrainingConfig",
             api_dependencies=[

llama_stack/providers/registry/scoring.py

@@ -38,7 +38,7 @@ def available_providers() -> list[ProviderSpec]:
         InlineProviderSpec(
             api=Api.scoring,
             provider_type="inline::braintrust",
-            pip_packages=["autoevals", "openai"],
+            pip_packages=["autoevals"],
             module="llama_stack.providers.inline.scoring.braintrust",
             config_class="llama_stack.providers.inline.scoring.braintrust.BraintrustScoringConfig",
             api_dependencies=[

llama_stack/providers/registry/tool_runtime.py

@@ -32,7 +32,7 @@ def available_providers() -> list[ProviderSpec]:
             ],
             module="llama_stack.providers.inline.tool_runtime.rag",
             config_class="llama_stack.providers.inline.tool_runtime.rag.config.RagToolRuntimeConfig",
-            api_dependencies=[Api.vector_io, Api.inference],
+            api_dependencies=[Api.vector_io, Api.inference, Api.files],
             description="RAG (Retrieval-Augmented Generation) tool runtime for document ingestion, chunking, and semantic search.",
         ),
         remote_provider_spec(

llama_stack/providers/remote/inference/anthropic/anthropic.py

@@ -5,12 +5,13 @@
 # the root directory of this source tree.
 
 from llama_stack.providers.utils.inference.litellm_openai_mixin import LiteLLMOpenAIMixin
+from llama_stack.providers.utils.inference.openai_mixin import OpenAIMixin
 
 from .config import AnthropicConfig
 from .models import MODEL_ENTRIES
 
 
-class AnthropicInferenceAdapter(LiteLLMOpenAIMixin):
+class AnthropicInferenceAdapter(OpenAIMixin, LiteLLMOpenAIMixin):
     def __init__(self, config: AnthropicConfig) -> None:
         LiteLLMOpenAIMixin.__init__(
             self,
@@ -26,3 +27,8 @@ class AnthropicInferenceAdapter(LiteLLMOpenAIMixin):
 
     async def shutdown(self) -> None:
         await super().shutdown()
+
+    get_api_key = LiteLLMOpenAIMixin.get_api_key
+
+    def get_base_url(self):
+        return "https://api.anthropic.com/v1"

llama_stack/providers/remote/inference/gemini/gemini.py

@@ -5,12 +5,13 @@
 # the root directory of this source tree.
 
 from llama_stack.providers.utils.inference.litellm_openai_mixin import LiteLLMOpenAIMixin
+from llama_stack.providers.utils.inference.openai_mixin import OpenAIMixin
 
 from .config import GeminiConfig
 from .models import MODEL_ENTRIES
 
 
-class GeminiInferenceAdapter(LiteLLMOpenAIMixin):
+class GeminiInferenceAdapter(OpenAIMixin, LiteLLMOpenAIMixin):
     def __init__(self, config: GeminiConfig) -> None:
         LiteLLMOpenAIMixin.__init__(
             self,
@@ -21,6 +22,11 @@ class GeminiInferenceAdapter(LiteLLMOpenAIMixin):
         )
         self.config = config
 
+    get_api_key = LiteLLMOpenAIMixin.get_api_key
+
+    def get_base_url(self):
+        return "https://generativelanguage.googleapis.com/v1beta/openai/"
+
     async def initialize(self) -> None:
         await super().initialize()
 

llama_stack/providers/remote/inference/groq/groq.py

@@ -4,30 +4,15 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
-from collections.abc import AsyncIterator
-from typing import Any
 
-from openai import AsyncOpenAI
-
-from llama_stack.apis.inference import (
-    OpenAIChatCompletion,
-    OpenAIChatCompletionChunk,
-    OpenAIChoiceDelta,
-    OpenAIChunkChoice,
-    OpenAIMessageParam,
-    OpenAIResponseFormatParam,
-    OpenAISystemMessageParam,
-)
 from llama_stack.providers.remote.inference.groq.config import GroqConfig
 from llama_stack.providers.utils.inference.litellm_openai_mixin import LiteLLMOpenAIMixin
-from llama_stack.providers.utils.inference.openai_compat import (
+from llama_stack.providers.utils.inference.openai_mixin import OpenAIMixin
-    prepare_openai_completion_params,
-)
 
 from .models import MODEL_ENTRIES
 
 
-class GroqInferenceAdapter(LiteLLMOpenAIMixin):
+class GroqInferenceAdapter(OpenAIMixin, LiteLLMOpenAIMixin):
     _config: GroqConfig
 
     def __init__(self, config: GroqConfig):
@@ -40,122 +25,14 @@ class GroqInferenceAdapter(LiteLLMOpenAIMixin):
         )
         self.config = config
 
+    # Delegate the client data handling get_api_key method to LiteLLMOpenAIMixin
+    get_api_key = LiteLLMOpenAIMixin.get_api_key
+
+    def get_base_url(self) -> str:
+        return f"{self.config.url}/openai/v1"
+
     async def initialize(self):
         await super().initialize()
 
     async def shutdown(self):
         await super().shutdown()
-
-    def _get_openai_client(self) -> AsyncOpenAI:
-        return AsyncOpenAI(
-            base_url=f"{self.config.url}/openai/v1",
-            api_key=self.get_api_key(),
-        )
-
-    async def openai_chat_completion(
-        self,
-        model: str,
-        messages: list[OpenAIMessageParam],
-        frequency_penalty: float | None = None,
-        function_call: str | dict[str, Any] | None = None,
-        functions: list[dict[str, Any]] | None = None,
-        logit_bias: dict[str, float] | None = None,
-        logprobs: bool | None = None,
-        max_completion_tokens: int | None = None,
-        max_tokens: int | None = None,
-        n: int | None = None,
-        parallel_tool_calls: bool | None = None,
-        presence_penalty: float | None = None,
-        response_format: OpenAIResponseFormatParam | None = None,
-        seed: int | None = None,
-        stop: str | list[str] | None = None,
-        stream: bool | None = None,
-        stream_options: dict[str, Any] | None = None,
-        temperature: float | None = None,
-        tool_choice: str | dict[str, Any] | None = None,
-        tools: list[dict[str, Any]] | None = None,
-        top_logprobs: int | None = None,
-        top_p: float | None = None,
-        user: str | None = None,
-    ) -> OpenAIChatCompletion | AsyncIterator[OpenAIChatCompletionChunk]:
-        model_obj = await self.model_store.get_model(model)
-
-        # Groq does not support json_schema response format, so we need to convert it to json_object
-        if response_format and response_format.type == "json_schema":
-            response_format.type = "json_object"
-            schema = response_format.json_schema.get("schema", {})
-            response_format.json_schema = None
-            json_instructions = f"\nYour response should be a JSON object that matches the following schema: {schema}"
-            if messages and messages[0].role == "system":
-                messages[0].content = messages[0].content + json_instructions
-            else:
-                messages.insert(0, OpenAISystemMessageParam(content=json_instructions))
-
-        # Groq returns a 400 error if tools are provided but none are called
-        # So, set tool_choice to "required" to attempt to force a call
-        if tools and (not tool_choice or tool_choice == "auto"):
-            tool_choice = "required"
-
-        params = await prepare_openai_completion_params(
-            model=model_obj.provider_resource_id,
-            messages=messages,
-            frequency_penalty=frequency_penalty,
-            function_call=function_call,
-            functions=functions,
-            logit_bias=logit_bias,
-            logprobs=logprobs,
-            max_completion_tokens=max_completion_tokens,
-            max_tokens=max_tokens,
-            n=n,
-            parallel_tool_calls=parallel_tool_calls,
-            presence_penalty=presence_penalty,
-            response_format=response_format,
-            seed=seed,
-            stop=stop,
-            stream=stream,
-            stream_options=stream_options,
-            temperature=temperature,
-            tool_choice=tool_choice,
-            tools=tools,
-            top_logprobs=top_logprobs,
-            top_p=top_p,
-            user=user,
-        )
-
-        # Groq does not support streaming requests that set response_format
-        fake_stream = False
-        if stream and response_format:
-            params["stream"] = False
-            fake_stream = True
-
-        response = await self._get_openai_client().chat.completions.create(**params)
-
-        if fake_stream:
-            chunk_choices = []
-            for choice in response.choices:
-                delta = OpenAIChoiceDelta(
-                    content=choice.message.content,
-                    role=choice.message.role,
-                    tool_calls=choice.message.tool_calls,
-                )
-                chunk_choice = OpenAIChunkChoice(
-                    delta=delta,
-                    finish_reason=choice.finish_reason,
-                    index=choice.index,
-                    logprobs=None,
-                )
-                chunk_choices.append(chunk_choice)
-            chunk = OpenAIChatCompletionChunk(
-                id=response.id,
-                choices=chunk_choices,
-                object="chat.completion.chunk",
-                created=response.created,
-                model=response.model,
-            )
-
-            async def _fake_stream_generator():
-                yield chunk
-
-            return _fake_stream_generator()
-        else:
-            return response

llama_stack/providers/remote/inference/ollama/ollama.py

@@ -118,10 +118,10 @@ class OllamaInferenceAdapter(
 
     async def initialize(self) -> None:
         logger.info(f"checking connectivity to Ollama at `{self.config.url}`...")
-        health_response = await self.health()
+        r = await self.health()
-        if health_response["status"] == HealthStatus.ERROR:
+        if r["status"] == HealthStatus.ERROR:
             logger.warning(
-                "Ollama Server is not running, make sure to start it using `ollama serve` in a separate terminal"
+                f"Ollama Server is not running (message: {r['message']}). Make sure to start it using `ollama serve` in a separate terminal"
             )
 
     async def should_refresh_models(self) -> bool:
@@ -156,7 +156,7 @@ class OllamaInferenceAdapter(
             ),
             Model(
                 identifier="nomic-embed-text",
-                provider_resource_id="nomic-embed-text",
+                provider_resource_id="nomic-embed-text:latest",
                 provider_id=provider_id,
                 metadata={
                     "embedding_dimension": 768,

llama_stack/providers/remote/inference/sambanova/sambanova.py

@@ -4,13 +4,26 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
+
 from llama_stack.providers.utils.inference.litellm_openai_mixin import LiteLLMOpenAIMixin
+from llama_stack.providers.utils.inference.openai_mixin import OpenAIMixin
 
 from .config import SambaNovaImplConfig
 from .models import MODEL_ENTRIES
 
 
-class SambaNovaInferenceAdapter(LiteLLMOpenAIMixin):
+class SambaNovaInferenceAdapter(OpenAIMixin, LiteLLMOpenAIMixin):
+    """
+    SambaNova Inference Adapter for Llama Stack.
+
+    Note: The inheritance order is important here. OpenAIMixin must come before
+    LiteLLMOpenAIMixin to ensure that OpenAIMixin.check_model_availability()
+    is used instead of LiteLLMOpenAIMixin.check_model_availability().
+
+    - OpenAIMixin.check_model_availability() queries the /v1/models to check if a model exists
+    - LiteLLMOpenAIMixin.check_model_availability() checks the static registry within LiteLLM
+    """
+
     def __init__(self, config: SambaNovaImplConfig):
         self.config = config
         self.environment_available_models = []
@@ -24,3 +37,14 @@ class SambaNovaInferenceAdapter(LiteLLMOpenAIMixin):
             download_images=True,  # SambaNova requires base64 image encoding
             json_schema_strict=False,  # SambaNova doesn't support strict=True yet
         )
+
+    # Delegate the client data handling get_api_key method to LiteLLMOpenAIMixin
+    get_api_key = LiteLLMOpenAIMixin.get_api_key
+
+    def get_base_url(self) -> str:
+        """
+        Get the base URL for OpenAI mixin.
+
+        :return: The SambaNova base URL
+        """
+        return self.config.url

llama_stack/providers/remote/inference/vertexai/vertexai.py

@@ -6,16 +6,20 @@
 
 from typing import Any
 
+import google.auth.transport.requests
+from google.auth import default
+
 from llama_stack.apis.inference import ChatCompletionRequest
 from llama_stack.providers.utils.inference.litellm_openai_mixin import (
     LiteLLMOpenAIMixin,
 )
+from llama_stack.providers.utils.inference.openai_mixin import OpenAIMixin
 
 from .config import VertexAIConfig
 from .models import MODEL_ENTRIES
 
 
-class VertexAIInferenceAdapter(LiteLLMOpenAIMixin):
+class VertexAIInferenceAdapter(OpenAIMixin, LiteLLMOpenAIMixin):
     def __init__(self, config: VertexAIConfig) -> None:
         LiteLLMOpenAIMixin.__init__(
             self,
@@ -27,10 +31,31 @@ class VertexAIInferenceAdapter(LiteLLMOpenAIMixin):
         self.config = config
 
     def get_api_key(self) -> str:
-        # Vertex AI doesn't use API keys, it uses Application Default Credentials
+        """
-        # Return empty string to let litellm handle authentication via ADC
+        Get an access token for Vertex AI using Application Default Credentials.
+
+        Vertex AI uses ADC instead of API keys. This method obtains an access token
+        from the default credentials and returns it for use with the OpenAI-compatible client.
+        """
+        try:
+            # Get default credentials - will read from GOOGLE_APPLICATION_CREDENTIALS
+            credentials, _ = default(scopes=["https://www.googleapis.com/auth/cloud-platform"])
+            credentials.refresh(google.auth.transport.requests.Request())
+            return str(credentials.token)
+        except Exception:
+            # If we can't get credentials, return empty string to let LiteLLM handle it
+            # This allows the LiteLLM mixin to work with ADC directly
             return ""
 
+    def get_base_url(self) -> str:
+        """
+        Get the Vertex AI OpenAI-compatible API base URL.
+
+        Returns the Vertex AI OpenAI-compatible endpoint URL.
+        Source: https://cloud.google.com/vertex-ai/generative-ai/docs/start/openai
+        """
+        return f"https://{self.config.location}-aiplatform.googleapis.com/v1/projects/{self.config.project}/locations/{self.config.location}/endpoints/openapi"
+
     async def _get_params(self, request: ChatCompletionRequest) -> dict[str, Any]:
         # Get base parameters from parent
         params = await super()._get_params(request)

llama_stack/providers/remote/inference/watsonx/watsonx.py

@@ -7,8 +7,8 @@
 from collections.abc import AsyncGenerator, AsyncIterator
 from typing import Any
 
-from ibm_watson_machine_learning.foundation_models import Model
+from ibm_watsonx_ai.foundation_models import Model
-from ibm_watson_machine_learning.metanames import GenTextParamsMetaNames as GenParams
+from ibm_watsonx_ai.metanames import GenTextParamsMetaNames as GenParams
 from openai import AsyncOpenAI
 
 from llama_stack.apis.common.content_types import InterleavedContent, InterleavedContentItem

llama_stack/providers/utils/bedrock/config.py

@@ -4,53 +4,55 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
+import os
+
 from pydantic import BaseModel, Field
 
 
 class BedrockBaseConfig(BaseModel):
     aws_access_key_id: str | None = Field(
-        default=None,
+        default_factory=lambda: os.getenv("AWS_ACCESS_KEY_ID"),
         description="The AWS access key to use. Default use environment variable: AWS_ACCESS_KEY_ID",
     )
     aws_secret_access_key: str | None = Field(
-        default=None,
+        default_factory=lambda: os.getenv("AWS_SECRET_ACCESS_KEY"),
         description="The AWS secret access key to use. Default use environment variable: AWS_SECRET_ACCESS_KEY",
     )
     aws_session_token: str | None = Field(
-        default=None,
+        default_factory=lambda: os.getenv("AWS_SESSION_TOKEN"),
         description="The AWS session token to use. Default use environment variable: AWS_SESSION_TOKEN",
     )
     region_name: str | None = Field(
-        default=None,
+        default_factory=lambda: os.getenv("AWS_DEFAULT_REGION"),
         description="The default AWS Region to use, for example, us-west-1 or us-west-2."
         "Default use environment variable: AWS_DEFAULT_REGION",
     )
     profile_name: str | None = Field(
-        default=None,
+        default_factory=lambda: os.getenv("AWS_PROFILE"),
         description="The profile name that contains credentials to use.Default use environment variable: AWS_PROFILE",
     )
     total_max_attempts: int | None = Field(
-        default=None,
+        default_factory=lambda: int(val) if (val := os.getenv("AWS_MAX_ATTEMPTS")) else None,
         description="An integer representing the maximum number of attempts that will be made for a single request, "
         "including the initial attempt. Default use environment variable: AWS_MAX_ATTEMPTS",
     )
     retry_mode: str | None = Field(
-        default=None,
+        default_factory=lambda: os.getenv("AWS_RETRY_MODE"),
         description="A string representing the type of retries Boto3 will perform."
         "Default use environment variable: AWS_RETRY_MODE",
     )
     connect_timeout: float | None = Field(
-        default=60,
+        default_factory=lambda: float(os.getenv("AWS_CONNECT_TIMEOUT", "60")),
         description="The time in seconds till a timeout exception is thrown when attempting to make a connection. "
         "The default is 60 seconds.",
     )
     read_timeout: float | None = Field(
-        default=60,
+        default_factory=lambda: float(os.getenv("AWS_READ_TIMEOUT", "60")),
         description="The time in seconds till a timeout exception is thrown when attempting to read from a connection."
         "The default is 60 seconds.",
     )
     session_ttl: int | None = Field(
-        default=3600,
+        default_factory=lambda: int(os.getenv("AWS_SESSION_TTL", "3600")),
         description="The time in seconds till a session expires. The default is 3600 seconds (1 hour).",
     )
 

llama_stack/providers/utils/inference/embedding_mixin.py

@@ -4,6 +4,7 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
+import asyncio
 import base64
 import struct
 from typing import TYPE_CHECKING
@@ -43,9 +44,11 @@ class SentenceTransformerEmbeddingMixin:
         task_type: EmbeddingTaskType | None = None,
     ) -> EmbeddingsResponse:
         model = await self.model_store.get_model(model_id)
-        embedding_model = self._load_sentence_transformer_model(model.provider_resource_id)
+        embedding_model = await self._load_sentence_transformer_model(model.provider_resource_id)
-        embeddings = embedding_model.encode(
+        embeddings = await asyncio.to_thread(
-            [interleaved_content_as_str(content) for content in contents], show_progress_bar=False
+            embedding_model.encode,
+            [interleaved_content_as_str(content) for content in contents],
+            show_progress_bar=False,
         )
         return EmbeddingsResponse(embeddings=embeddings)
 
@@ -64,8 +67,8 @@ class SentenceTransformerEmbeddingMixin:
 
         # Get the model and generate embeddings
         model_obj = await self.model_store.get_model(model)
-        embedding_model = self._load_sentence_transformer_model(model_obj.provider_resource_id)
+        embedding_model = await self._load_sentence_transformer_model(model_obj.provider_resource_id)
-        embeddings = embedding_model.encode(input_list, show_progress_bar=False)
+        embeddings = await asyncio.to_thread(embedding_model.encode, input_list, show_progress_bar=False)
 
         # Convert embeddings to the requested format
         data = []
@@ -93,7 +96,7 @@ class SentenceTransformerEmbeddingMixin:
             usage=usage,
         )
 
-    def _load_sentence_transformer_model(self, model: str) -> "SentenceTransformer":
+    async def _load_sentence_transformer_model(self, model: str) -> "SentenceTransformer":
         global EMBEDDING_MODELS
 
         loaded_model = EMBEDDING_MODELS.get(model)
@@ -101,8 +104,12 @@ class SentenceTransformerEmbeddingMixin:
             return loaded_model
 
         log.info(f"Loading sentence transformer for {model}...")
+
+        def _load_model():
             from sentence_transformers import SentenceTransformer
 
-        loaded_model = SentenceTransformer(model)
+            return SentenceTransformer(model)
+
+        loaded_model = await asyncio.to_thread(_load_model)
         EMBEDDING_MODELS[model] = loaded_model
         return loaded_model

llama_stack/providers/utils/inference/inference_store.py

@@ -3,6 +3,11 @@
 #
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
+import asyncio
+from typing import Any
+
+from sqlalchemy.exc import IntegrityError
+
 from llama_stack.apis.inference import (
     ListOpenAIChatCompletionResponse,
     OpenAIChatCompletion,
@@ -10,24 +15,43 @@ from llama_stack.apis.inference import (
     OpenAIMessageParam,
     Order,
 )
-from llama_stack.core.datatypes import AccessRule
+from llama_stack.core.datatypes import AccessRule, InferenceStoreConfig
-from llama_stack.core.utils.config_dirs import RUNTIME_BASE_DIR
+from llama_stack.log import get_logger
 
 from ..sqlstore.api import ColumnDefinition, ColumnType
 from ..sqlstore.authorized_sqlstore import AuthorizedSqlStore
-from ..sqlstore.sqlstore import SqliteSqlStoreConfig, SqlStoreConfig, sqlstore_impl
+from ..sqlstore.sqlstore import SqlStoreConfig, SqlStoreType, sqlstore_impl
+
+logger = get_logger(name=__name__, category="inference_store")
 
 
 class InferenceStore:
-    def __init__(self, sql_store_config: SqlStoreConfig, policy: list[AccessRule]):
+    def __init__(
-        if not sql_store_config:
+        self,
-            sql_store_config = SqliteSqlStoreConfig(
+        config: InferenceStoreConfig | SqlStoreConfig,
-                db_path=(RUNTIME_BASE_DIR / "sqlstore.db").as_posix(),
+        policy: list[AccessRule],
+    ):
+        # Handle backward compatibility
+        if not isinstance(config, InferenceStoreConfig):
+            # Legacy: SqlStoreConfig passed directly as config
+            config = InferenceStoreConfig(
+                sql_store_config=config,
             )
-        self.sql_store_config = sql_store_config
+
+        self.config = config
+        self.sql_store_config = config.sql_store_config
         self.sql_store = None
         self.policy = policy
 
+        # Disable write queue for SQLite to avoid concurrency issues
+        self.enable_write_queue = self.sql_store_config.type != SqlStoreType.sqlite
+
+        # Async write queue and worker control
+        self._queue: asyncio.Queue[tuple[OpenAIChatCompletion, list[OpenAIMessageParam]]] | None = None
+        self._worker_tasks: list[asyncio.Task[Any]] = []
+        self._max_write_queue_size: int = config.max_write_queue_size
+        self._num_writers: int = max(1, config.num_writers)
+
     async def initialize(self):
         """Create the necessary tables if they don't exist."""
         self.sql_store = AuthorizedSqlStore(sqlstore_impl(self.sql_store_config))
@@ -42,23 +66,109 @@ class InferenceStore:
             },
         )
 
+        if self.enable_write_queue:
+            self._queue = asyncio.Queue(maxsize=self._max_write_queue_size)
+            for _ in range(self._num_writers):
+                self._worker_tasks.append(asyncio.create_task(self._worker_loop()))
+        else:
+            logger.info("Write queue disabled for SQLite to avoid concurrency issues")
+
+    async def shutdown(self) -> None:
+        if not self._worker_tasks:
+            return
+        if self._queue is not None:
+            await self._queue.join()
+        for t in self._worker_tasks:
+            if not t.done():
+                t.cancel()
+        for t in self._worker_tasks:
+            try:
+                await t
+            except asyncio.CancelledError:
+                pass
+        self._worker_tasks.clear()
+
+    async def flush(self) -> None:
+        """Wait for all queued writes to complete. Useful for testing."""
+        if self.enable_write_queue and self._queue is not None:
+            await self._queue.join()
+
     async def store_chat_completion(
         self, chat_completion: OpenAIChatCompletion, input_messages: list[OpenAIMessageParam]
     ) -> None:
-        if not self.sql_store:
+        if self.enable_write_queue:
+            if self._queue is None:
+                raise ValueError("Inference store is not initialized")
+            try:
+                self._queue.put_nowait((chat_completion, input_messages))
+            except asyncio.QueueFull:
+                logger.warning(
+                    f"Write queue full; adding chat completion id={getattr(chat_completion, 'id', '<unknown>')}"
+                )
+                await self._queue.put((chat_completion, input_messages))
+        else:
+            await self._write_chat_completion(chat_completion, input_messages)
+
+    async def _worker_loop(self) -> None:
+        assert self._queue is not None
+        while True:
+            try:
+                item = await self._queue.get()
+            except asyncio.CancelledError:
+                break
+            chat_completion, input_messages = item
+            try:
+                await self._write_chat_completion(chat_completion, input_messages)
+            except Exception as e:  # noqa: BLE001
+                logger.error(f"Error writing chat completion: {e}")
+            finally:
+                self._queue.task_done()
+
+    async def _write_chat_completion(
+        self, chat_completion: OpenAIChatCompletion, input_messages: list[OpenAIMessageParam]
+    ) -> None:
+        if self.sql_store is None:
             raise ValueError("Inference store is not initialized")
 
         data = chat_completion.model_dump()
-
+        record_data = {
-        await self.sql_store.insert(
-            table="chat_completions",
-            data={
             "id": data["id"],
             "created": data["created"],
             "model": data["model"],
             "choices": data["choices"],
             "input_messages": [message.model_dump() for message in input_messages],
-            },
+        }
+
+        try:
+            await self.sql_store.insert(
+                table="chat_completions",
+                data=record_data,
+            )
+        except IntegrityError as e:
+            # Duplicate chat completion IDs can be generated during tests especially if they are replaying
+            # recorded responses across different tests. No need to warn or error under those circumstances.
+            # In the wild, this is not likely to happen at all (no evidence) so we aren't really hiding any problem.
+
+            # Check if it's a unique constraint violation
+            error_message = str(e.orig) if e.orig else str(e)
+            if self._is_unique_constraint_error(error_message):
+                # Update the existing record instead
+                await self.sql_store.update(table="chat_completions", data=record_data, where={"id": data["id"]})
+            else:
+                # Re-raise if it's not a unique constraint error
+                raise
+
+    def _is_unique_constraint_error(self, error_message: str) -> bool:
+        """Check if the error is specifically a unique constraint violation."""
+        error_lower = error_message.lower()
+        return any(
+            indicator in error_lower
+            for indicator in [
+                "unique constraint failed",  # SQLite
+                "duplicate key",  # PostgreSQL
+                "unique violation",  # PostgreSQL alternative
+                "duplicate entry",  # MySQL
+            ]
         )
 
     async def list_chat_completions(

llama_stack/providers/utils/sqlstore/authorized_sqlstore.py

@@ -172,6 +172,20 @@ class AuthorizedSqlStore:
 
         return results.data[0] if results.data else None
 
+    async def update(self, table: str, data: Mapping[str, Any], where: Mapping[str, Any]) -> None:
+        """Update rows with automatic access control attribute capture."""
+        enhanced_data = dict(data)
+
+        current_user = get_authenticated_user()
+        if current_user:
+            enhanced_data["owner_principal"] = current_user.principal
+            enhanced_data["access_attributes"] = current_user.attributes
+        else:
+            enhanced_data["owner_principal"] = None
+            enhanced_data["access_attributes"] = None
+
+        await self.sql_store.update(table, enhanced_data, where)
+
     async def delete(self, table: str, where: Mapping[str, Any]) -> None:
         """Delete rows with automatic access control filtering."""
         await self.sql_store.delete(table, where)

llama_stack/providers/utils/telemetry/tracing.py

@@ -18,6 +18,7 @@ from functools import wraps
 from typing import Any
 
 from llama_stack.apis.telemetry import (
+    Event,
     LogSeverity,
     Span,
     SpanEndPayload,
@@ -98,7 +99,7 @@ class BackgroundLogger:
     def __init__(self, api: Telemetry, capacity: int = 100000):
         self.api = api
         self.log_queue: queue.Queue[Any] = queue.Queue(maxsize=capacity)
-        self.worker_thread = threading.Thread(target=self._process_logs, daemon=True)
+        self.worker_thread = threading.Thread(target=self._worker, daemon=True)
         self.worker_thread.start()
         self._last_queue_full_log_time: float = 0.0
         self._dropped_since_last_notice: int = 0
@@ -118,12 +119,16 @@ class BackgroundLogger:
                 self._last_queue_full_log_time = current_time
                 self._dropped_since_last_notice = 0
 
-    def _process_logs(self):
+    def _worker(self):
+        loop = asyncio.new_event_loop()
+        asyncio.set_event_loop(loop)
+        loop.run_until_complete(self._process_logs())
+
+    async def _process_logs(self):
         while True:
             try:
                 event = self.log_queue.get()
-                # figure out how to use a thread's native loop
+                await self.api.log_event(event)
-                asyncio.run(self.api.log_event(event))
             except Exception:
                 import traceback
 
@@ -136,6 +141,19 @@ class BackgroundLogger:
         self.log_queue.join()
 
 
+def enqueue_event(event: Event) -> None:
+    """Enqueue a telemetry event to the background logger if available.
+
+    This provides a non-blocking path for routers and other hot paths to
+    submit telemetry without awaiting the Telemetry API, reducing contention
+    with the main event loop.
+    """
+    global BACKGROUND_LOGGER
+    if BACKGROUND_LOGGER is None:
+        raise RuntimeError("Telemetry API not initialized")
+    BACKGROUND_LOGGER.log_event(event)
+
+
 class TraceContext:
     spans: list[Span] = []
 
@@ -256,11 +274,7 @@ class TelemetryHandler(logging.Handler):
         if record.module in ("asyncio", "selector_events"):
             return
 
-        global CURRENT_TRACE_CONTEXT, BACKGROUND_LOGGER
+        global CURRENT_TRACE_CONTEXT
-
-        if BACKGROUND_LOGGER is None:
-            raise RuntimeError("Telemetry API not initialized")
-
         context = CURRENT_TRACE_CONTEXT.get()
         if context is None:
             return
@@ -269,7 +283,7 @@ class TelemetryHandler(logging.Handler):
         if span is None:
             return
 
-        BACKGROUND_LOGGER.log_event(
+        enqueue_event(
             UnstructuredLogEvent(
                 trace_id=span.trace_id,
                 span_id=span.span_id,

llama_stack/providers/utils/tools/mcp.py

@@ -67,6 +67,38 @@ async def client_wrapper(endpoint: str, headers: dict[str, str]) -> AsyncGenerat
                     raise AuthenticationRequiredError(exc) from exc
             if i == len(connection_strategies) - 1:
                 raise
+        except* httpx.ConnectError as eg:
+            # Connection refused, server down, network unreachable
+            if i == len(connection_strategies) - 1:
+                error_msg = f"Failed to connect to MCP server at {endpoint}: Connection refused"
+                logger.error(f"MCP connection error: {error_msg}")
+                raise ConnectionError(error_msg) from eg
+            else:
+                logger.warning(
+                    f"failed to connect to MCP server at {endpoint} via {strategy.name}, falling back to {connection_strategies[i + 1].name}"
+                )
+        except* httpx.TimeoutException as eg:
+            # Request timeout, server too slow
+            if i == len(connection_strategies) - 1:
+                error_msg = f"MCP server at {endpoint} timed out"
+                logger.error(f"MCP timeout error: {error_msg}")
+                raise TimeoutError(error_msg) from eg
+            else:
+                logger.warning(
+                    f"MCP server at {endpoint} timed out via {strategy.name}, falling back to {connection_strategies[i + 1].name}"
+                )
+        except* httpx.RequestError as eg:
+            # DNS resolution failures, network errors, invalid URLs
+            if i == len(connection_strategies) - 1:
+                # Get the first exception's message for the error string
+                exc_msg = str(eg.exceptions[0]) if eg.exceptions else "Unknown error"
+                error_msg = f"Network error connecting to MCP server at {endpoint}: {exc_msg}"
+                logger.error(f"MCP network error: {error_msg}")
+                raise ConnectionError(error_msg) from eg
+            else:
+                logger.warning(
+                    f"network error connecting to MCP server at {endpoint} via {strategy.name}, falling back to {connection_strategies[i + 1].name}"
+                )
         except* McpError:
             if i < len(connection_strategies) - 1:
                 logger.warning(

llama_stack/testing/inference_recorder.py

@@ -30,6 +30,9 @@ from openai.types.completion_choice import CompletionChoice
 CompletionChoice.model_fields["finish_reason"].annotation = Literal["stop", "length", "content_filter"] | None
 CompletionChoice.model_rebuild()
 
+REPO_ROOT = Path(__file__).parent.parent.parent
+DEFAULT_STORAGE_DIR = REPO_ROOT / "tests/integration/recordings"
+
 
 class InferenceMode(StrEnum):
     LIVE = "live"
@@ -51,7 +54,7 @@ def normalize_request(method: str, url: str, headers: dict[str, Any], body: dict
 
 
 def get_inference_mode() -> InferenceMode:
-    return InferenceMode(os.environ.get("LLAMA_STACK_TEST_INFERENCE_MODE", "live").lower())
+    return InferenceMode(os.environ.get("LLAMA_STACK_TEST_INFERENCE_MODE", "replay").lower())
 
 
 def setup_inference_recording():
@@ -60,28 +63,18 @@ def setup_inference_recording():
     to increase their reliability and reduce reliance on expensive, external services.
 
     Currently, this is only supported for OpenAI and Ollama clients. These should cover the vast majority of use cases.
-    Calls to the /models endpoint are not currently trapped. We probably need to add support for this.
 
-    Two environment variables are required:
+    Two environment variables are supported:
-    - LLAMA_STACK_TEST_INFERENCE_MODE: The mode to run in. Must be 'live', 'record', or 'replay'.
+    - LLAMA_STACK_TEST_INFERENCE_MODE: The mode to run in. Must be 'live', 'record', or 'replay'. Default is 'replay'.
-    - LLAMA_STACK_TEST_RECORDING_DIR: The directory to store the recordings in.
+    - LLAMA_STACK_TEST_RECORDING_DIR: The directory to store the recordings in. Default is 'tests/integration/recordings'.
 
-    The recordings are stored in a SQLite database and a JSON file for each request. The SQLite database is used to
+    The recordings are stored as JSON files.
-    quickly find the correct recording for a given request. The JSON files are used to store the request and response
-    bodies.
     """
     mode = get_inference_mode()
-
-    if mode not in InferenceMode:
-        raise ValueError(f"Invalid LLAMA_STACK_TEST_INFERENCE_MODE: {mode}. Must be 'live', 'record', or 'replay'")
-
     if mode == InferenceMode.LIVE:
         return None
 
-    if "LLAMA_STACK_TEST_RECORDING_DIR" not in os.environ:
+    storage_dir = os.environ.get("LLAMA_STACK_TEST_RECORDING_DIR", DEFAULT_STORAGE_DIR)
-        raise ValueError("LLAMA_STACK_TEST_RECORDING_DIR must be set for recording or replaying")
-    storage_dir = os.environ["LLAMA_STACK_TEST_RECORDING_DIR"]
-
     return inference_recording(mode=mode, storage_dir=storage_dir)
 
 
@@ -134,8 +127,8 @@ class ResponseStorage:
     def store_recording(self, request_hash: str, request: dict[str, Any], response: dict[str, Any]):
         """Store a request/response pair."""
         # Generate unique response filename
-        response_file = f"{request_hash[:12]}.json"
+        short_hash = request_hash[:12]
-        response_path = self.responses_dir / response_file
+        response_file = f"{short_hash}.json"
 
         # Serialize response body if needed
         serialized_response = dict(response)
@@ -147,6 +140,14 @@ class ResponseStorage:
                 # Handle single response
                 serialized_response["body"] = _serialize_response(serialized_response["body"])
 
+        # If this is an Ollama /api/tags recording, include models digest in filename to distinguish variants
+        endpoint = request.get("endpoint")
+        if endpoint in ("/api/tags", "/v1/models"):
+            digest = _model_identifiers_digest(endpoint, response)
+            response_file = f"models-{short_hash}-{digest}.json"
+
+        response_path = self.responses_dir / response_file
+
         # Save response to JSON file
         with open(response_path, "w") as f:
             json.dump({"request": request, "response": serialized_response}, f, indent=2)
@@ -161,6 +162,17 @@ class ResponseStorage:
         if not response_path.exists():
             return None
 
+        return _recording_from_file(response_path)
+
+    def _model_list_responses(self, short_hash: str) -> list[dict[str, Any]]:
+        results: list[dict[str, Any]] = []
+        for path in self.responses_dir.glob(f"models-{short_hash}-*.json"):
+            data = _recording_from_file(path)
+            results.append(data)
+        return results
+
+
+def _recording_from_file(response_path) -> dict[str, Any]:
     with open(response_path) as f:
         data = json.load(f)
 
@@ -176,6 +188,61 @@ class ResponseStorage:
     return cast(dict[str, Any], data)
 
 
+def _model_identifiers_digest(endpoint: str, response: dict[str, Any]) -> str:
+    def _extract_model_identifiers():
+        """Extract a stable set of identifiers for model-list endpoints.
+
+        Supported endpoints:
+        - '/api/tags' (Ollama): response body has 'models': [ { name/model/digest/id/... }, ... ]
+        - '/v1/models' (OpenAI): response body has 'data': [ { id: ... }, ... ]
+        Returns a list of unique identifiers or None if structure doesn't match.
+        """
+        body = response["body"]
+        if endpoint == "/api/tags":
+            items = body.get("models")
+            idents = [m.model for m in items]
+        else:
+            items = body.get("data")
+            idents = [m.id for m in items]
+        return sorted(set(idents))
+
+    identifiers = _extract_model_identifiers()
+    return hashlib.sha1(("|".join(identifiers)).encode("utf-8")).hexdigest()[:8]
+
+
+def _combine_model_list_responses(endpoint: str, records: list[dict[str, Any]]) -> dict[str, Any] | None:
+    """Return a single, unioned recording for supported model-list endpoints."""
+    seen: dict[str, dict[str, Any]] = {}
+    for rec in records:
+        body = rec["response"]["body"]
+        if endpoint == "/api/tags":
+            items = body.models
+        elif endpoint == "/v1/models":
+            items = body.data
+        else:
+            items = []
+
+        for m in items:
+            if endpoint == "/v1/models":
+                key = m.id
+            else:
+                key = m.model
+            seen[key] = m
+
+    ordered = [seen[k] for k in sorted(seen.keys())]
+    canonical = records[0]
+    canonical_req = canonical.get("request", {})
+    if isinstance(canonical_req, dict):
+        canonical_req["endpoint"] = endpoint
+    if endpoint == "/v1/models":
+        body = {"data": ordered, "object": "list"}
+    else:
+        from ollama import ListResponse
+
+        body = ListResponse(models=ordered)
+    return {"request": canonical_req, "response": {"body": body, "is_streaming": False}}
+
+
 async def _patched_inference_method(original_method, self, client_type, endpoint, *args, **kwargs):
     global _current_mode, _current_storage
 
@@ -195,8 +262,6 @@ async def _patched_inference_method(original_method, self, client_type, endpoint
         raise ValueError(f"Unknown client type: {client_type}")
 
     url = base_url.rstrip("/") + endpoint
-
-    # Normalize request for matching
     method = "POST"
     headers = {}
     body = kwargs
@@ -204,6 +269,11 @@ async def _patched_inference_method(original_method, self, client_type, endpoint
     request_hash = normalize_request(method, url, headers, body)
 
     if _current_mode == InferenceMode.REPLAY:
+        # Special handling for model-list endpoints: return union of all responses
+        if endpoint in ("/api/tags", "/v1/models"):
+            records = _current_storage._model_list_responses(request_hash[:12])
+            recording = _combine_model_list_responses(endpoint, records)
+        else:
             recording = _current_storage.find_recording(request_hash)
         if recording:
             response_body = recording["response"]["body"]
@@ -222,7 +292,7 @@ async def _patched_inference_method(original_method, self, client_type, endpoint
                 f"No recorded response found for request hash: {request_hash}\n"
                 f"Request: {method} {url} {body}\n"
                 f"Model: {body.get('model', 'unknown')}\n"
-                f"To record this response, run with LLAMA_STACK_INFERENCE_MODE=record"
+                f"To record this response, run with LLAMA_STACK_TEST_INFERENCE_MODE=record"
             )
 
     elif _current_mode == InferenceMode.RECORD:
@@ -274,12 +344,14 @@ def patch_inference_clients():
     from openai.resources.chat.completions import AsyncCompletions as AsyncChatCompletions
     from openai.resources.completions import AsyncCompletions
     from openai.resources.embeddings import AsyncEmbeddings
+    from openai.resources.models import AsyncModels
 
     # Store original methods for both OpenAI and Ollama clients
     _original_methods = {
         "chat_completions_create": AsyncChatCompletions.create,
         "completions_create": AsyncCompletions.create,
         "embeddings_create": AsyncEmbeddings.create,
+        "models_list": AsyncModels.list,
         "ollama_generate": OllamaAsyncClient.generate,
         "ollama_chat": OllamaAsyncClient.chat,
         "ollama_embed": OllamaAsyncClient.embed,
@@ -304,10 +376,16 @@ def patch_inference_clients():
             _original_methods["embeddings_create"], self, "openai", "/v1/embeddings", *args, **kwargs
         )
 
+    async def patched_models_list(self, *args, **kwargs):
+        return await _patched_inference_method(
+            _original_methods["models_list"], self, "openai", "/v1/models", *args, **kwargs
+        )
+
     # Apply OpenAI patches
     AsyncChatCompletions.create = patched_chat_completions_create
     AsyncCompletions.create = patched_completions_create
     AsyncEmbeddings.create = patched_embeddings_create
+    AsyncModels.list = patched_models_list
 
     # Create patched methods for Ollama client
     async def patched_ollama_generate(self, *args, **kwargs):
@@ -361,11 +439,13 @@ def unpatch_inference_clients():
     from openai.resources.chat.completions import AsyncCompletions as AsyncChatCompletions
     from openai.resources.completions import AsyncCompletions
     from openai.resources.embeddings import AsyncEmbeddings
+    from openai.resources.models import AsyncModels
 
     # Restore OpenAI client methods
     AsyncChatCompletions.create = _original_methods["chat_completions_create"]
     AsyncCompletions.create = _original_methods["completions_create"]
     AsyncEmbeddings.create = _original_methods["embeddings_create"]
+    AsyncModels.list = _original_methods["models_list"]
 
     # Restore Ollama client methods if they were patched
     OllamaAsyncClient.generate = _original_methods["ollama_generate"]
@@ -379,16 +459,10 @@ def unpatch_inference_clients():
 
 
 @contextmanager
-def inference_recording(mode: str = "live", storage_dir: str | Path | None = None) -> Generator[None, None, None]:
+def inference_recording(mode: str, storage_dir: str | Path | None = None) -> Generator[None, None, None]:
     """Context manager for inference recording/replaying."""
     global _current_mode, _current_storage
 
-    # Set defaults
-    if storage_dir is None:
-        storage_dir_path = Path.home() / ".llama" / "recordings"
-    else:
-        storage_dir_path = Path(storage_dir)
-
     # Store previous state
     prev_mode = _current_mode
     prev_storage = _current_storage
@@ -397,7 +471,9 @@ def inference_recording(mode: str = "live", storage_dir: str | Path | None = Non
         _current_mode = mode
 
         if mode in ["record", "replay"]:
-            _current_storage = ResponseStorage(storage_dir_path)
+            if storage_dir is None:
+                raise ValueError("storage_dir is required for record and replay modes")
+            _current_storage = ResponseStorage(Path(storage_dir))
             patch_inference_clients()
 
         yield

llama_stack/ui/package-lock.json

@@ -10,7 +10,7 @@
       "dependencies": {
         "@radix-ui/react-collapsible": "^1.1.12",
         "@radix-ui/react-dialog": "^1.1.13",
-        "@radix-ui/react-dropdown-menu": "^2.1.14",
+        "@radix-ui/react-dropdown-menu": "^2.1.16",
         "@radix-ui/react-select": "^2.2.5",
         "@radix-ui/react-separator": "^1.1.7",
         "@radix-ui/react-slot": "^1.2.3",
@@ -18,18 +18,18 @@
         "class-variance-authority": "^0.7.1",
         "clsx": "^2.1.1",
         "framer-motion": "^12.23.12",
-        "llama-stack-client": "^0.2.20",
+        "llama-stack-client": "^0.2.21",
-        "lucide-react": "^0.510.0",
+        "lucide-react": "^0.542.0",
         "next": "15.3.3",
         "next-auth": "^4.24.11",
         "next-themes": "^0.4.6",
         "react": "^19.0.0",
-        "react-dom": "^19.0.0",
+        "react-dom": "^19.1.1",
         "react-markdown": "^10.1.0",
         "remark-gfm": "^4.0.1",
         "remeda": "^2.30.0",
         "shiki": "^1.29.2",
-        "sonner": "^2.0.6",
+        "sonner": "^2.0.7",
         "tailwind-merge": "^3.3.1"
       },
       "devDependencies": {
@@ -2066,12 +2066,35 @@
       "license": "MIT"
     },
     "node_modules/@radix-ui/react-arrow": {
-      "version": "1.1.6",
+      "version": "1.1.7",
-      "resolved": "https://registry.npmjs.org/@radix-ui/react-arrow/-/react-arrow-1.1.6.tgz",
+      "resolved": "https://registry.npmjs.org/@radix-ui/react-arrow/-/react-arrow-1.1.7.tgz",
-      "integrity": "sha512-2JMfHJf/eVnwq+2dewT3C0acmCWD3XiVA1Da+jTDqo342UlU13WvXtqHhG+yJw5JeQmu4ue2eMy6gcEArLBlcw==",
+      "integrity": "sha512-F+M1tLhO+mlQaOWspE8Wstg+z6PwxwRd8oQ8IXceWz92kfAmalTRf0EjrouQeo7QssEPfCn05B4Ihs1K9WQ/7w==",
       "license": "MIT",
       "dependencies": {
-        "@radix-ui/react-primitive": "2.1.2"
+        "@radix-ui/react-primitive": "2.1.3"
+      },
+      "peerDependencies": {
+        "@types/react": "*",
+        "@types/react-dom": "*",
+        "react": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc",
+        "react-dom": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc"
+      },
+      "peerDependenciesMeta": {
+        "@types/react": {
+          "optional": true
+        },
+        "@types/react-dom": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/@radix-ui/react-arrow/node_modules/@radix-ui/react-primitive": {
+      "version": "2.1.3",
+      "resolved": "https://registry.npmjs.org/@radix-ui/react-primitive/-/react-primitive-2.1.3.tgz",
+      "integrity": "sha512-m9gTwRkhy2lvCPe6QJp4d3G1TYEUHn/FzJUtq9MjH46an1wJU+GdoGC5VLof8RX8Ft/DlpshApkhswDLZzHIcQ==",
+      "license": "MIT",
+      "dependencies": {
+        "@radix-ui/react-slot": "1.2.3"
       },
       "peerDependencies": {
         "@types/react": "*",
@@ -2172,15 +2195,15 @@
       }
     },
     "node_modules/@radix-ui/react-collection": {
-      "version": "1.1.6",
+      "version": "1.1.7",
-      "resolved": "https://registry.npmjs.org/@radix-ui/react-collection/-/react-collection-1.1.6.tgz",
+      "resolved": "https://registry.npmjs.org/@radix-ui/react-collection/-/react-collection-1.1.7.tgz",
-      "integrity": "sha512-PbhRFK4lIEw9ADonj48tiYWzkllz81TM7KVYyyMMw2cwHO7D5h4XKEblL8NlaRisTK3QTe6tBEhDccFUryxHBQ==",
+      "integrity": "sha512-Fh9rGN0MoI4ZFUNyfFVNU4y9LUz93u9/0K+yLgA2bwRojxM8JU1DyvvMBabnZPBgMWREAJvU2jjVzq+LrFUglw==",
       "license": "MIT",
       "dependencies": {
         "@radix-ui/react-compose-refs": "1.1.2",
         "@radix-ui/react-context": "1.1.2",
-        "@radix-ui/react-primitive": "2.1.2",
+        "@radix-ui/react-primitive": "2.1.3",
-        "@radix-ui/react-slot": "1.2.2"
+        "@radix-ui/react-slot": "1.2.3"
       },
       "peerDependencies": {
         "@types/react": "*",
@@ -2197,21 +2220,26 @@
         }
       }
     },
-    "node_modules/@radix-ui/react-collection/node_modules/@radix-ui/react-slot": {
+    "node_modules/@radix-ui/react-collection/node_modules/@radix-ui/react-primitive": {
-      "version": "1.2.2",
+      "version": "2.1.3",
-      "resolved": "https://registry.npmjs.org/@radix-ui/react-slot/-/react-slot-1.2.2.tgz",
+      "resolved": "https://registry.npmjs.org/@radix-ui/react-primitive/-/react-primitive-2.1.3.tgz",
-      "integrity": "sha512-y7TBO4xN4Y94FvcWIOIh18fM4R1A8S4q1jhoz4PNzOoHsFcN8pogcFmZrTYAm4F9VRUrWP/Mw7xSKybIeRI+CQ==",
+      "integrity": "sha512-m9gTwRkhy2lvCPe6QJp4d3G1TYEUHn/FzJUtq9MjH46an1wJU+GdoGC5VLof8RX8Ft/DlpshApkhswDLZzHIcQ==",
       "license": "MIT",
       "dependencies": {
-        "@radix-ui/react-compose-refs": "1.1.2"
+        "@radix-ui/react-slot": "1.2.3"
       },
       "peerDependencies": {
         "@types/react": "*",
-        "react": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc"
+        "@types/react-dom": "*",
+        "react": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc",
+        "react-dom": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc"
       },
       "peerDependenciesMeta": {
         "@types/react": {
           "optional": true
+        },
+        "@types/react-dom": {
+          "optional": true
         }
       }
     },
@@ -2342,17 +2370,17 @@
       }
     },
     "node_modules/@radix-ui/react-dropdown-menu": {
-      "version": "2.1.14",
+      "version": "2.1.16",
-      "resolved": "https://registry.npmjs.org/@radix-ui/react-dropdown-menu/-/react-dropdown-menu-2.1.14.tgz",
+      "resolved": "https://registry.npmjs.org/@radix-ui/react-dropdown-menu/-/react-dropdown-menu-2.1.16.tgz",
-      "integrity": "sha512-lzuyNjoWOoaMFE/VC5FnAAYM16JmQA8ZmucOXtlhm2kKR5TSU95YLAueQ4JYuRmUJmBvSqXaVFGIfuukybwZJQ==",
+      "integrity": "sha512-1PLGQEynI/3OX/ftV54COn+3Sud/Mn8vALg2rWnBLnRaGtJDduNW/22XjlGgPdpcIbiQxjKtb7BkcjP00nqfJw==",
       "license": "MIT",
       "dependencies": {
-        "@radix-ui/primitive": "1.1.2",
+        "@radix-ui/primitive": "1.1.3",
         "@radix-ui/react-compose-refs": "1.1.2",
         "@radix-ui/react-context": "1.1.2",
         "@radix-ui/react-id": "1.1.1",
-        "@radix-ui/react-menu": "2.1.14",
+        "@radix-ui/react-menu": "2.1.16",
-        "@radix-ui/react-primitive": "2.1.2",
+        "@radix-ui/react-primitive": "2.1.3",
         "@radix-ui/react-use-controllable-state": "1.2.2"
       },
       "peerDependencies": {
@@ -2370,6 +2398,35 @@
         }
       }
     },
+    "node_modules/@radix-ui/react-dropdown-menu/node_modules/@radix-ui/primitive": {
+      "version": "1.1.3",
+      "resolved": "https://registry.npmjs.org/@radix-ui/primitive/-/primitive-1.1.3.tgz",
+      "integrity": "sha512-JTF99U/6XIjCBo0wqkU5sK10glYe27MRRsfwoiq5zzOEZLHU3A3KCMa5X/azekYRCJ0HlwI0crAXS/5dEHTzDg==",
+      "license": "MIT"
+    },
+    "node_modules/@radix-ui/react-dropdown-menu/node_modules/@radix-ui/react-primitive": {
+      "version": "2.1.3",
+      "resolved": "https://registry.npmjs.org/@radix-ui/react-primitive/-/react-primitive-2.1.3.tgz",
+      "integrity": "sha512-m9gTwRkhy2lvCPe6QJp4d3G1TYEUHn/FzJUtq9MjH46an1wJU+GdoGC5VLof8RX8Ft/DlpshApkhswDLZzHIcQ==",
+      "license": "MIT",
+      "dependencies": {
+        "@radix-ui/react-slot": "1.2.3"
+      },
+      "peerDependencies": {
+        "@types/react": "*",
+        "@types/react-dom": "*",
+        "react": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc",
+        "react-dom": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc"
+      },
+      "peerDependenciesMeta": {
+        "@types/react": {
+          "optional": true
+        },
+        "@types/react-dom": {
+          "optional": true
+        }
+      }
+    },
     "node_modules/@radix-ui/react-focus-guards": {
       "version": "1.1.2",
       "resolved": "https://registry.npmjs.org/@radix-ui/react-focus-guards/-/react-focus-guards-1.1.2.tgz",
@@ -2429,26 +2486,26 @@
       }
     },
     "node_modules/@radix-ui/react-menu": {
-      "version": "2.1.14",
+      "version": "2.1.16",
-      "resolved": "https://registry.npmjs.org/@radix-ui/react-menu/-/react-menu-2.1.14.tgz",
+      "resolved": "https://registry.npmjs.org/@radix-ui/react-menu/-/react-menu-2.1.16.tgz",
-      "integrity": "sha512-0zSiBAIFq9GSKoSH5PdEaQeRB3RnEGxC+H2P0egtnKoKKLNBH8VBHyVO6/jskhjAezhOIplyRUj7U2lds9A+Yg==",
+      "integrity": "sha512-72F2T+PLlphrqLcAotYPp0uJMr5SjP5SL01wfEspJbru5Zs5vQaSHb4VB3ZMJPimgHHCHG7gMOeOB9H3Hdmtxg==",
       "license": "MIT",
       "dependencies": {
-        "@radix-ui/primitive": "1.1.2",
+        "@radix-ui/primitive": "1.1.3",
-        "@radix-ui/react-collection": "1.1.6",
+        "@radix-ui/react-collection": "1.1.7",
         "@radix-ui/react-compose-refs": "1.1.2",
         "@radix-ui/react-context": "1.1.2",
         "@radix-ui/react-direction": "1.1.1",
-        "@radix-ui/react-dismissable-layer": "1.1.9",
+        "@radix-ui/react-dismissable-layer": "1.1.11",
-        "@radix-ui/react-focus-guards": "1.1.2",
+        "@radix-ui/react-focus-guards": "1.1.3",
-        "@radix-ui/react-focus-scope": "1.1.6",
+        "@radix-ui/react-focus-scope": "1.1.7",
         "@radix-ui/react-id": "1.1.1",
-        "@radix-ui/react-popper": "1.2.6",
+        "@radix-ui/react-popper": "1.2.8",
-        "@radix-ui/react-portal": "1.1.8",
+        "@radix-ui/react-portal": "1.1.9",
-        "@radix-ui/react-presence": "1.1.4",
+        "@radix-ui/react-presence": "1.1.5",
-        "@radix-ui/react-primitive": "2.1.2",
+        "@radix-ui/react-primitive": "2.1.3",
-        "@radix-ui/react-roving-focus": "1.1.9",
+        "@radix-ui/react-roving-focus": "1.1.11",
-        "@radix-ui/react-slot": "1.2.2",
+        "@radix-ui/react-slot": "1.2.3",
         "@radix-ui/react-use-callback-ref": "1.1.1",
         "aria-hidden": "^1.2.4",
         "react-remove-scroll": "^2.6.3"
@@ -2468,14 +2525,44 @@
         }
       }
     },
-    "node_modules/@radix-ui/react-menu/node_modules/@radix-ui/react-slot": {
+    "node_modules/@radix-ui/react-menu/node_modules/@radix-ui/primitive": {
-      "version": "1.2.2",
+      "version": "1.1.3",
-      "resolved": "https://registry.npmjs.org/@radix-ui/react-slot/-/react-slot-1.2.2.tgz",
+      "resolved": "https://registry.npmjs.org/@radix-ui/primitive/-/primitive-1.1.3.tgz",
-      "integrity": "sha512-y7TBO4xN4Y94FvcWIOIh18fM4R1A8S4q1jhoz4PNzOoHsFcN8pogcFmZrTYAm4F9VRUrWP/Mw7xSKybIeRI+CQ==",
+      "integrity": "sha512-JTF99U/6XIjCBo0wqkU5sK10glYe27MRRsfwoiq5zzOEZLHU3A3KCMa5X/azekYRCJ0HlwI0crAXS/5dEHTzDg==",
+      "license": "MIT"
+    },
+    "node_modules/@radix-ui/react-menu/node_modules/@radix-ui/react-dismissable-layer": {
+      "version": "1.1.11",
+      "resolved": "https://registry.npmjs.org/@radix-ui/react-dismissable-layer/-/react-dismissable-layer-1.1.11.tgz",
+      "integrity": "sha512-Nqcp+t5cTB8BinFkZgXiMJniQH0PsUt2k51FUhbdfeKvc4ACcG2uQniY/8+h1Yv6Kza4Q7lD7PQV0z0oicE0Mg==",
       "license": "MIT",
       "dependencies": {
-        "@radix-ui/react-compose-refs": "1.1.2"
+        "@radix-ui/primitive": "1.1.3",
+        "@radix-ui/react-compose-refs": "1.1.2",
+        "@radix-ui/react-primitive": "2.1.3",
+        "@radix-ui/react-use-callback-ref": "1.1.1",
+        "@radix-ui/react-use-escape-keydown": "1.1.1"
       },
+      "peerDependencies": {
+        "@types/react": "*",
+        "@types/react-dom": "*",
+        "react": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc",
+        "react-dom": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc"
+      },
+      "peerDependenciesMeta": {
+        "@types/react": {
+          "optional": true
+        },
+        "@types/react-dom": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/@radix-ui/react-menu/node_modules/@radix-ui/react-focus-guards": {
+      "version": "1.1.3",
+      "resolved": "https://registry.npmjs.org/@radix-ui/react-focus-guards/-/react-focus-guards-1.1.3.tgz",
+      "integrity": "sha512-0rFg/Rj2Q62NCm62jZw0QX7a3sz6QCQU0LpZdNrJX8byRGaGVTqbrW9jAoIAHyMQqsNpeZ81YgSizOt5WXq0Pw==",
+      "license": "MIT",
       "peerDependencies": {
         "@types/react": "*",
         "react": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc"
@@ -2486,17 +2573,113 @@
         }
       }
     },
+    "node_modules/@radix-ui/react-menu/node_modules/@radix-ui/react-focus-scope": {
+      "version": "1.1.7",
+      "resolved": "https://registry.npmjs.org/@radix-ui/react-focus-scope/-/react-focus-scope-1.1.7.tgz",
+      "integrity": "sha512-t2ODlkXBQyn7jkl6TNaw/MtVEVvIGelJDCG41Okq/KwUsJBwQ4XVZsHAVUkK4mBv3ewiAS3PGuUWuY2BoK4ZUw==",
+      "license": "MIT",
+      "dependencies": {
+        "@radix-ui/react-compose-refs": "1.1.2",
+        "@radix-ui/react-primitive": "2.1.3",
+        "@radix-ui/react-use-callback-ref": "1.1.1"
+      },
+      "peerDependencies": {
+        "@types/react": "*",
+        "@types/react-dom": "*",
+        "react": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc",
+        "react-dom": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc"
+      },
+      "peerDependenciesMeta": {
+        "@types/react": {
+          "optional": true
+        },
+        "@types/react-dom": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/@radix-ui/react-menu/node_modules/@radix-ui/react-portal": {
+      "version": "1.1.9",
+      "resolved": "https://registry.npmjs.org/@radix-ui/react-portal/-/react-portal-1.1.9.tgz",
+      "integrity": "sha512-bpIxvq03if6UNwXZ+HTK71JLh4APvnXntDc6XOX8UVq4XQOVl7lwok0AvIl+b8zgCw3fSaVTZMpAPPagXbKmHQ==",
+      "license": "MIT",
+      "dependencies": {
+        "@radix-ui/react-primitive": "2.1.3",
+        "@radix-ui/react-use-layout-effect": "1.1.1"
+      },
+      "peerDependencies": {
+        "@types/react": "*",
+        "@types/react-dom": "*",
+        "react": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc",
+        "react-dom": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc"
+      },
+      "peerDependenciesMeta": {
+        "@types/react": {
+          "optional": true
+        },
+        "@types/react-dom": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/@radix-ui/react-menu/node_modules/@radix-ui/react-presence": {
+      "version": "1.1.5",
+      "resolved": "https://registry.npmjs.org/@radix-ui/react-presence/-/react-presence-1.1.5.tgz",
+      "integrity": "sha512-/jfEwNDdQVBCNvjkGit4h6pMOzq8bHkopq458dPt2lMjx+eBQUohZNG9A7DtO/O5ukSbxuaNGXMjHicgwy6rQQ==",
+      "license": "MIT",
+      "dependencies": {
+        "@radix-ui/react-compose-refs": "1.1.2",
+        "@radix-ui/react-use-layout-effect": "1.1.1"
+      },
+      "peerDependencies": {
+        "@types/react": "*",
+        "@types/react-dom": "*",
+        "react": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc",
+        "react-dom": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc"
+      },
+      "peerDependenciesMeta": {
+        "@types/react": {
+          "optional": true
+        },
+        "@types/react-dom": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/@radix-ui/react-menu/node_modules/@radix-ui/react-primitive": {
+      "version": "2.1.3",
+      "resolved": "https://registry.npmjs.org/@radix-ui/react-primitive/-/react-primitive-2.1.3.tgz",
+      "integrity": "sha512-m9gTwRkhy2lvCPe6QJp4d3G1TYEUHn/FzJUtq9MjH46an1wJU+GdoGC5VLof8RX8Ft/DlpshApkhswDLZzHIcQ==",
+      "license": "MIT",
+      "dependencies": {
+        "@radix-ui/react-slot": "1.2.3"
+      },
+      "peerDependencies": {
+        "@types/react": "*",
+        "@types/react-dom": "*",
+        "react": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc",
+        "react-dom": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc"
+      },
+      "peerDependenciesMeta": {
+        "@types/react": {
+          "optional": true
+        },
+        "@types/react-dom": {
+          "optional": true
+        }
+      }
+    },
     "node_modules/@radix-ui/react-popper": {
-      "version": "1.2.6",
+      "version": "1.2.8",
-      "resolved": "https://registry.npmjs.org/@radix-ui/react-popper/-/react-popper-1.2.6.tgz",
+      "resolved": "https://registry.npmjs.org/@radix-ui/react-popper/-/react-popper-1.2.8.tgz",
-      "integrity": "sha512-7iqXaOWIjDBfIG7aq8CUEeCSsQMLFdn7VEE8TaFz704DtEzpPHR7w/uuzRflvKgltqSAImgcmxQ7fFX3X7wasg==",
+      "integrity": "sha512-0NJQ4LFFUuWkE7Oxf0htBKS6zLkkjBH+hM1uk7Ng705ReR8m/uelduy1DBo0PyBXPKVnBA6YBlU94MBGXrSBCw==",
       "license": "MIT",
       "dependencies": {
         "@floating-ui/react-dom": "^2.0.0",
-        "@radix-ui/react-arrow": "1.1.6",
+        "@radix-ui/react-arrow": "1.1.7",
         "@radix-ui/react-compose-refs": "1.1.2",
         "@radix-ui/react-context": "1.1.2",
-        "@radix-ui/react-primitive": "2.1.2",
+        "@radix-ui/react-primitive": "2.1.3",
         "@radix-ui/react-use-callback-ref": "1.1.1",
         "@radix-ui/react-use-layout-effect": "1.1.1",
         "@radix-ui/react-use-rect": "1.1.1",
@@ -2518,6 +2701,29 @@
         }
       }
     },
+    "node_modules/@radix-ui/react-popper/node_modules/@radix-ui/react-primitive": {
+      "version": "2.1.3",
+      "resolved": "https://registry.npmjs.org/@radix-ui/react-primitive/-/react-primitive-2.1.3.tgz",
+      "integrity": "sha512-m9gTwRkhy2lvCPe6QJp4d3G1TYEUHn/FzJUtq9MjH46an1wJU+GdoGC5VLof8RX8Ft/DlpshApkhswDLZzHIcQ==",
+      "license": "MIT",
+      "dependencies": {
+        "@radix-ui/react-slot": "1.2.3"
+      },
+      "peerDependencies": {
+        "@types/react": "*",
+        "@types/react-dom": "*",
+        "react": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc",
+        "react-dom": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc"
+      },
+      "peerDependenciesMeta": {
+        "@types/react": {
+          "optional": true
+        },
+        "@types/react-dom": {
+          "optional": true
+        }
+      }
+    },
     "node_modules/@radix-ui/react-portal": {
       "version": "1.1.8",
       "resolved": "https://registry.npmjs.org/@radix-ui/react-portal/-/react-portal-1.1.8.tgz",
@@ -2608,18 +2814,18 @@
       }
     },
     "node_modules/@radix-ui/react-roving-focus": {
-      "version": "1.1.9",
+      "version": "1.1.11",
-      "resolved": "https://registry.npmjs.org/@radix-ui/react-roving-focus/-/react-roving-focus-1.1.9.tgz",
+      "resolved": "https://registry.npmjs.org/@radix-ui/react-roving-focus/-/react-roving-focus-1.1.11.tgz",
-      "integrity": "sha512-ZzrIFnMYHHCNqSNCsuN6l7wlewBEq0O0BCSBkabJMFXVO51LRUTq71gLP1UxFvmrXElqmPjA5VX7IqC9VpazAQ==",
+      "integrity": "sha512-7A6S9jSgm/S+7MdtNDSb+IU859vQqJ/QAtcYQcfFC6W8RS4IxIZDldLR0xqCFZ6DCyrQLjLPsxtTNch5jVA4lA==",
       "license": "MIT",
       "dependencies": {
-        "@radix-ui/primitive": "1.1.2",
+        "@radix-ui/primitive": "1.1.3",
-        "@radix-ui/react-collection": "1.1.6",
+        "@radix-ui/react-collection": "1.1.7",
         "@radix-ui/react-compose-refs": "1.1.2",
         "@radix-ui/react-context": "1.1.2",
         "@radix-ui/react-direction": "1.1.1",
         "@radix-ui/react-id": "1.1.1",
-        "@radix-ui/react-primitive": "2.1.2",
+        "@radix-ui/react-primitive": "2.1.3",
         "@radix-ui/react-use-callback-ref": "1.1.1",
         "@radix-ui/react-use-controllable-state": "1.2.2"
       },
@@ -2638,6 +2844,35 @@
         }
       }
     },
+    "node_modules/@radix-ui/react-roving-focus/node_modules/@radix-ui/primitive": {
+      "version": "1.1.3",
+      "resolved": "https://registry.npmjs.org/@radix-ui/primitive/-/primitive-1.1.3.tgz",
+      "integrity": "sha512-JTF99U/6XIjCBo0wqkU5sK10glYe27MRRsfwoiq5zzOEZLHU3A3KCMa5X/azekYRCJ0HlwI0crAXS/5dEHTzDg==",
+      "license": "MIT"
+    },
+    "node_modules/@radix-ui/react-roving-focus/node_modules/@radix-ui/react-primitive": {
+      "version": "2.1.3",
+      "resolved": "https://registry.npmjs.org/@radix-ui/react-primitive/-/react-primitive-2.1.3.tgz",
+      "integrity": "sha512-m9gTwRkhy2lvCPe6QJp4d3G1TYEUHn/FzJUtq9MjH46an1wJU+GdoGC5VLof8RX8Ft/DlpshApkhswDLZzHIcQ==",
+      "license": "MIT",
+      "dependencies": {
+        "@radix-ui/react-slot": "1.2.3"
+      },
+      "peerDependencies": {
+        "@types/react": "*",
+        "@types/react-dom": "*",
+        "react": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc",
+        "react-dom": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc"
+      },
+      "peerDependenciesMeta": {
+        "@types/react": {
+          "optional": true
+        },
+        "@types/react-dom": {
+          "optional": true
+        }
+      }
+    },
     "node_modules/@radix-ui/react-select": {
       "version": "2.2.5",
       "resolved": "https://registry.npmjs.org/@radix-ui/react-select/-/react-select-2.2.5.tgz",
@@ -2681,55 +2916,6 @@
         }
       }
     },
-    "node_modules/@radix-ui/react-select/node_modules/@radix-ui/react-arrow": {
-      "version": "1.1.7",
-      "resolved": "https://registry.npmjs.org/@radix-ui/react-arrow/-/react-arrow-1.1.7.tgz",
-      "integrity": "sha512-F+M1tLhO+mlQaOWspE8Wstg+z6PwxwRd8oQ8IXceWz92kfAmalTRf0EjrouQeo7QssEPfCn05B4Ihs1K9WQ/7w==",
-      "license": "MIT",
-      "dependencies": {
-        "@radix-ui/react-primitive": "2.1.3"
-      },
-      "peerDependencies": {
-        "@types/react": "*",
-        "@types/react-dom": "*",
-        "react": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc",
-        "react-dom": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc"
-      },
-      "peerDependenciesMeta": {
-        "@types/react": {
-          "optional": true
-        },
-        "@types/react-dom": {
-          "optional": true
-        }
-      }
-    },
-    "node_modules/@radix-ui/react-select/node_modules/@radix-ui/react-collection": {
-      "version": "1.1.7",
-      "resolved": "https://registry.npmjs.org/@radix-ui/react-collection/-/react-collection-1.1.7.tgz",
-      "integrity": "sha512-Fh9rGN0MoI4ZFUNyfFVNU4y9LUz93u9/0K+yLgA2bwRojxM8JU1DyvvMBabnZPBgMWREAJvU2jjVzq+LrFUglw==",
-      "license": "MIT",
-      "dependencies": {
-        "@radix-ui/react-compose-refs": "1.1.2",
-        "@radix-ui/react-context": "1.1.2",
-        "@radix-ui/react-primitive": "2.1.3",
-        "@radix-ui/react-slot": "1.2.3"
-      },
-      "peerDependencies": {
-        "@types/react": "*",
-        "@types/react-dom": "*",
-        "react": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc",
-        "react-dom": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc"
-      },
-      "peerDependenciesMeta": {
-        "@types/react": {
-          "optional": true
-        },
-        "@types/react-dom": {
-          "optional": true
-        }
-      }
-    },
     "node_modules/@radix-ui/react-select/node_modules/@radix-ui/react-dismissable-layer": {
       "version": "1.1.10",
       "resolved": "https://registry.npmjs.org/@radix-ui/react-dismissable-layer/-/react-dismissable-layer-1.1.10.tgz",
@@ -2965,29 +3151,6 @@
       "integrity": "sha512-JTF99U/6XIjCBo0wqkU5sK10glYe27MRRsfwoiq5zzOEZLHU3A3KCMa5X/azekYRCJ0HlwI0crAXS/5dEHTzDg==",
       "license": "MIT"
     },
-    "node_modules/@radix-ui/react-tooltip/node_modules/@radix-ui/react-arrow": {
-      "version": "1.1.7",
-      "resolved": "https://registry.npmjs.org/@radix-ui/react-arrow/-/react-arrow-1.1.7.tgz",
-      "integrity": "sha512-F+M1tLhO+mlQaOWspE8Wstg+z6PwxwRd8oQ8IXceWz92kfAmalTRf0EjrouQeo7QssEPfCn05B4Ihs1K9WQ/7w==",
-      "license": "MIT",
-      "dependencies": {
-        "@radix-ui/react-primitive": "2.1.3"
-      },
-      "peerDependencies": {
-        "@types/react": "*",
-        "@types/react-dom": "*",
-        "react": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc",
-        "react-dom": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc"
-      },
-      "peerDependenciesMeta": {
-        "@types/react": {
-          "optional": true
-        },
-        "@types/react-dom": {
-          "optional": true
-        }
-      }
-    },
     "node_modules/@radix-ui/react-tooltip/node_modules/@radix-ui/react-dismissable-layer": {
       "version": "1.1.11",
       "resolved": "https://registry.npmjs.org/@radix-ui/react-dismissable-layer/-/react-dismissable-layer-1.1.11.tgz",
@@ -3015,38 +3178,6 @@
         }
       }
     },
-    "node_modules/@radix-ui/react-tooltip/node_modules/@radix-ui/react-popper": {
-      "version": "1.2.8",
-      "resolved": "https://registry.npmjs.org/@radix-ui/react-popper/-/react-popper-1.2.8.tgz",
-      "integrity": "sha512-0NJQ4LFFUuWkE7Oxf0htBKS6zLkkjBH+hM1uk7Ng705ReR8m/uelduy1DBo0PyBXPKVnBA6YBlU94MBGXrSBCw==",
-      "license": "MIT",
-      "dependencies": {
-        "@floating-ui/react-dom": "^2.0.0",
-        "@radix-ui/react-arrow": "1.1.7",
-        "@radix-ui/react-compose-refs": "1.1.2",
-        "@radix-ui/react-context": "1.1.2",
-        "@radix-ui/react-primitive": "2.1.3",
-        "@radix-ui/react-use-callback-ref": "1.1.1",
-        "@radix-ui/react-use-layout-effect": "1.1.1",
-        "@radix-ui/react-use-rect": "1.1.1",
-        "@radix-ui/react-use-size": "1.1.1",
-        "@radix-ui/rect": "1.1.1"
-      },
-      "peerDependencies": {
-        "@types/react": "*",
-        "@types/react-dom": "*",
-        "react": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc",
-        "react-dom": "^16.8 || ^17.0 || ^18.0 || ^19.0 || ^19.0.0-rc"
-      },
-      "peerDependenciesMeta": {
-        "@types/react": {
-          "optional": true
-        },
-        "@types/react-dom": {
-          "optional": true
-        }
-      }
-    },
     "node_modules/@radix-ui/react-tooltip/node_modules/@radix-ui/react-portal": {
       "version": "1.1.9",
       "resolved": "https://registry.npmjs.org/@radix-ui/react-portal/-/react-portal-1.1.9.tgz",
@@ -3447,6 +3578,13 @@
         "tailwindcss": "4.1.6"
       }
     },
+    "node_modules/@tailwindcss/node/node_modules/tailwindcss": {
+      "version": "4.1.6",
+      "resolved": "https://registry.npmjs.org/tailwindcss/-/tailwindcss-4.1.6.tgz",
+      "integrity": "sha512-j0cGLTreM6u4OWzBeLBpycK0WIh8w7kSwcUsQZoGLHZ7xDTdM69lN64AgoIEEwFi0tnhs4wSykUa5YWxAzgFYg==",
+      "dev": true,
+      "license": "MIT"
+    },
     "node_modules/@tailwindcss/oxide": {
       "version": "4.1.6",
       "resolved": "https://registry.npmjs.org/@tailwindcss/oxide/-/oxide-4.1.6.tgz",
@@ -3707,6 +3845,13 @@
         "tailwindcss": "4.1.6"
       }
     },
+    "node_modules/@tailwindcss/postcss/node_modules/tailwindcss": {
+      "version": "4.1.6",
+      "resolved": "https://registry.npmjs.org/tailwindcss/-/tailwindcss-4.1.6.tgz",
+      "integrity": "sha512-j0cGLTreM6u4OWzBeLBpycK0WIh8w7kSwcUsQZoGLHZ7xDTdM69lN64AgoIEEwFi0tnhs4wSykUa5YWxAzgFYg==",
+      "dev": true,
+      "license": "MIT"
+    },
     "node_modules/@testing-library/dom": {
       "version": "10.4.1",
       "resolved": "https://registry.npmjs.org/@testing-library/dom/-/dom-10.4.1.tgz",
@@ -4079,9 +4224,9 @@
       }
     },
     "node_modules/@types/react-dom": {
-      "version": "19.1.5",
+      "version": "19.1.9",
-      "resolved": "https://registry.npmjs.org/@types/react-dom/-/react-dom-19.1.5.tgz",
+      "resolved": "https://registry.npmjs.org/@types/react-dom/-/react-dom-19.1.9.tgz",
-      "integrity": "sha512-CMCjrWucUBZvohgZxkjd6S9h0nZxXjzus6yDfUb+xLxYM7VvjKNH1tQrE9GWLql1XoOP4/Ds3bwFqShHUYraGg==",
+      "integrity": "sha512-qXRuZaOsAdXKFyOhRBg6Lqqc0yay13vN7KrIg4L7N4aaHN68ma9OK3NE1BoDFgFOTfM7zg+3/8+2n8rLUH3OKQ==",
       "devOptional": true,
       "license": "MIT",
       "peerDependencies": {
@@ -10147,9 +10292,9 @@
       "license": "MIT"
     },
     "node_modules/llama-stack-client": {
-      "version": "0.2.20",
+      "version": "0.2.21",
-      "resolved": "https://registry.npmjs.org/llama-stack-client/-/llama-stack-client-0.2.20.tgz",
+      "resolved": "https://registry.npmjs.org/llama-stack-client/-/llama-stack-client-0.2.21.tgz",
-      "integrity": "sha512-1vD5nizTX5JEW8TADxKgy/P1W8YZoPSpdnmfxbdYbWgpQ3BWtbvLS6jmDk7VwVA5fRC4895VfHsRDfS1liHarw==",
+      "integrity": "sha512-rjU2Vx5xStxDYavU8K1An/SYXiQQjroLcK98B+p0Paz/a7OgRao2S0YwvThJjPUyChY4fO03UIXP9LpmHqlXWQ==",
       "license": "MIT",
       "dependencies": {
         "@types/node": "^18.11.18",
@@ -10240,9 +10385,9 @@
       "license": "ISC"
     },
     "node_modules/lucide-react": {
-      "version": "0.510.0",
+      "version": "0.542.0",
-      "resolved": "https://registry.npmjs.org/lucide-react/-/lucide-react-0.510.0.tgz",
+      "resolved": "https://registry.npmjs.org/lucide-react/-/lucide-react-0.542.0.tgz",
-      "integrity": "sha512-p8SQRAMVh7NhsAIETokSqDrc5CHnDLbV29mMnzaXx+Vc/hnqQzwI2r0FMWCcoTXnbw2KEjy48xwpGdEL+ck06Q==",
+      "integrity": "sha512-w3hD8/SQB7+lzU2r4VdFyzzOzKnUjTZIF/MQJGSSvni7Llewni4vuViRppfRAa2guOsY5k4jZyxw/i9DQHv+dw==",
       "license": "ISC",
       "peerDependencies": {
         "react": "^16.5.1 || ^17.0.0 || ^18.0.0 || ^19.0.0"
@@ -12448,24 +12593,24 @@
       }
     },
     "node_modules/react": {
-      "version": "19.1.0",
+      "version": "19.1.1",
-      "resolved": "https://registry.npmjs.org/react/-/react-19.1.0.tgz",
+      "resolved": "https://registry.npmjs.org/react/-/react-19.1.1.tgz",
-      "integrity": "sha512-FS+XFBNvn3GTAWq26joslQgWNoFu08F4kl0J4CgdNKADkdSGXQyTCnKteIAJy96Br6YbpEU1LSzV5dYtjMkMDg==",
+      "integrity": "sha512-w8nqGImo45dmMIfljjMwOGtbmC/mk4CMYhWIicdSflH91J9TyCyczcPFXJzrZ/ZXcgGRFeP6BU0BEJTw6tZdfQ==",
       "license": "MIT",
       "engines": {
         "node": ">=0.10.0"
       }
     },
     "node_modules/react-dom": {
-      "version": "19.1.0",
+      "version": "19.1.1",
-      "resolved": "https://registry.npmjs.org/react-dom/-/react-dom-19.1.0.tgz",
+      "resolved": "https://registry.npmjs.org/react-dom/-/react-dom-19.1.1.tgz",
-      "integrity": "sha512-Xs1hdnE+DyKgeHJeJznQmYMIBG3TKIHJJT95Q58nHLSrElKlGQqDTR2HQ9fx5CN/Gk6Vh/kupBTDLU11/nDk/g==",
+      "integrity": "sha512-Dlq/5LAZgF0Gaz6yiqZCf6VCcZs1ghAJyrsu84Q/GT0gV+mCxbfmKNoGRKBYMJ8IEdGPqu49YWXD02GCknEDkw==",
       "license": "MIT",
       "dependencies": {
         "scheduler": "^0.26.0"
       },
       "peerDependencies": {
-        "react": "^19.1.0"
+        "react": "^19.1.1"
       }
     },
     "node_modules/react-is": {
@@ -13285,9 +13430,9 @@
       }
     },
     "node_modules/sonner": {
-      "version": "2.0.6",
+      "version": "2.0.7",
-      "resolved": "https://registry.npmjs.org/sonner/-/sonner-2.0.6.tgz",
+      "resolved": "https://registry.npmjs.org/sonner/-/sonner-2.0.7.tgz",
-      "integrity": "sha512-yHFhk8T/DK3YxjFQXIrcHT1rGEeTLliVzWbO0xN8GberVun2RiBnxAjXAYpZrqwEVHBG9asI/Li8TAAhN9m59Q==",
+      "integrity": "sha512-W6ZN4p58k8aDKA4XPcx2hpIQXBRAgyiWVkYhT7CvK6D3iAu7xjvVyhQHg2/iaKJZ1XVJ4r7XuwGL+WGEK37i9w==",
       "license": "MIT",
       "peerDependencies": {
         "react": "^18.0.0 || ^19.0.0 || ^19.0.0-rc",
@@ -13712,9 +13857,9 @@
       }
     },
     "node_modules/tailwindcss": {
-      "version": "4.1.6",
+      "version": "4.1.13",
-      "resolved": "https://registry.npmjs.org/tailwindcss/-/tailwindcss-4.1.6.tgz",
+      "resolved": "https://registry.npmjs.org/tailwindcss/-/tailwindcss-4.1.13.tgz",
-      "integrity": "sha512-j0cGLTreM6u4OWzBeLBpycK0WIh8w7kSwcUsQZoGLHZ7xDTdM69lN64AgoIEEwFi0tnhs4wSykUa5YWxAzgFYg==",
+      "integrity": "sha512-i+zidfmTqtwquj4hMEwdjshYYgMbOrPzb9a0M3ZgNa0JMoZeFC6bxZvO8yr8ozS6ix2SDz0+mvryPeBs2TFE+w==",
       "dev": true,
       "license": "MIT"
     },

llama_stack/ui/package.json

@@ -15,7 +15,7 @@
   "dependencies": {
     "@radix-ui/react-collapsible": "^1.1.12",
     "@radix-ui/react-dialog": "^1.1.13",
-    "@radix-ui/react-dropdown-menu": "^2.1.14",
+    "@radix-ui/react-dropdown-menu": "^2.1.16",
     "@radix-ui/react-select": "^2.2.5",
     "@radix-ui/react-separator": "^1.1.7",
     "@radix-ui/react-slot": "^1.2.3",
@@ -23,18 +23,18 @@
     "class-variance-authority": "^0.7.1",
     "clsx": "^2.1.1",
     "framer-motion": "^12.23.12",
-    "llama-stack-client": "^0.2.20",
+    "llama-stack-client": "^0.2.21",
-    "lucide-react": "^0.510.0",
+    "lucide-react": "^0.542.0",
     "next": "15.3.3",
     "next-auth": "^4.24.11",
     "next-themes": "^0.4.6",
     "react": "^19.0.0",
-    "react-dom": "^19.0.0",
+    "react-dom": "^19.1.1",
     "react-markdown": "^10.1.0",
     "remark-gfm": "^4.0.1",
     "remeda": "^2.30.0",
     "shiki": "^1.29.2",
-    "sonner": "^2.0.6",
+    "sonner": "^2.0.7",
     "tailwind-merge": "^3.3.1"
   },
   "devDependencies": {

pyproject.toml

@@ -7,7 +7,7 @@ required-version = ">=0.7.0"
 
 [project]
 name = "llama_stack"
-version = "0.2.20"
+version = "0.2.21"
 authors = [{ name = "Meta Llama", email = "llama-oss@meta.com" }]
 description = "Llama Stack"
 readme = "README.md"
@@ -31,9 +31,8 @@ dependencies = [
     "huggingface-hub>=0.34.0,<1.0",
     "jinja2>=3.1.6",
     "jsonschema",
-    "llama-stack-client>=0.2.20",
+    "llama-stack-client>=0.2.21",
-    "llama-api-client>=0.1.2",
+    "openai>=1.100.0",                                # for expires_after support
-    "openai>=1.99.6",
     "prompt-toolkit",
     "python-dotenv",
     "python-jose[cryptography]",
@@ -56,7 +55,7 @@ dependencies = [
 ui = [
     "streamlit",
     "pandas",
-    "llama-stack-client>=0.2.20",
+    "llama-stack-client>=0.2.21",
     "streamlit-option-menu",
 ]
 
@@ -81,7 +80,6 @@ dev = [
 unit = [
     "sqlite-vec",
     "ollama",
-    "openai",
     "aiosqlite",
     "aiohttp",
     "psycopg2-binary>=2.9.0",
@@ -93,7 +91,7 @@ unit = [
     "sqlalchemy[asyncio]>=2.0.41",
     "blobfile",
     "faiss-cpu",
-    "pymilvus>=2.5.12",
+    "pymilvus>=2.6.1",
     "milvus-lite>=2.5.0",
     "litellm",
     "together",
@@ -106,7 +104,6 @@ unit = [
 # separately. If you are using "uv" to execute your tests, you can use the "--group" flag to specify extra
 # dependencies.
 test = [
-    "openai>=1.100.0",  # for expires_after support
     "aiosqlite",
     "aiohttp",
     "torch>=2.6.0",
@@ -115,13 +112,13 @@ test = [
     "psycopg2-binary>=2.9.0",
     "pypdf",
     "mcp",
-    "datasets",
+    "datasets>=4.0.0",
     "autoevals",
     "transformers",
     "sqlalchemy",
     "sqlalchemy[asyncio]>=2.0.41",
     "requests",
-    "pymilvus>=2.5.12",
+    "pymilvus>=2.6.1",
     "milvus-lite>=2.5.0",
     "weaviate-client>=4.16.4",
 ]
@@ -146,7 +143,7 @@ docs = [
 ]
 codegen = ["rich", "pydantic", "jinja2>=3.1.6"]
 benchmark = [
-    "locust>=2.37.14",
+    "locust>=2.39.1",
 ]
 
 [project.urls]

scripts/get_setup_env.py

@@ -0,0 +1,71 @@
+#!/usr/bin/env python3
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+"""
+Small helper script to extract environment variables from a test setup.
+Used by integration-tests.sh to set environment variables before starting the server.
+"""
+
+import argparse
+import sys
+
+from tests.integration.suites import SETUP_DEFINITIONS, SUITE_DEFINITIONS
+
+
+def get_setup_env_vars(setup_name, suite_name=None):
+    """
+    Get environment variables for a setup, with optional suite default fallback.
+
+    Args:
+        setup_name: Name of the setup (e.g., 'ollama', 'gpt')
+        suite_name: Optional suite name to get default setup if setup_name is None
+
+    Returns:
+        Dictionary of environment variables
+    """
+    # If no setup specified, try to get default from suite
+    if not setup_name and suite_name:
+        suite = SUITE_DEFINITIONS.get(suite_name)
+        if suite and suite.default_setup:
+            setup_name = suite.default_setup
+
+    if not setup_name:
+        return {}
+
+    setup = SETUP_DEFINITIONS.get(setup_name)
+    if not setup:
+        print(
+            f"Error: Unknown setup '{setup_name}'. Available: {', '.join(sorted(SETUP_DEFINITIONS.keys()))}",
+            file=sys.stderr,
+        )
+        sys.exit(1)
+
+    return setup.env
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Extract environment variables from a test setup")
+    parser.add_argument("--setup", help="Setup name (e.g., ollama, gpt)")
+    parser.add_argument("--suite", help="Suite name to get default setup from if --setup not provided")
+    parser.add_argument("--format", choices=["bash", "json"], default="bash", help="Output format (default: bash)")
+
+    args = parser.parse_args()
+
+    env_vars = get_setup_env_vars(args.setup, args.suite)
+
+    if args.format == "bash":
+        # Output as bash export statements
+        for key, value in env_vars.items():
+            print(f"export {key}='{value}'")
+    elif args.format == "json":
+        import json
+
+        print(json.dumps(env_vars))
+
+
+if __name__ == "__main__":
+    main()

scripts/github/schedule-record-workflow.sh

@@ -14,8 +14,8 @@ set -euo pipefail
 # Default values
 BRANCH=""
 TEST_SUBDIRS=""
-TEST_PROVIDER="ollama"
+TEST_SETUP="ollama"
-RUN_VISION_TESTS=false
+TEST_SUITE="base"
 TEST_PATTERN=""
 
 # Help function
@@ -27,24 +27,24 @@ Trigger the integration test recording workflow remotely. This way you do not ne
 
 OPTIONS:
     -b, --branch BRANCH         Branch to run the workflow on (defaults to current branch)
-    -s, --test-subdirs DIRS     Comma-separated list of test subdirectories to run (REQUIRED)
+    -t, --suite SUITE           Test suite to use: base, responses, vision, etc. (default: base)
-    -p, --test-provider PROVIDER Test provider to use: vllm or ollama (default: ollama)
+    -p, --setup SETUP           Test setup to use: vllm, ollama, gpt, etc. (default: ollama)
-    -v, --run-vision-tests      Include vision tests in the recording
+    -s, --subdirs DIRS          Comma-separated list of test subdirectories to run (overrides suite)
-    -k, --test-pattern PATTERN  Regex pattern to pass to pytest -k
+    -k, --pattern PATTERN       Regex pattern to pass to pytest -k
     -h, --help                  Show this help message
 
 EXAMPLES:
     # Record tests for current branch with agents subdirectory
-    $0 --test-subdirs "agents"
+    $0 --subdirs "agents"
 
     # Record tests for specific branch with vision tests
-    $0 -b my-feature-branch --test-subdirs "inference" --run-vision-tests
+    $0 -b my-feature-branch --suite vision
 
-    # Record multiple test subdirectories with specific provider
+    # Record multiple test subdirectories with specific setup
-    $0 --test-subdirs "agents,inference" --test-provider vllm
+    $0 --subdirs "agents,inference" --setup vllm
 
     # Record tests matching a specific pattern
-    $0 --test-subdirs "inference" --test-pattern "test_streaming"
+    $0 --subdirs "inference" --pattern "test_streaming"
 
 EOF
 }
@@ -63,19 +63,19 @@ while [[ $# -gt 0 ]]; do
             BRANCH="$2"
             shift 2
             ;;
-        -s|--test-subdirs)
+        -s|--subdirs)
             TEST_SUBDIRS="$2"
             shift 2
             ;;
-        -p|--test-provider)
+        -p|--setup)
-            TEST_PROVIDER="$2"
+            TEST_SETUP="$2"
             shift 2
             ;;
-        -v|--run-vision-tests)
+        -t|--suite)
-            RUN_VISION_TESTS=true
+            TEST_SUITE="$2"
-            shift
+            shift 2
             ;;
-        -k|--test-pattern)
+        -k|--pattern)
             TEST_PATTERN="$2"
             shift 2
             ;;
@@ -92,22 +92,17 @@ while [[ $# -gt 0 ]]; do
 done
 
 # Validate required parameters
-if [[ -z "$TEST_SUBDIRS" ]]; then
+if [[ -z "$TEST_SUBDIRS" && -z "$TEST_SUITE" ]]; then
-    echo "Error: --test-subdirs is required"
+    echo "Error: --subdirs or --suite is required"
-    echo "Please specify which test subdirectories to run, e.g.:"
+    echo "Please specify which test subdirectories to run or test suite to use, e.g.:"
-    echo "  $0 --test-subdirs \"agents,inference\""
+    echo "  $0 --subdirs \"agents,inference\""
-    echo "  $0 --test-subdirs \"inference\" --run-vision-tests"
+    echo "  $0 --suite vision"
     echo ""
     exit 1
 fi
 
-# Validate test provider
+# Validate test setup (optional - setups are validated by the workflow itself)
-if [[ "$TEST_PROVIDER" != "vllm" && "$TEST_PROVIDER" != "ollama" ]]; then
+# Common setups: ollama, vllm, gpt, etc.
-    echo "❌ Error: Invalid test provider '$TEST_PROVIDER'"
-    echo "   Supported providers: vllm, ollama"
-    echo "   Example: $0 --test-subdirs \"agents\" --test-provider vllm"
-    exit 1
-fi
 
 # Check if required tools are installed
 if ! command -v gh &> /dev/null; then
@@ -237,22 +232,25 @@ fi
 # Build the workflow dispatch command
 echo "Triggering integration test recording workflow..."
 echo "Branch: $BRANCH"
-echo "Test provider: $TEST_PROVIDER"
+echo "Test setup: $TEST_SETUP"
 echo "Test subdirs: $TEST_SUBDIRS"
-echo "Run vision tests: $RUN_VISION_TESTS"
+echo "Test suite: $TEST_SUITE"
 echo "Test pattern: ${TEST_PATTERN:-"(none)"}"
 echo ""
 
 # Prepare inputs for gh workflow run
-INPUTS="-f test-subdirs='$TEST_SUBDIRS'"
+INPUTS=
-if [[ -n "$TEST_PROVIDER" ]]; then
+if [[ -n "$TEST_SUBDIRS" ]]; then
-    INPUTS="$INPUTS -f test-provider='$TEST_PROVIDER'"
+    INPUTS="$INPUTS -f subdirs='$TEST_SUBDIRS'"
 fi
-if [[ "$RUN_VISION_TESTS" == "true" ]]; then
+if [[ -n "$TEST_SETUP" ]]; then
-    INPUTS="$INPUTS -f run-vision-tests=true"
+    INPUTS="$INPUTS -f test-setup='$TEST_SETUP'"
+fi
+if [[ -n "$TEST_SUITE" ]]; then
+    INPUTS="$INPUTS -f suite='$TEST_SUITE'"
 fi
 if [[ -n "$TEST_PATTERN" ]]; then
-    INPUTS="$INPUTS -f test-pattern='$TEST_PATTERN'"
+    INPUTS="$INPUTS -f pattern='$TEST_PATTERN'"
 fi
 
 # Run the workflow

scripts/integration-tests.sh

@@ -13,10 +13,10 @@ set -euo pipefail
 
 # Default values
 STACK_CONFIG=""
-PROVIDER=""
+TEST_SUITE="base"
+TEST_SETUP=""
 TEST_SUBDIRS=""
 TEST_PATTERN=""
-RUN_VISION_TESTS="false"
 INFERENCE_MODE="replay"
 EXTRA_PARAMS=""
 
@@ -27,25 +27,30 @@ Usage: $0 [OPTIONS]
 
 Options:
     --stack-config STRING    Stack configuration to use (required)
-    --provider STRING        Provider to use (ollama, vllm, etc.) (required)
+    --suite STRING           Test suite to run (default: 'base')
-    --test-subdirs STRING    Comma-separated list of test subdirectories to run (default: 'inference')
+    --setup STRING           Test setup (models, env) to use (e.g., 'ollama', 'ollama-vision', 'gpt', 'vllm')
-    --run-vision-tests       Run vision tests instead of regular tests
     --inference-mode STRING  Inference mode: record or replay (default: replay)
-    --test-pattern STRING    Regex pattern to pass to pytest -k
+    --subdirs STRING         Comma-separated list of test subdirectories to run (overrides suite)
+    --pattern STRING         Regex pattern to pass to pytest -k
     --help                   Show this help message
 
+Suites are defined in tests/integration/suites.py and define which tests to run.
+Setups are defined in tests/integration/setups.py and provide global configuration (models, env).
+
+You can also specify subdirectories (of tests/integration) to select tests from, which will override the suite.
+
 Examples:
     # Basic inference tests with ollama
-    $0 --stack-config server:ci-tests --provider ollama
+    $0 --stack-config server:ci-tests --suite base --setup ollama
 
     # Multiple test directories with vllm
-    $0 --stack-config server:ci-tests --provider vllm --test-subdirs 'inference,agents'
+    $0 --stack-config server:ci-tests --subdirs 'inference,agents' --setup vllm
 
     # Vision tests with ollama
-    $0 --stack-config server:ci-tests --provider ollama --run-vision-tests
+    $0 --stack-config server:ci-tests --suite vision  # default setup for this suite is ollama-vision
 
     # Record mode for updating test recordings
-    $0 --stack-config server:ci-tests --provider ollama --inference-mode record
+    $0 --stack-config server:ci-tests --suite base --inference-mode record
 EOF
 }
 
@@ -56,23 +61,23 @@ while [[ $# -gt 0 ]]; do
             STACK_CONFIG="$2"
             shift 2
             ;;
-        --provider)
+        --setup)
-            PROVIDER="$2"
+            TEST_SETUP="$2"
             shift 2
             ;;
-        --test-subdirs)
+        --subdirs)
             TEST_SUBDIRS="$2"
             shift 2
             ;;
-        --run-vision-tests)
+        --suite)
-            RUN_VISION_TESTS="true"
+            TEST_SUITE="$2"
-            shift
+            shift 2
             ;;
         --inference-mode)
             INFERENCE_MODE="$2"
             shift 2
             ;;
-        --test-pattern)
+        --pattern)
             TEST_PATTERN="$2"
             shift 2
             ;;
@@ -96,18 +101,23 @@ if [[ -z "$STACK_CONFIG" ]]; then
     exit 1
 fi
 
-if [[ -z "$PROVIDER" ]]; then
+if [[ -z "$TEST_SETUP" && -n "$TEST_SUBDIRS" ]]; then
-    echo "Error: --provider is required"
+    echo "Error: --test-setup is required when --test-subdirs is provided"
     usage
     exit 1
 fi
 
+if [[ -z "$TEST_SUITE" && -z "$TEST_SUBDIRS" ]]; then
+    echo "Error: --test-suite or --test-subdirs is required"
+    exit 1
+fi
+
 echo "=== Llama Stack Integration Test Runner ==="
 echo "Stack Config: $STACK_CONFIG"
-echo "Provider: $PROVIDER"
+echo "Setup: $TEST_SETUP"
-echo "Test Subdirs: $TEST_SUBDIRS"
-echo "Vision Tests: $RUN_VISION_TESTS"
 echo "Inference Mode: $INFERENCE_MODE"
+echo "Test Suite: $TEST_SUITE"
+echo "Test Subdirs: $TEST_SUBDIRS"
 echo "Test Pattern: $TEST_PATTERN"
 echo ""
 
@@ -122,31 +132,28 @@ echo ""
 
 # Set environment variables
 export LLAMA_STACK_CLIENT_TIMEOUT=300
-export LLAMA_STACK_TEST_INFERENCE_MODE="$INFERENCE_MODE"
-
-# Configure provider-specific settings
-if [[ "$PROVIDER" == "ollama" ]]; then
-    export OLLAMA_URL="http://0.0.0.0:11434"
-    export TEXT_MODEL="ollama/llama3.2:3b-instruct-fp16"
-    export SAFETY_MODEL="ollama/llama-guard3:1b"
-    EXTRA_PARAMS="--safety-shield=llama-guard"
-else
-    export VLLM_URL="http://localhost:8000/v1"
-    export TEXT_MODEL="vllm/meta-llama/Llama-3.2-1B-Instruct"
-    EXTRA_PARAMS=""
-fi
 
 THIS_DIR=$(dirname "$0")
+
+if [[ -n "$TEST_SETUP" ]]; then
+    EXTRA_PARAMS="--setup=$TEST_SETUP"
+fi
+
+# Apply setup-specific environment variables (needed for server startup and tests)
+echo "=== Applying Setup Environment Variables ==="
+
+# the server needs this
+export LLAMA_STACK_TEST_INFERENCE_MODE="$INFERENCE_MODE"
+
+SETUP_ENV=$(PYTHONPATH=$THIS_DIR/.. python "$THIS_DIR/get_setup_env.py" --suite "$TEST_SUITE" --setup "$TEST_SETUP" --format bash)
+echo "Setting up environment variables:"
+echo "$SETUP_ENV"
+eval "$SETUP_ENV"
+echo ""
+
 ROOT_DIR="$THIS_DIR/.."
 cd $ROOT_DIR
 
-# Set recording directory
-if [[ "$RUN_VISION_TESTS" == "true" ]]; then
-    export LLAMA_STACK_TEST_RECORDING_DIR="tests/integration/recordings/vision"
-else
-    export LLAMA_STACK_TEST_RECORDING_DIR="tests/integration/recordings"
-fi
-
 # check if "llama" and "pytest" are available. this script does not use `uv run` given
 # it can be used in a pre-release environment where we have not been able to tell
 # uv about pre-release dependencies properly (yet).
@@ -162,6 +169,18 @@ fi
 
 # Start Llama Stack Server if needed
 if [[ "$STACK_CONFIG" == *"server:"* ]]; then
+    stop_server() {
+        echo "Stopping Llama Stack Server..."
+        pids=$(lsof -i :8321 | awk 'NR>1 {print $2}')
+        if [[ -n "$pids" ]]; then
+            echo "Killing Llama Stack Server processes: $pids"
+            kill -9 $pids
+        else
+            echo "No Llama Stack Server processes found ?!"
+        fi
+        echo "Llama Stack Server stopped"
+    }
+
     # check if server is already running
     if curl -s http://localhost:8321/v1/health 2>/dev/null | grep -q "OK"; then
         echo "Llama Stack Server is already running, skipping start"
@@ -185,14 +204,16 @@ if [[ "$STACK_CONFIG" == *"server:"* ]]; then
         done
         echo ""
     fi
+
+    trap stop_server EXIT ERR INT TERM
 fi
 
 # Run tests
 echo "=== Running Integration Tests ==="
 EXCLUDE_TESTS="builtin_tool or safety_with_image or code_interpreter or test_rag"
 
-# Additional exclusions for vllm provider
+# Additional exclusions for vllm setup
-if [[ "$PROVIDER" == "vllm" ]]; then
+if [[ "$TEST_SETUP" == "vllm" ]]; then
     EXCLUDE_TESTS="${EXCLUDE_TESTS} or test_inference_store_tool_calls"
 fi
 
@@ -201,56 +222,12 @@ if [[ -n "$TEST_PATTERN" ]]; then
     PYTEST_PATTERN="${PYTEST_PATTERN} and $TEST_PATTERN"
 fi
 
-# Run vision tests if specified
-if [[ "$RUN_VISION_TESTS" == "true" ]]; then
-    echo "Running vision tests..."
-    set +e
-    pytest -s -v tests/integration/inference/test_vision_inference.py \
-        --stack-config="$STACK_CONFIG" \
-        -k "$PYTEST_PATTERN" \
-        --vision-model=ollama/llama3.2-vision:11b \
-        --embedding-model=sentence-transformers/all-MiniLM-L6-v2 \
-        --color=yes $EXTRA_PARAMS \
-        --capture=tee-sys
-    exit_code=$?
-    set -e
-
-    if [ $exit_code -eq 0 ]; then
-        echo "✅ Vision tests completed successfully"
-    elif [ $exit_code -eq 5 ]; then
-        echo "⚠️ No vision tests collected (pattern matched no tests)"
-    else
-        echo "❌ Vision tests failed"
-        exit 1
-    fi
-    exit 0
-fi
-
-# Run regular tests
-if [[ -z "$TEST_SUBDIRS" ]]; then
-   TEST_SUBDIRS=$(find tests/integration -maxdepth 1 -mindepth 1 -type d |
-            sed 's|tests/integration/||' |
-            grep -Ev "^(__pycache__|fixtures|test_cases|recordings|non_ci|post_training)$" |
-            sort)
-fi
 echo "Test subdirs to run: $TEST_SUBDIRS"
 
+if [[ -n "$TEST_SUBDIRS" ]]; then
     # Collect all test files for the specified test types
     TEST_FILES=""
     for test_subdir in $(echo "$TEST_SUBDIRS" | tr ',' '\n'); do
-    # Skip certain test types for vllm provider
-    if [[ "$PROVIDER" == "vllm" ]]; then
-        if [[ "$test_subdir" == "safety" ]] || [[ "$test_subdir" == "post_training" ]] || [[ "$test_subdir" == "tool_runtime" ]]; then
-            echo "Skipping $test_subdir for vllm provider"
-            continue
-        fi
-    fi
-
-    if [[ "$STACK_CONFIG" != *"server:"* ]] && [[ "$test_subdir" == "batches" ]]; then
-        echo "Skipping $test_subdir for library client until types are supported"
-        continue
-    fi
-
         if [[ -d "tests/integration/$test_subdir" ]]; then
             # Find all Python test files in this directory
             test_files=$(find tests/integration/$test_subdir -name "test_*.py" -o -name "*_test.py")
@@ -272,15 +249,23 @@ echo ""
     echo "=== Running all collected tests in a single pytest command ==="
     echo "Total test files: $(echo $TEST_FILES | wc -w)"
 
+    PYTEST_TARGET="$TEST_FILES"
+else
+    PYTEST_TARGET="tests/integration/"
+    EXTRA_PARAMS="$EXTRA_PARAMS --suite=$TEST_SUITE"
+fi
+
 set +e
-pytest -s -v $TEST_FILES \
+set -x
+pytest -s -v $PYTEST_TARGET \
     --stack-config="$STACK_CONFIG" \
+    --inference-mode="$INFERENCE_MODE" \
     -k "$PYTEST_PATTERN" \
-    --text-model="$TEXT_MODEL" \
+    $EXTRA_PARAMS \
-    --embedding-model=sentence-transformers/all-MiniLM-L6-v2 \
+    --color=yes \
-    --color=yes $EXTRA_PARAMS \
     --capture=tee-sys
 exit_code=$?
+set +x
 set -e
 
 if [ $exit_code -eq 0 ]; then

tests/README.md

@@ -38,26 +38,15 @@ For running integration tests, you must provide a few things:
   - a distribution name (e.g., `starter`) or a path to a `run.yaml` file
   - a comma-separated list of api=provider pairs, e.g. `inference=fireworks,safety=llama-guard,agents=meta-reference`. This is most useful for testing a single API surface.
 
-- Whether you are using replay or live mode for inference. This is specified with the LLAMA_STACK_TEST_INFERENCE_MODE environment variable. The default mode currently is "live" -- that is certainly surprising, but we will fix this soon.
-
 - Any API keys you need to use should be set in the environment, or can be passed in with the --env option.
 
 You can run the integration tests in replay mode with:
 ```bash
 # Run all tests with existing recordings
-LLAMA_STACK_TEST_INFERENCE_MODE=replay \
-  LLAMA_STACK_TEST_RECORDING_DIR=tests/integration/recordings \
   uv run --group test \
   pytest -sv tests/integration/ --stack-config=starter
 ```
 
-If you don't specify LLAMA_STACK_TEST_INFERENCE_MODE, by default it will be in "live" mode -- that is, it will make real API calls.
-
-```bash
-# Test against live APIs
-FIREWORKS_API_KEY=your_key pytest -sv tests/integration/inference --stack-config=starter
-```
-
 ### Re-recording tests
 
 #### Local Re-recording (Manual Setup Required)
@@ -66,7 +55,6 @@ If you want to re-record tests locally, you can do so with:
 
 ```bash
 LLAMA_STACK_TEST_INFERENCE_MODE=record \
-  LLAMA_STACK_TEST_RECORDING_DIR=tests/integration/recordings \
   uv run --group test \
   pytest -sv tests/integration/ --stack-config=starter -k "<appropriate test name>"
 ```
@@ -89,7 +77,7 @@ You must be careful when re-recording. CI workflows assume a specific setup for
 ./scripts/github/schedule-record-workflow.sh --test-subdirs "agents,inference"
 
 # Record with vision tests enabled
-./scripts/github/schedule-record-workflow.sh --test-subdirs "inference" --run-vision-tests
+./scripts/github/schedule-record-workflow.sh --test-suite vision
 
 # Record with specific provider
 ./scripts/github/schedule-record-workflow.sh --test-subdirs "agents" --test-provider vllm

tests/integration/README.md

@@ -6,8 +6,6 @@ Integration tests verify complete workflows across different providers using Lla
 
 ```bash
 # Run all integration tests with existing recordings
-LLAMA_STACK_TEST_INFERENCE_MODE=replay \
-  LLAMA_STACK_TEST_RECORDING_DIR=tests/integration/recordings \
 uv run --group test \
   pytest -sv tests/integration/ --stack-config=starter
 ```
@@ -42,6 +40,37 @@ Model parameters can be influenced by the following options:
 Each of these are comma-separated lists and can be used to generate multiple parameter combinations. Note that tests will be skipped
 if no model is specified.
 
+### Suites and Setups
+
+- `--suite`: single named suite that narrows which tests are collected.
+- Available suites:
+  - `base`: collects most tests (excludes responses and post_training)
+  - `responses`: collects tests under `tests/integration/responses` (needs strong tool-calling models)
+  - `vision`: collects only `tests/integration/inference/test_vision_inference.py`
+- `--setup`: global configuration that can be used with any suite. Setups prefill model/env defaults; explicit CLI flags always win.
+  - Available setups:
+    - `ollama`: Local Ollama provider with lightweight models (sets OLLAMA_URL, uses llama3.2:3b-instruct-fp16)
+    - `vllm`: VLLM provider for efficient local inference (sets VLLM_URL, uses Llama-3.2-1B-Instruct)
+    - `gpt`: OpenAI GPT models for high-quality responses (uses gpt-4o)
+    - `claude`: Anthropic Claude models for high-quality responses (uses claude-3-5-sonnet)
+
+Examples
+
+```bash
+# Fast responses run with a strong tool-calling model
+pytest -s -v tests/integration --stack-config=server:starter --suite=responses --setup=gpt
+
+# Fast single-file vision run with Ollama defaults
+pytest -s -v tests/integration --stack-config=server:starter --suite=vision --setup=ollama
+
+# Base suite with VLLM for performance
+pytest -s -v tests/integration --stack-config=server:starter --suite=base --setup=vllm
+
+# Override a default from setup
+pytest -s -v tests/integration --stack-config=server:starter \
+  --suite=responses --setup=gpt --embedding-model=text-embedding-3-small
+```
+
 ## Examples
 
 ### Testing against a Server
@@ -98,29 +127,24 @@ pytest -s -v tests/integration/vector_io/ \
 
 The testing system supports three modes controlled by environment variables:
 
-### LIVE Mode (Default)
+### REPLAY Mode (Default)
-Tests make real API calls:
+Uses cached responses instead of making API calls:
 ```bash
-LLAMA_STACK_TEST_INFERENCE_MODE=live pytest tests/integration/
+pytest tests/integration/
 ```
-
 ### RECORD Mode
 Captures API interactions for later replay:
 ```bash
-LLAMA_STACK_TEST_INFERENCE_MODE=record \
+pytest tests/integration/inference/test_new_feature.py --inference-mode=record
-LLAMA_STACK_TEST_RECORDING_DIR=tests/integration/recordings \
-pytest tests/integration/inference/test_new_feature.py
 ```
 
-### REPLAY Mode
+### LIVE Mode
-Uses cached responses instead of making API calls:
+Tests make real API calls (but not recorded):
 ```bash
-LLAMA_STACK_TEST_INFERENCE_MODE=replay \
+pytest tests/integration/ --inference-mode=live
-LLAMA_STACK_TEST_RECORDING_DIR=tests/integration/recordings \
-pytest tests/integration/
 ```
 
-Note that right now you must specify the recording directory. This is because different tests use different recording directories and we don't (yet) have a fool-proof way to map a test to a recording directory. We are working on this.
+By default, the recording directory is `tests/integration/recordings`. You can override this by setting the `LLAMA_STACK_TEST_RECORDING_DIR` environment variable.
 
 ## Managing Recordings
 
@@ -138,16 +162,14 @@ cat recordings/responses/abc123.json | jq '.'
 #### Remote Re-recording (Recommended)
 Use the automated workflow script for easier re-recording:
 ```bash
-./scripts/github/schedule-record-workflow.sh --test-subdirs "inference,agents"
+./scripts/github/schedule-record-workflow.sh --subdirs "inference,agents"
 ```
 See the [main testing guide](../README.md#remote-re-recording-recommended) for full details.
 
 #### Local Re-recording
 ```bash
 # Re-record specific tests
-LLAMA_STACK_TEST_INFERENCE_MODE=record \
+pytest -s -v --stack-config=server:starter tests/integration/inference/test_modified.py --inference-mode=record
-LLAMA_STACK_TEST_RECORDING_DIR=tests/integration/recordings \
-pytest -s -v --stack-config=server:starter tests/integration/inference/test_modified.py
 ```
 
 Note that when re-recording tests, you must use a Stack pointing to a server (i.e., `server:starter`). This subtlety exists because the set of tests run in server are a superset of the set of tests run in the library client.

tests/integration/batches/test_batches.py

@@ -268,3 +268,58 @@ class TestBatchesIntegration:
 
         deleted_error_file = openai_client.files.delete(final_batch.error_file_id)
         assert deleted_error_file.deleted, f"Error file {final_batch.error_file_id} was not deleted successfully"
+
+    def test_batch_e2e_completions(self, openai_client, batch_helper, text_model_id):
+        """Run an end-to-end batch with a single successful text completion request."""
+        request_body = {"model": text_model_id, "prompt": "Say completions", "max_tokens": 20}
+
+        batch_requests = [
+            {
+                "custom_id": "success-1",
+                "method": "POST",
+                "url": "/v1/completions",
+                "body": request_body,
+            }
+        ]
+
+        with batch_helper.create_file(batch_requests) as uploaded_file:
+            batch = openai_client.batches.create(
+                input_file_id=uploaded_file.id,
+                endpoint="/v1/completions",
+                completion_window="24h",
+                metadata={"test": "e2e_completions_success"},
+            )
+
+            final_batch = batch_helper.wait_for(
+                batch.id,
+                max_wait_time=3 * 60,
+                expected_statuses={"completed"},
+                timeout_action="skip",
+            )
+
+        assert final_batch.status == "completed"
+        assert final_batch.request_counts is not None
+        assert final_batch.request_counts.total == 1
+        assert final_batch.request_counts.completed == 1
+        assert final_batch.output_file_id is not None
+
+        output_content = openai_client.files.content(final_batch.output_file_id)
+        if isinstance(output_content, str):
+            output_text = output_content
+        else:
+            output_text = output_content.content.decode("utf-8")
+
+        output_lines = output_text.strip().split("\n")
+        assert len(output_lines) == 1
+
+        result = json.loads(output_lines[0])
+        assert result["custom_id"] == "success-1"
+        assert "response" in result
+        assert result["response"]["status_code"] == 200
+
+        deleted_output_file = openai_client.files.delete(final_batch.output_file_id)
+        assert deleted_output_file.deleted
+
+        if final_batch.error_file_id is not None:
+            deleted_error_file = openai_client.files.delete(final_batch.error_file_id)
+            assert deleted_error_file.deleted

tests/integration/conftest.py

@@ -6,15 +6,17 @@
 import inspect
 import itertools
 import os
-import platform
 import textwrap
 import time
+from pathlib import Path
 
 import pytest
 from dotenv import load_dotenv
 
 from llama_stack.log import get_logger
 
+from .suites import SETUP_DEFINITIONS, SUITE_DEFINITIONS
+
 logger = get_logger(__name__, category="tests")
 
 
@@ -30,6 +32,8 @@ def pytest_runtest_makereport(item, call):
 def pytest_sessionstart(session):
     # stop macOS from complaining about duplicate OpenMP libraries
     os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"
+    if "LLAMA_STACK_TEST_INFERENCE_MODE" not in os.environ:
+        os.environ["LLAMA_STACK_TEST_INFERENCE_MODE"] = "replay"
 
 
 def pytest_runtest_teardown(item):
@@ -59,9 +63,36 @@ def pytest_configure(config):
         key, value = env_var.split("=", 1)
         os.environ[key] = value
 
-    if platform.system() == "Darwin":  # Darwin is the system name for macOS
+    inference_mode = config.getoption("--inference-mode")
-        os.environ["DISABLE_CODE_SANDBOX"] = "1"
+    os.environ["LLAMA_STACK_TEST_INFERENCE_MODE"] = inference_mode
-        logger.info("Setting DISABLE_CODE_SANDBOX=1 for macOS")
+
+    suite = config.getoption("--suite")
+    if suite:
+        if suite not in SUITE_DEFINITIONS:
+            raise pytest.UsageError(f"Unknown suite: {suite}. Available: {', '.join(sorted(SUITE_DEFINITIONS.keys()))}")
+
+    # Apply setups (global parameterizations): env + defaults
+    setup = config.getoption("--setup")
+    if suite and not setup:
+        setup = SUITE_DEFINITIONS[suite].default_setup
+
+    if setup:
+        if setup not in SETUP_DEFINITIONS:
+            raise pytest.UsageError(
+                f"Unknown setup '{setup}'. Available: {', '.join(sorted(SETUP_DEFINITIONS.keys()))}"
+            )
+
+        setup_obj = SETUP_DEFINITIONS[setup]
+        logger.info(f"Applying setup '{setup}'{' for suite ' + suite if suite else ''}")
+        # Apply env first
+        for k, v in setup_obj.env.items():
+            if k not in os.environ:
+                os.environ[k] = str(v)
+        # Apply defaults if not provided explicitly
+        for dest, value in setup_obj.defaults.items():
+            current = getattr(config.option, dest, None)
+            if not current:
+                setattr(config.option, dest, value)
 
 
 def pytest_addoption(parser):
@@ -103,16 +134,32 @@ def pytest_addoption(parser):
         default=384,
         help="Output dimensionality of the embedding model to use for testing. Default: 384",
     )
+
     parser.addoption(
-        "--record-responses",
+        "--inference-mode",
-        action="store_true",
+        help="Inference mode: { record, replay, live } (default: replay)",
-        help="Record new API responses instead of using cached ones.",
+        choices=["record", "replay", "live"],
+        default="replay",
     )
     parser.addoption(
         "--report",
         help="Path where the test report should be written, e.g. --report=/path/to/report.md",
     )
 
+    available_suites = ", ".join(sorted(SUITE_DEFINITIONS.keys()))
+    suite_help = (
+        f"Single test suite to run (narrows collection). Available: {available_suites}. Example: --suite=responses"
+    )
+    parser.addoption("--suite", help=suite_help)
+
+    # Global setups for any suite
+    available_setups = ", ".join(sorted(SETUP_DEFINITIONS.keys()))
+    setup_help = (
+        f"Global test setup configuration. Available: {available_setups}. "
+        "Can be used with any suite. Example: --setup=ollama"
+    )
+    parser.addoption("--setup", help=setup_help)
+
 
 MODEL_SHORT_IDS = {
     "meta-llama/Llama-3.2-3B-Instruct": "3B",
@@ -195,3 +242,36 @@ def pytest_generate_tests(metafunc):
 
 
 pytest_plugins = ["tests.integration.fixtures.common"]
+
+
+def pytest_ignore_collect(path: str, config: pytest.Config) -> bool:
+    """Skip collecting paths outside the selected suite roots for speed."""
+    suite = config.getoption("--suite")
+    if not suite:
+        return False
+
+    sobj = SUITE_DEFINITIONS.get(suite)
+    roots: list[str] = sobj.get("roots", []) if isinstance(sobj, dict) else getattr(sobj, "roots", [])
+    if not roots:
+        return False
+
+    p = Path(str(path)).resolve()
+
+    # Only constrain within tests/integration to avoid ignoring unrelated tests
+    integration_root = (Path(str(config.rootpath)) / "tests" / "integration").resolve()
+    if not p.is_relative_to(integration_root):
+        return False
+
+    for r in roots:
+        rp = (Path(str(config.rootpath)) / r).resolve()
+        if rp.is_file():
+            # Allow the exact file and any ancestor directories so pytest can walk into it.
+            if p == rp:
+                return False
+            if p.is_dir() and rp.is_relative_to(p):
+                return False
+        else:
+            # Allow anything inside an allowed directory
+            if p.is_relative_to(rp):
+                return False
+    return True

tests/integration/inference/test_openai_completion.py

@@ -5,6 +5,8 @@
 # the root directory of this source tree.
 
 
+import time
+
 import pytest
 
 from ..test_cases.test_case import TestCase
@@ -35,6 +37,11 @@ def skip_if_model_doesnt_support_openai_completion(client_with_models, model_id)
         "remote::sambanova",
         "remote::tgi",
         "remote::vertexai",
+        # {"error":{"message":"Unknown request URL: GET /openai/v1/completions. Please check the URL for typos,
+        # or see the docs at https://console.groq.com/docs/","type":"invalid_request_error","code":"unknown_url"}}
+        "remote::groq",
+        "remote::gemini",  # https://generativelanguage.googleapis.com/v1beta/openai/completions -> 404
+        "remote::anthropic",  # at least claude-3-{5,7}-{haiku,sonnet}-* / claude-{sonnet,opus}-4-* are not supported
     ):
         pytest.skip(f"Model {model_id} hosted by {provider.provider_type} doesn't support OpenAI completions.")
 
@@ -56,6 +63,26 @@ def skip_if_model_doesnt_support_suffix(client_with_models, model_id):
         pytest.skip(f"Provider {provider.provider_type} doesn't support suffix.")
 
 
+def skip_if_doesnt_support_n(client_with_models, model_id):
+    provider = provider_from_model(client_with_models, model_id)
+    if provider.provider_type in (
+        "remote::sambanova",
+        "remote::ollama",
+        # https://console.groq.com/docs/openai#currently-unsupported-openai-features
+        # -> Error code: 400 - {'error': {'message': "'n' : number must be at most 1", 'type': 'invalid_request_error'}}
+        "remote::groq",
+        # Error code: 400 - [{'error': {'code': 400, 'message': 'Only one candidate can be specified in the
+        # current model', 'status': 'INVALID_ARGUMENT'}}]
+        "remote::gemini",
+        # https://docs.anthropic.com/en/api/openai-sdk#simple-fields
+        "remote::anthropic",
+        "remote::vertexai",
+        #  Error code: 400 - [{'error': {'code': 400, 'message': 'Unable to submit request because candidateCount must be 1 but
+        #  the entered value was 2. Update the candidateCount value and try again.', 'status': 'INVALID_ARGUMENT'}
+    ):
+        pytest.skip(f"Model {model_id} hosted by {provider.provider_type} doesn't support n param.")
+
+
 def skip_if_model_doesnt_support_openai_chat_completion(client_with_models, model_id):
     provider = provider_from_model(client_with_models, model_id)
     if provider.provider_type in (
@@ -260,10 +287,7 @@ def test_openai_chat_completion_streaming(compat_client, client_with_models, tex
 )
 def test_openai_chat_completion_streaming_with_n(compat_client, client_with_models, text_model_id, test_case):
     skip_if_model_doesnt_support_openai_chat_completion(client_with_models, text_model_id)
-
+    skip_if_doesnt_support_n(client_with_models, text_model_id)
-    provider = provider_from_model(client_with_models, text_model_id)
-    if provider.provider_type == "remote::ollama":
-        pytest.skip(f"Model {text_model_id} hosted by {provider.provider_type} doesn't support n > 1.")
 
     tc = TestCase(test_case)
     question = tc["question"]
@@ -323,8 +347,15 @@ def test_inference_store(compat_client, client_with_models, text_model_id, strea
         response_id = response.id
         content = response.choices[0].message.content
 
+    tries = 0
+    while tries < 10:
         responses = client.chat.completions.list(limit=1000)
-    assert response_id in [r.id for r in responses.data]
+        if response_id in [r.id for r in responses.data]:
+            break
+        else:
+            tries += 1
+            time.sleep(0.1)
+    assert tries < 10, f"Response {response_id} not found after 1 second"
 
     retrieved_response = client.chat.completions.retrieve(response_id)
     assert retrieved_response.id == response_id
@@ -388,6 +419,18 @@ def test_inference_store_tool_calls(compat_client, client_with_models, text_mode
         response_id = response.id
         content = response.choices[0].message.content
 
+    # wait for the response to be stored
+    tries = 0
+    while tries < 10:
+        responses = client.chat.completions.list(limit=1000)
+        if response_id in [r.id for r in responses.data]:
+            break
+        else:
+            tries += 1
+            time.sleep(0.1)
+
+    assert tries < 10, f"Response {response_id} not found after 1 second"
+
     responses = client.chat.completions.list(limit=1000)
     assert response_id in [r.id for r in responses.data]
 

tests/integration/recordings/responses/00ba04f74a96.json

@@ -20,15 +20,15 @@
       "__type__": "ollama._types.GenerateResponse",
       "__data__": {
         "model": "llama-guard3:1b",
-        "created_at": "2025-08-01T23:12:53.860911Z",
+        "created_at": "2025-09-03T17:37:35.23084Z",
         "done": true,
         "done_reason": "stop",
-        "total_duration": 249137667,
+        "total_duration": 195981375,
-        "load_duration": 152509542,
+        "load_duration": 110522917,
         "prompt_eval_count": 216,
-        "prompt_eval_duration": 71000000,
+        "prompt_eval_duration": 72393958,
         "eval_count": 2,
-        "eval_duration": 24000000,
+        "eval_duration": 11843000,
         "response": "safe",
         "thinking": null,
         "context": null

tests/integration/recordings/responses/04172112ffbb.json

@@ -21,7 +21,7 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:18.033900164Z",
+          "created_at": "2025-09-03T17:41:43.950283Z",
           "done": false,
           "done_reason": null,
           "total_duration": null,
@@ -39,7 +39,7 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:18.213371151Z",
+          "created_at": "2025-09-03T17:41:43.991122Z",
           "done": false,
           "done_reason": null,
           "total_duration": null,
@@ -57,7 +57,7 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:18.387513976Z",
+          "created_at": "2025-09-03T17:41:44.031378Z",
           "done": false,
           "done_reason": null,
           "total_duration": null,
@@ -75,7 +75,7 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:18.564344287Z",
+          "created_at": "2025-09-03T17:41:44.073098Z",
           "done": false,
           "done_reason": null,
           "total_duration": null,
@@ -93,7 +93,7 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:18.746579415Z",
+          "created_at": "2025-09-03T17:41:44.115961Z",
           "done": false,
           "done_reason": null,
           "total_duration": null,
@@ -111,7 +111,7 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:18.923276047Z",
+          "created_at": "2025-09-03T17:41:44.156517Z",
           "done": false,
           "done_reason": null,
           "total_duration": null,
@@ -129,7 +129,7 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:19.099961963Z",
+          "created_at": "2025-09-03T17:41:44.197079Z",
           "done": false,
           "done_reason": null,
           "total_duration": null,
@@ -147,7 +147,7 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:19.275621884Z",
+          "created_at": "2025-09-03T17:41:44.237565Z",
           "done": false,
           "done_reason": null,
           "total_duration": null,
@@ -165,7 +165,7 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:19.452204196Z",
+          "created_at": "2025-09-03T17:41:44.277755Z",
           "done": false,
           "done_reason": null,
           "total_duration": null,
@@ -183,7 +183,7 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:19.626937514Z",
+          "created_at": "2025-09-03T17:41:44.318476Z",
           "done": false,
           "done_reason": null,
           "total_duration": null,
@@ -201,7 +201,7 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:19.805566767Z",
+          "created_at": "2025-09-03T17:41:44.358628Z",
           "done": false,
           "done_reason": null,
           "total_duration": null,
@@ -219,7 +219,7 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:19.985987477Z",
+          "created_at": "2025-09-03T17:41:44.398984Z",
           "done": false,
           "done_reason": null,
           "total_duration": null,
@@ -237,7 +237,7 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:20.166458601Z",
+          "created_at": "2025-09-03T17:41:44.439232Z",
           "done": false,
           "done_reason": null,
           "total_duration": null,
@@ -255,7 +255,7 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:20.343346795Z",
+          "created_at": "2025-09-03T17:41:44.479478Z",
           "done": false,
           "done_reason": null,
           "total_duration": null,
@@ -273,7 +273,7 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:20.525008091Z",
+          "created_at": "2025-09-03T17:41:44.520202Z",
           "done": false,
           "done_reason": null,
           "total_duration": null,
@@ -291,7 +291,7 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:20.709087695Z",
+          "created_at": "2025-09-03T17:41:44.560517Z",
           "done": false,
           "done_reason": null,
           "total_duration": null,
@@ -309,7 +309,7 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:20.887074305Z",
+          "created_at": "2025-09-03T17:41:44.601592Z",
           "done": false,
           "done_reason": null,
           "total_duration": null,
@@ -327,15 +327,15 @@
         "__type__": "ollama._types.GenerateResponse",
         "__data__": {
           "model": "llama3.2:3b-instruct-fp16",
-          "created_at": "2025-07-31T17:59:21.065244925Z",
+          "created_at": "2025-09-03T17:41:44.642064Z",
           "done": true,
           "done_reason": "stop",
-          "total_duration": 4373531496,
+          "total_duration": 887142667,
-          "load_duration": 44438132,
+          "load_duration": 119331417,
           "prompt_eval_count": 56,
-          "prompt_eval_duration": 1296273199,
+          "prompt_eval_duration": 74294709,
           "eval_count": 18,
-          "eval_duration": 3032321735,
+          "eval_duration": 692842791,
           "response": "",
           "thinking": null,
           "context": null

tests/integration/recordings/responses/0b27fd737699.json

@@ -20,15 +20,15 @@
       "__type__": "ollama._types.GenerateResponse",
       "__data__": {
         "model": "llama-guard3:1b",
-        "created_at": "2025-08-01T23:13:57.556416Z",
+        "created_at": "2025-09-03T17:37:47.461886Z",
         "done": true,
         "done_reason": "stop",
-        "total_duration": 432363250,
+        "total_duration": 338927833,
-        "load_duration": 159296417,
+        "load_duration": 100895125,
         "prompt_eval_count": 223,
-        "prompt_eval_duration": 257000000,
+        "prompt_eval_duration": 221583042,
         "eval_count": 2,
-        "eval_duration": 14000000,
+        "eval_duration": 12341416,
         "response": "safe",
         "thinking": null,
         "context": null

tests/integration/recordings/responses/0b3f2e4754ff.json

@@ -24,7 +24,7 @@
       {
         "__type__": "openai.types.chat.chat_completion_chunk.ChatCompletionChunk",
         "__data__": {
-          "id": "chatcmpl-29",
+          "id": "chatcmpl-414",
           "choices": [
             {
               "delta": {
@@ -39,7 +39,7 @@
               "logprobs": null
             }
           ],
-          "created": 1754090031,
+          "created": 1756921333,
           "model": "llama3.2:3b-instruct-fp16",
           "object": "chat.completion.chunk",
           "service_tier": null,
@@ -50,7 +50,7 @@
       {
         "__type__": "openai.types.chat.chat_completion_chunk.ChatCompletionChunk",
         "__data__": {
-          "id": "chatcmpl-29",
+          "id": "chatcmpl-414",
           "choices": [
             {
               "delta": {
@@ -65,7 +65,7 @@
               "logprobs": null
             }
           ],
-          "created": 1754090031,
+          "created": 1756921333,
           "model": "llama3.2:3b-instruct-fp16",
           "object": "chat.completion.chunk",
           "service_tier": null,
@@ -76,7 +76,7 @@
       {
         "__type__": "openai.types.chat.chat_completion_chunk.ChatCompletionChunk",
         "__data__": {
-          "id": "chatcmpl-29",
+          "id": "chatcmpl-414",
           "choices": [
             {
               "delta": {
@@ -91,7 +91,7 @@
               "logprobs": null
             }
           ],
-          "created": 1754090031,
+          "created": 1756921333,
           "model": "llama3.2:3b-instruct-fp16",
           "object": "chat.completion.chunk",
           "service_tier": null,
@@ -102,7 +102,7 @@
       {
         "__type__": "openai.types.chat.chat_completion_chunk.ChatCompletionChunk",
         "__data__": {
-          "id": "chatcmpl-29",
+          "id": "chatcmpl-414",
           "choices": [
             {
               "delta": {
@@ -117,7 +117,7 @@
               "logprobs": null
             }
           ],
-          "created": 1754090031,
+          "created": 1756921333,
           "model": "llama3.2:3b-instruct-fp16",
           "object": "chat.completion.chunk",
           "service_tier": null,
@@ -128,7 +128,7 @@
       {
         "__type__": "openai.types.chat.chat_completion_chunk.ChatCompletionChunk",
         "__data__": {
-          "id": "chatcmpl-29",
+          "id": "chatcmpl-414",
           "choices": [
             {
               "delta": {
@@ -143,7 +143,7 @@
               "logprobs": null
             }
           ],
-          "created": 1754090031,
+          "created": 1756921334,
           "model": "llama3.2:3b-instruct-fp16",
           "object": "chat.completion.chunk",
           "service_tier": null,
@@ -154,7 +154,7 @@
       {
         "__type__": "openai.types.chat.chat_completion_chunk.ChatCompletionChunk",
         "__data__": {
-          "id": "chatcmpl-29",
+          "id": "chatcmpl-414",
           "choices": [
             {
               "delta": {
@@ -169,7 +169,7 @@
               "logprobs": null
             }
           ],
-          "created": 1754090031,
+          "created": 1756921334,
           "model": "llama3.2:3b-instruct-fp16",
           "object": "chat.completion.chunk",
           "service_tier": null,
@@ -180,7 +180,7 @@
       {
         "__type__": "openai.types.chat.chat_completion_chunk.ChatCompletionChunk",
         "__data__": {
-          "id": "chatcmpl-29",
+          "id": "chatcmpl-414",
           "choices": [
             {
               "delta": {
@@ -195,7 +195,7 @@
               "logprobs": null
             }
           ],
-          "created": 1754090031,
+          "created": 1756921334,
           "model": "llama3.2:3b-instruct-fp16",
           "object": "chat.completion.chunk",
           "service_tier": null,
@@ -206,7 +206,7 @@
       {
         "__type__": "openai.types.chat.chat_completion_chunk.ChatCompletionChunk",
         "__data__": {
-          "id": "chatcmpl-29",
+          "id": "chatcmpl-414",
           "choices": [
             {
               "delta": {
@@ -221,7 +221,7 @@
               "logprobs": null
             }
           ],
-          "created": 1754090031,
+          "created": 1756921334,
           "model": "llama3.2:3b-instruct-fp16",
           "object": "chat.completion.chunk",
           "service_tier": null,

tests/integration/recordings/responses/0e8f2b001dd9.json

@@ -1,7 +1,7 @@
 {
   "request": {
     "method": "POST",
-    "url": "http://localhost:11434/v1/v1/chat/completions",
+    "url": "http://0.0.0.0:11434/v1/v1/chat/completions",
     "headers": {},
     "body": {
       "model": "llama3.2:3b-instruct-fp16",
@@ -20,14 +20,14 @@
     "body": {
       "__type__": "openai.types.chat.chat_completion.ChatCompletion",
       "__data__": {
-        "id": "chatcmpl-368",
+        "id": "chatcmpl-161",
         "choices": [
           {
             "finish_reason": "stop",
             "index": 0,
             "logprobs": null,
             "message": {
-              "content": "Saturn is known for its extensive ring system.",
+              "content": "The answer is Saturn.",
               "refusal": null,
               "role": "assistant",
               "annotations": null,
@@ -37,15 +37,15 @@
             }
           }
         ],
-        "created": 1754081853,
+        "created": 1756921364,
         "model": "llama3.2:3b-instruct-fp16",
         "object": "chat.completion",
         "service_tier": null,
         "system_fingerprint": "fp_ollama",
         "usage": {
-          "completion_tokens": 11,
+          "completion_tokens": 6,
           "prompt_tokens": 39,
-          "total_tokens": 50,
+          "total_tokens": 45,
           "completion_tokens_details": null,
           "prompt_tokens_details": null
         }

tests/integration/recordings/responses/10eea8c15ddc.json

@@ -20,15 +20,15 @@
       "__type__": "ollama._types.GenerateResponse",
       "__data__": {
         "model": "llama-guard3:1b",
-        "created_at": "2025-08-01T23:12:51.682357Z",
+        "created_at": "2025-09-03T17:37:33.473237Z",
         "done": true,
         "done_reason": "stop",
-        "total_duration": 238161000,
+        "total_duration": 279025042,
-        "load_duration": 72494750,
+        "load_duration": 162673250,
         "prompt_eval_count": 212,
-        "prompt_eval_duration": 87000000,
+        "prompt_eval_duration": 73595834,
         "eval_count": 5,
-        "eval_duration": 74000000,
+        "eval_duration": 41950291,
         "response": "unsafe\nS8",
         "thinking": null,
         "context": null