Merge branch 'main' into allow-dynamic-models-nvidia

.github/CODEOWNERS

@@ -2,4 +2,4 @@
 
 # These owners will be the default owners for everything in
 # the repo. Unless a later match takes precedence,
-* @ashwinb @yanxi0830 @hardikjshah @raghotham @ehhuang @terrytangyuan @leseb @bbrowning @reluctantfuturist
+* @ashwinb @yanxi0830 @hardikjshah @raghotham @ehhuang @terrytangyuan @leseb @bbrowning @reluctantfuturist @mattf

.github/ISSUE_TEMPLATE/tech-debt.yml

@@ -0,0 +1,30 @@
+name: 🔧 Tech Debt
+description: Something that is functional but should be improved or optimizied
+labels: ["tech-debt"]
+body:
+- type: textarea
+  id: tech-debt-explanation
+  attributes:
+    label: 🤔 What is the technical debt you think should be addressed?
+    description: >
+      A clear and concise description of _what_ needs to be addressed - ensure you are describing
+      constitutes [technical debt](https://en.wikipedia.org/wiki/Technical_debt) and is not a bug
+      or feature request.
+  validations:
+    required: true
+
+- type: textarea
+  id: tech-debt-motivation
+  attributes:
+    label: 💡 What is the benefit of addressing this technical debt?
+    description: >
+      A clear and concise description of _why_ this work is needed.
+  validations:
+    required: true
+
+- type: textarea
+  id: other-thoughts
+  attributes:
+    label: Other thoughts
+    description: >
+      Any thoughts about how this may result in complexity in the codebase, or other trade-offs.

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

@@ -7,3 +7,7 @@ runs:
       shell: bash
       run: |
         docker run -d --name ollama -p 11434:11434 docker.io/leseb/ollama-with-models
+        # TODO: rebuild an ollama image with llama-guard3:1b
+        echo "Verifying Ollama status..."
+        timeout 30 bash -c 'while ! curl -s -L http://127.0.0.1:11434; do sleep 1 && echo "."; done'
+        docker exec ollama ollama pull llama-guard3:1b

.github/workflows/install-script-ci.yml

@@ -3,10 +3,10 @@ name: Installer CI
 on:
   pull_request:
     paths:
-      - 'install.sh'
+      - 'scripts/install.sh'
   push:
     paths:
-      - 'install.sh'
+      - 'scripts/install.sh'
   schedule:
     - cron: '0 2 * * *'  # every day at 02:00 UTC
 
@@ -16,11 +16,11 @@ jobs:
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # 4.2.2
       - name: Run ShellCheck on install.sh
-        run: shellcheck install.sh
+        run: shellcheck scripts/install.sh
   smoke-test:
     needs: lint
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # 4.2.2
       - name: Run installer end-to-end
-        run: ./install.sh
+        run: ./scripts/install.sh

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

@@ -35,7 +35,7 @@ jobs:
 
       - name: Install minikube
         if: ${{ matrix.auth-provider == 'kubernetes' }}
-        uses: medyagh/setup-minikube@cea33675329b799adccc9526aa5daccc26cd5052 # v0.0.19
+        uses: medyagh/setup-minikube@e3c7f79eb1e997eabccc536a6cf318a2b0fe19d9 # v0.0.20
 
       - name: Start minikube
         if: ${{ matrix.auth-provider == 'oauth2_token' }}

.github/workflows/integration-tests.yml

@@ -18,16 +18,33 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
-  test-matrix:
+  discover-tests:
     runs-on: ubuntu-latest
+    outputs:
+      test-type: ${{ steps.generate-matrix.outputs.test-type }}
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+
+      - name: Generate test matrix
+        id: generate-matrix
+        run: |
+          # Get test directories dynamically, excluding non-test directories
+          TEST_TYPES=$(find tests/integration -maxdepth 1 -mindepth 1 -type d -printf "%f\n" |
+            grep -Ev "^(__pycache__|fixtures|test_cases)$" |
+            sort | jq -R -s -c 'split("\n")[:-1]')
+          echo "test-type=$TEST_TYPES" >> $GITHUB_OUTPUT
+
+  test-matrix:
+    needs: discover-tests
+    runs-on: ubuntu-latest
+
     strategy:
+      fail-fast: false
       matrix:
-        # Listing tests manually since some of them currently fail
+        test-type: ${{ fromJson(needs.discover-tests.outputs.test-type) }}
-        # TODO: generate matrix list from tests/integration when fixed
-        test-type: [agents, inference, datasets, inspect, scoring, post_training, providers, tool_runtime, vector_io]
         client-type: [library, server]
         python-version: ["3.12", "3.13"]
-      fail-fast: false # we want to run all tests regardless of failure
 
     steps:
       - name: Checkout repository
@@ -53,9 +70,11 @@ jobs:
 
       - name: Run Integration Tests
         env:
-          OLLAMA_INFERENCE_MODEL: "meta-llama/Llama-3.2-3B-Instruct" # for server tests
+          OLLAMA_INFERENCE_MODEL: "llama3.2:3b-instruct-fp16" # for server tests
           ENABLE_OLLAMA: "ollama" # for server tests
           OLLAMA_URL: "http://0.0.0.0:11434"
+          SAFETY_MODEL: "llama-guard3:1b"
+          LLAMA_STACK_CLIENT_TIMEOUT: "300" # Increased timeout for eval operations
         # Use 'shell' to get pipefail behavior
         # https://docs.github.com/en/actions/reference/workflow-syntax-for-github-actions#exit-codes-and-error-action-preference
         # TODO: write a precommit hook to detect if a test contains a pipe but does not use 'shell: bash'
@@ -68,8 +87,9 @@ jobs:
           fi
           uv run pytest -s -v tests/integration/${{ matrix.test-type }} --stack-config=${stack_config} \
             -k "not(builtin_tool or safety_with_image or code_interpreter or test_rag)" \
-            --text-model="ollama/meta-llama/Llama-3.2-3B-Instruct" \
+            --text-model="ollama/llama3.2:3b-instruct-fp16" \
             --embedding-model=all-MiniLM-L6-v2 \
+            --safety-shield=$SAFETY_MODEL \
             --color=yes \
             --capture=tee-sys | tee pytest-${{ matrix.test-type }}.log
 

.pre-commit-config.yaml

@@ -29,7 +29,7 @@ repos:
     -   id: check-toml
 
 -   repo: https://github.com/Lucas-C/pre-commit-hooks
-    rev: v1.5.4
+    rev: v1.5.5
     hooks:
     -   id: insert-license
         files: \.py$|\.sh$
@@ -38,7 +38,7 @@ repos:
           - docs/license_header.txt
 
 -   repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.9.4
+    rev: v0.12.2
     hooks:
     -   id: ruff
         args: [ --fix ]
@@ -46,14 +46,14 @@ repos:
     -   id: ruff-format
 
 -   repo: https://github.com/adamchainz/blacken-docs
-    rev: 1.19.0
+    rev: 1.19.1
     hooks:
     -   id: blacken-docs
         additional_dependencies:
         - black==24.3.0
 
 -   repo: https://github.com/astral-sh/uv-pre-commit
-    rev: 0.7.8
+    rev: 0.7.20
     hooks:
     -   id: uv-lock
     -   id: uv-export
@@ -66,7 +66,7 @@ repos:
         ]
 
 -   repo: https://github.com/pre-commit/mirrors-mypy
-    rev: v1.15.0
+    rev: v1.16.1
     hooks:
     -   id: mypy
         additional_dependencies:
@@ -133,3 +133,8 @@ repos:
 ci:
     autofix_commit_msg: 🎨 [pre-commit.ci] Auto format from pre-commit.com hooks
     autoupdate_commit_msg: ⬆ [pre-commit.ci] pre-commit autoupdate
+    autofix_prs: true
+    autoupdate_branch: ''
+    autoupdate_schedule: weekly
+    skip: []
+    submodules: false

CONTRIBUTING.md

@@ -66,7 +66,7 @@ You can install the dependencies by running:
 
 ```bash
 cd llama-stack
-uv sync --extra dev
+uv sync --group dev
 uv pip install -e .
 source .venv/bin/activate
 ```
@@ -168,7 +168,7 @@ manually as they are auto-generated.
 
 ### Updating the provider documentation
 
-If you have made changes to a provider's configuration, you should run `./scripts/distro_codegen.py`
+If you have made changes to a provider's configuration, you should run `./scripts/provider_codegen.py`
 to re-generate the documentation. You should not change `docs/source/.../providers/` files manually
 as they are auto-generated.
 Note that the provider "description" field will be used to generate the provider documentation.

README.md

@@ -77,7 +77,7 @@ As more providers start supporting Llama 4, you can use them in Llama Stack as w
 To try Llama Stack locally, run:
 
 ```bash
-curl -LsSf https://github.com/meta-llama/llama-stack/raw/main/install.sh | bash
+curl -LsSf https://github.com/meta-llama/llama-stack/raw/main/scripts/install.sh | bash
 ```
 
 ### Overview

docs/_static/llama-stack-spec.html

@@ -14796,7 +14796,8 @@
                         "description": "Template for formatting each retrieved chunk in the context. Available placeholders: {index} (1-based chunk ordinal), {chunk.content} (chunk content string), {metadata} (chunk metadata dict). Default: \"Result {index}\\nContent: {chunk.content}\\nMetadata: {metadata}\\n\""
                     },
                     "mode": {
-                        "type": "string",
+                        "$ref": "#/components/schemas/RAGSearchMode",
+                        "default": "vector",
                         "description": "Search mode for retrieval—either \"vector\", \"keyword\", or \"hybrid\". Default \"vector\"."
                     },
                     "ranker": {
@@ -14831,6 +14832,16 @@
                     }
                 }
             },
+            "RAGSearchMode": {
+                "type": "string",
+                "enum": [
+                    "vector",
+                    "keyword",
+                    "hybrid"
+                ],
+                "title": "RAGSearchMode",
+                "description": "Search modes for RAG query retrieval: - VECTOR: Uses vector similarity search for semantic matching - KEYWORD: Uses keyword-based search for exact matching - HYBRID: Combines both vector and keyword search for better results"
+            },
             "RRFRanker": {
                 "type": "object",
                 "properties": {

docs/_static/llama-stack-spec.yaml

@@ -10346,7 +10346,8 @@ components:
             content string), {metadata} (chunk metadata dict). Default: "Result {index}\nContent:
             {chunk.content}\nMetadata: {metadata}\n"
         mode:
-          type: string
+          $ref: '#/components/schemas/RAGSearchMode'
+          default: vector
           description: >-
             Search mode for retrieval—either "vector", "keyword", or "hybrid". Default
             "vector".
@@ -10373,6 +10374,17 @@ components:
         mapping:
           default: '#/components/schemas/DefaultRAGQueryGeneratorConfig'
           llm: '#/components/schemas/LLMRAGQueryGeneratorConfig'
+    RAGSearchMode:
+      type: string
+      enum:
+        - vector
+        - keyword
+        - hybrid
+      title: RAGSearchMode
+      description: >-
+        Search modes for RAG query retrieval: - VECTOR: Uses vector similarity search
+        for semantic matching - KEYWORD: Uses keyword-based search for exact matching
+        - HYBRID: Combines both vector and keyword search for better results
     RRFRanker:
       type: object
       properties:

docs/original_rfc.md

@@ -1,5 +1,7 @@
 # The Llama Stack API
 
+*Originally authored Jul 23, 2024*
+
 **Authors:**
 
 * Meta: @raghotham, @ashwinb, @hjshah, @jspisak
@@ -24,7 +26,7 @@ Meta releases weights of both the pretrained and instruction fine-tuned Llama mo
 
 ### Model Lifecycle
 
-![Figure 1: Model Life Cycle](../docs/resources/model-lifecycle.png)
+![Figure 1: Model Life Cycle](resources/model-lifecycle.png)
 
 For each of the operations that need to be performed (e.g. fine tuning, inference, evals etc) during the model life cycle, we identified the capabilities as toolchain APIs that are needed. Some of these capabilities are primitive operations like inference while other capabilities like synthetic data generation are composed of other capabilities. The list of APIs we have identified to support the lifecycle of Llama models is below:
 
@@ -37,7 +39,7 @@ For each of the operations that need to be performed (e.g. fine tuning, inferenc
 
 ### Agentic System
 
-![Figure 2: Agentic System](../docs/resources/agentic-system.png)
+![Figure 2: Agentic System](resources/agentic-system.png)
 
 In addition to the model lifecycle, we considered the different components involved in an agentic system. Specifically around tool calling and shields. Since the model may decide to call tools, a single model inference call is not enough. What’s needed is an agentic loop consisting of tool calls and inference. The model provides separate tokens representing end-of-message and end-of-turn. A message represents a possible stopping point for execution where the model can inform the execution environment that a tool call needs to be made. The execution environment, upon execution, adds back the result to the context window and makes another inference call. This process can get repeated until an end-of-turn token is generated.
 Note that as of today, in the OSS world, such a “loop” is often coded explicitly via elaborate prompt engineering using a ReAct pattern (typically) or preconstructed execution graph. Llama 3.1 (and future Llamas) attempts to absorb this multi-step reasoning loop inside the main model itself.
@@ -63,9 +65,9 @@ The sequence diagram that details the steps is [here](https://github.com/meta-ll
 
 We define the Llama Stack as a layer cake shown below.
 
-![Figure 3: Llama Stack](../docs/resources/llama-stack.png)
+![Figure 3: Llama Stack](resources/llama-stack.png)
 
-The API is defined in the [YAML](../docs/_static/llama-stack-spec.yaml) and [HTML](../docs/_static/llama-stack-spec.html) files.
+The API is defined in the [YAML](_static/llama-stack-spec.yaml) and [HTML](_static/llama-stack-spec.html) files.
 
 ## Sample implementations
 

docs/source/distributions/building_distro.md

@@ -145,6 +145,10 @@ $ llama stack build --template starter
 ...
 You can now edit ~/.llama/distributions/llamastack-starter/starter-run.yaml and run `llama stack run ~/.llama/distributions/llamastack-starter/starter-run.yaml`
 ```
+
+```{tip}
+The generated `run.yaml` file is a starting point for your configuration. For comprehensive guidance on customizing it for your specific needs, infrastructure, and deployment scenarios, see [Customizing Your run.yaml Configuration](customizing_run_yaml.md).
+```
 :::
 :::{tab-item} Building from Scratch
 

docs/source/distributions/configuration.md

@@ -2,6 +2,10 @@
 
 The Llama Stack runtime configuration is specified as a YAML file. Here is a simplified version of an example configuration file for the Ollama distribution:
 
+```{note}
+The default `run.yaml` files generated by templates are starting points for your configuration. For guidance on customizing these files for your specific needs, see [Customizing Your run.yaml Configuration](customizing_run_yaml.md).
+```
+
 ```{dropdown} 👋 Click here for a Sample Configuration File
 
 ```yaml

docs/source/distributions/customizing_run_yaml.md

@@ -0,0 +1,40 @@
+# Customizing run.yaml Files
+
+The `run.yaml` files generated by Llama Stack templates are **starting points** designed to be customized for your specific needs. They are not meant to be used as-is in production environments.
+
+## Key Points
+
+- **Templates are starting points**: Generated `run.yaml` files contain defaults for development/testing
+- **Customization expected**: Update URLs, credentials, models, and settings for your environment
+- **Version control separately**: Keep customized configs in your own repository
+- **Environment-specific**: Create different configurations for dev, staging, production
+
+## What You Can Customize
+
+You can customize:
+- **Provider endpoints**: Change `http://localhost:8000` to your actual servers
+- **Swap providers**: Replace default providers (e.g., swap Tavily with Brave for search)
+- **Storage paths**: Move from `/tmp/` to production directories
+- **Authentication**: Add API keys, SSL, timeouts
+- **Models**: Different model sizes for dev vs prod
+- **Database settings**: Switch from SQLite to PostgreSQL
+- **Tool configurations**: Add custom tools and integrations
+
+## Best Practices
+
+- Use environment variables for secrets and environment-specific values
+- Create separate `run.yaml` files for different environments (dev, staging, prod)
+- Document your changes with comments
+- Test configurations before deployment
+- Keep your customized configs in version control
+
+Example structure:
+```
+your-project/
+├── configs/
+│   ├── dev-run.yaml
+│   ├── prod-run.yaml
+└── README.md
+```
+
+The goal is to take the generated template and adapt it to your specific infrastructure and operational needs.

docs/source/distributions/index.md

@@ -9,6 +9,7 @@ This section provides an overview of the distributions available in Llama Stack.
 
 importing_as_library
 configuration
+customizing_run_yaml
 list_of_distributions
 kubernetes_deployment
 building_distro

docs/source/distributions/k8s/apply.sh

@@ -6,12 +6,12 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
-export POSTGRES_USER=${POSTGRES_USER:-llamastack}
+export POSTGRES_USER=llamastack
-export POSTGRES_DB=${POSTGRES_DB:-llamastack}
+export POSTGRES_DB=llamastack
-export POSTGRES_PASSWORD=${POSTGRES_PASSWORD:-llamastack}
+export POSTGRES_PASSWORD=llamastack
 
-export INFERENCE_MODEL=${INFERENCE_MODEL:-meta-llama/Llama-3.2-3B-Instruct}
+export INFERENCE_MODEL=meta-llama/Llama-3.2-3B-Instruct
-export SAFETY_MODEL=${SAFETY_MODEL:-meta-llama/Llama-Guard-3-1B}
+export SAFETY_MODEL=meta-llama/Llama-Guard-3-1B
 
 # HF_TOKEN should be set by the user; base64 encode it for the secret
 if [ -n "${HF_TOKEN:-}" ]; then

docs/source/distributions/k8s/vllm-k8s.yaml.template

@@ -32,7 +32,7 @@ spec:
         image: vllm/vllm-openai:latest
         command: ["/bin/sh", "-c"]
         args:
-        - "vllm serve ${INFERENCE_MODEL} --dtype float16 --enforce-eager --max-model-len 4096 --gpu-memory-utilization 0.6"
+        - "vllm serve ${INFERENCE_MODEL} --dtype float16 --enforce-eager --max-model-len 4096 --gpu-memory-utilization 0.6 --enable-auto-tool-choice --tool-call-parser llama4_pythonic"
         env:
         - name: INFERENCE_MODEL
           value: "${INFERENCE_MODEL}"

docs/source/distributions/ondevice_distro/android_sdk.md

@@ -13,7 +13,7 @@ Latest Release Notes: [link](https://github.com/meta-llama/llama-stack-client-ko
 *Tagged releases are stable versions of the project. While we strive to maintain a stable main branch, it's not guaranteed to be free of bugs or issues.*
 
 ## Android Demo App
-Check out our demo app to see how to integrate Llama Stack into your Android app: [Android Demo App](https://github.com/meta-llama/llama-stack-client-kotlin/tree/examples/android_app)
+Check out our demo app to see how to integrate Llama Stack into your Android app: [Android Demo App](https://github.com/meta-llama/llama-stack-client-kotlin/tree/latest-release/examples/android_app)
 
 The key files in the app are `ExampleLlamaStackLocalInference.kt`, `ExampleLlamaStackRemoteInference.kts`, and `MainActivity.java`. With encompassed business logic, the app shows how to use Llama Stack for both the environments.
 
@@ -68,7 +68,7 @@ Ensure the Llama Stack server version is the same as the Kotlin SDK Library for
 
 Other inference providers: [Table](https://llama-stack.readthedocs.io/en/latest/index.html#supported-llama-stack-implementations)
 
-How to set remote localhost in Demo App: [Settings](https://github.com/meta-llama/llama-stack-apps/tree/main/examples/android_app#settings)
+How to set remote localhost in Demo App: [Settings](https://github.com/meta-llama/llama-stack-client-kotlin/tree/latest-release/examples/android_app#settings)
 
 ### Initialize the Client
 A client serves as the primary interface for interacting with a specific inference type and its associated parameters. Only after client is initialized then you can configure and start inferences.
@@ -135,7 +135,7 @@ val result = client!!.inference().chatCompletionStreaming(
 
 ### Setup Custom Tool Calling
 
-Android demo app for more details: [Custom Tool Calling](https://github.com/meta-llama/llama-stack-apps/tree/main/examples/android_app#tool-calling)
+Android demo app for more details: [Custom Tool Calling](https://github.com/meta-llama/llama-stack-client-kotlin/tree/latest-release/examples/android_app#tool-calling)
 
 ## Advanced Users
 

docs/source/getting_started/detailed_tutorial.md

@@ -54,7 +54,7 @@ Llama Stack is a server that exposes multiple APIs, you connect with it using th
 You can use Python to build and run the Llama Stack server, which is useful for testing and development.
 
 Llama Stack uses a [YAML configuration file](../distributions/configuration.md) to specify the stack setup,
-which defines the providers and their settings.
+which defines the providers and their settings. The generated configuration serves as a starting point that you can [customize for your specific needs](../distributions/customizing_run_yaml.md).
 Now let's build and run the Llama Stack config for Ollama.
 We use `starter` as template. By default all providers are disabled, this requires enable ollama by passing environment variables.
 
@@ -77,7 +77,7 @@ ENABLE_OLLAMA=ollama INFERENCE_MODEL="llama3.2:3b" llama stack build --template
 You can use a container image to run the Llama Stack server. We provide several container images for the server
 component that works with different inference providers out of the box. For this guide, we will use
 `llamastack/distribution-starter` as the container image. If you'd like to build your own image or customize the
-configurations, please check out [this guide](../references/index.md).
+configurations, please check out [this guide](../distributions/building_distro.md).
 First lets setup some environment variables and create a local directory to mount into the container’s file system.
 ```bash
 export INFERENCE_MODEL="llama3.2:3b"

docs/source/providers/vector_io/inline_milvus.md

@@ -11,7 +11,7 @@ Please refer to the remote provider documentation.
 | Field | Type | Required | Default | Description |
 |-------|------|----------|---------|-------------|
 | `db_path` | `<class 'str'>` | No | PydanticUndefined |  |
-| `kvstore` | `utils.kvstore.config.RedisKVStoreConfig \| utils.kvstore.config.SqliteKVStoreConfig \| utils.kvstore.config.PostgresKVStoreConfig \| utils.kvstore.config.MongoDBKVStoreConfig` | No | sqlite |  |
+| `kvstore` | `utils.kvstore.config.RedisKVStoreConfig \| utils.kvstore.config.SqliteKVStoreConfig \| utils.kvstore.config.PostgresKVStoreConfig \| utils.kvstore.config.MongoDBKVStoreConfig` | No | sqlite | Config for KV store backend (SQLite only for now) |
 | `consistency_level` | `<class 'str'>` | No | Strong | The consistency level of the Milvus server |
 
 ## Sample Configuration

docs/source/providers/vector_io/inline_sqlite-vec.md

@@ -205,12 +205,16 @@ See [sqlite-vec's GitHub repo](https://github.com/asg017/sqlite-vec/tree/main) f
 
 | Field | Type | Required | Default | Description |
 |-------|------|----------|---------|-------------|
-| `db_path` | `<class 'str'>` | No | PydanticUndefined |  |
+| `db_path` | `<class 'str'>` | No | PydanticUndefined | Path to the SQLite database file |
+| `kvstore` | `utils.kvstore.config.RedisKVStoreConfig \| utils.kvstore.config.SqliteKVStoreConfig \| utils.kvstore.config.PostgresKVStoreConfig \| utils.kvstore.config.MongoDBKVStoreConfig` | No | sqlite | Config for KV store backend (SQLite only for now) |
 
 ## Sample Configuration
 
 ```yaml
 db_path: ${env.SQLITE_STORE_DIR:=~/.llama/dummy}/sqlite_vec.db
+kvstore:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/dummy}/sqlite_vec_registry.db
 
 ```
 

docs/source/providers/vector_io/inline_sqlite_vec.md

@@ -10,12 +10,16 @@ Please refer to the sqlite-vec provider documentation.
 
 | Field | Type | Required | Default | Description |
 |-------|------|----------|---------|-------------|
-| `db_path` | `<class 'str'>` | No | PydanticUndefined |  |
+| `db_path` | `<class 'str'>` | No | PydanticUndefined | Path to the SQLite database file |
+| `kvstore` | `utils.kvstore.config.RedisKVStoreConfig \| utils.kvstore.config.SqliteKVStoreConfig \| utils.kvstore.config.PostgresKVStoreConfig \| utils.kvstore.config.MongoDBKVStoreConfig` | No | sqlite | Config for KV store backend (SQLite only for now) |
 
 ## Sample Configuration
 
 ```yaml
 db_path: ${env.SQLITE_STORE_DIR:=~/.llama/dummy}/sqlite_vec.db
+kvstore:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/dummy}/sqlite_vec_registry.db
 
 ```
 

docs/source/providers/vector_io/remote_pgvector.md

@@ -40,6 +40,7 @@ See [PGVector's documentation](https://github.com/pgvector/pgvector) for more de
 | `db` | `str \| None` | No | postgres |  |
 | `user` | `str \| None` | No | postgres |  |
 | `password` | `str \| None` | No | mysecretpassword |  |
+| `kvstore` | `utils.kvstore.config.RedisKVStoreConfig \| utils.kvstore.config.SqliteKVStoreConfig \| utils.kvstore.config.PostgresKVStoreConfig \| utils.kvstore.config.MongoDBKVStoreConfig, annotation=NoneType, required=False, default='sqlite', discriminator='type'` | No |  | Config for KV store backend (SQLite only for now) |
 
 ## Sample Configuration
 
@@ -49,6 +50,9 @@ port: ${env.PGVECTOR_PORT:=5432}
 db: ${env.PGVECTOR_DB}
 user: ${env.PGVECTOR_USER}
 password: ${env.PGVECTOR_PASSWORD}
+kvstore:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/dummy}/pgvector_registry.db
 
 ```
 

docs/source/providers/vector_io/remote_weaviate.md

@@ -36,7 +36,9 @@ See [Weaviate's documentation](https://weaviate.io/developers/weaviate) for more
 ## Sample Configuration
 
 ```yaml
-{}
+kvstore:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/dummy}/weaviate_registry.db
 
 ```
 

llama_stack/apis/tools/rag_tool.py

@@ -87,6 +87,20 @@ class RAGQueryGenerator(Enum):
     custom = "custom"
 
 
+@json_schema_type
+class RAGSearchMode(Enum):
+    """
+    Search modes for RAG query retrieval:
+    - VECTOR: Uses vector similarity search for semantic matching
+    - KEYWORD: Uses keyword-based search for exact matching
+    - HYBRID: Combines both vector and keyword search for better results
+    """
+
+    VECTOR = "vector"
+    KEYWORD = "keyword"
+    HYBRID = "hybrid"
+
+
 @json_schema_type
 class DefaultRAGQueryGeneratorConfig(BaseModel):
     type: Literal["default"] = "default"
@@ -128,7 +142,7 @@ class RAGQueryConfig(BaseModel):
     max_tokens_in_context: int = 4096
     max_chunks: int = 5
     chunk_template: str = "Result {index}\nContent: {chunk.content}\nMetadata: {metadata}\n"
-    mode: str | None = None
+    mode: RAGSearchMode | None = RAGSearchMode.VECTOR
     ranker: Ranker | None = Field(default=None)  # Only used for hybrid mode
 
     @field_validator("chunk_template")

llama_stack/cli/stack/_build.py

@@ -93,7 +93,7 @@ def run_stack_build_command(args: argparse.Namespace) -> None:
             )
             sys.exit(1)
     elif args.providers:
-        providers = dict()
+        providers_list: dict[str, str | list[str]] = dict()
         for api_provider in args.providers.split(","):
             if "=" not in api_provider:
                 cprint(
@@ -112,7 +112,15 @@ def run_stack_build_command(args: argparse.Namespace) -> None:
                 )
                 sys.exit(1)
             if provider in providers_for_api:
-                providers.setdefault(api, []).append(provider)
+                if api not in providers_list:
+                    providers_list[api] = []
+                # Use type guarding to ensure we have a list
+                provider_value = providers_list[api]
+                if isinstance(provider_value, list):
+                    provider_value.append(provider)
+                else:
+                    # Convert string to list and append
+                    providers_list[api] = [provider_value, provider]
             else:
                 cprint(
                     f"{provider} is not a valid provider for the {api} API.",
@@ -121,7 +129,7 @@ def run_stack_build_command(args: argparse.Namespace) -> None:
                 )
                 sys.exit(1)
         distribution_spec = DistributionSpec(
-            providers=providers,
+            providers=providers_list,
             description=",".join(args.providers),
         )
         if not args.image_type:
@@ -182,7 +190,7 @@ def run_stack_build_command(args: argparse.Namespace) -> None:
 
         cprint("Tip: use <TAB> to see options for the providers.\n", color="green", file=sys.stderr)
 
-        providers = dict()
+        providers: dict[str, str | list[str]] = dict()
         for api, providers_for_api in get_provider_registry().items():
             available_providers = [x for x in providers_for_api.keys() if x not in ("remote", "remote::sample")]
             if not available_providers:
@@ -371,10 +379,16 @@ def _run_stack_build_command_from_build_config(
         if not image_name:
             raise ValueError("Please specify an image name when building a venv image")
 
+    # At this point, image_name should be guaranteed to be a string
+    if image_name is None:
+        raise ValueError("image_name should not be None after validation")
+
     if template_name:
         build_dir = DISTRIBS_BASE_DIR / template_name
         build_file_path = build_dir / f"{template_name}-build.yaml"
     else:
+        if image_name is None:
+            raise ValueError("image_name cannot be None")
         build_dir = DISTRIBS_BASE_DIR / image_name
         build_file_path = build_dir / f"{image_name}-build.yaml"
 
@@ -395,7 +409,7 @@ def _run_stack_build_command_from_build_config(
         build_file_path,
         image_name,
         template_or_config=template_name or config_path or str(build_file_path),
-        run_config=run_config_file,
+        run_config=run_config_file.as_posix() if run_config_file else None,
     )
     if return_code != 0:
         raise RuntimeError(f"Failed to build image {image_name}")

llama_stack/cli/stack/run.py

@@ -83,22 +83,13 @@ class StackRun(Subcommand):
             return ImageType.CONDA.value, args.image_name
         return args.image_type, args.image_name
 
-    def _run_stack_run_cmd(self, args: argparse.Namespace) -> None:
+    def _resolve_config_and_template(self, args: argparse.Namespace) -> tuple[Path | None, str | None]:
-        import yaml
+        """Resolve config file path and template name from args.config"""
-
-        from llama_stack.distribution.configure import parse_and_maybe_upgrade_config
         from llama_stack.distribution.utils.config_dirs import DISTRIBS_BASE_DIR
-        from llama_stack.distribution.utils.exec import formulate_run_args, run_command
 
-        if args.enable_ui:
+        if not args.config:
-            self._start_ui_development_server(args.port)
+            return None, None
-        image_type, image_name = self._get_image_type_and_name(args)
 
-        # Check if config is required based on image type
-        if (image_type in [ImageType.CONDA.value, ImageType.VENV.value]) and not args.config:
-            self.parser.error("Config file is required for venv and conda environments")
-
-        if args.config:
         config_file = Path(args.config)
         has_yaml_suffix = args.config.endswith(".yaml")
         template_name = None
@@ -123,6 +114,26 @@ class StackRun(Subcommand):
                 f"Config file must be a valid file path, '{config_file}' is not a file: type={type(config_file)}"
             )
 
+        return config_file, template_name
+
+    def _run_stack_run_cmd(self, args: argparse.Namespace) -> None:
+        import yaml
+
+        from llama_stack.distribution.configure import parse_and_maybe_upgrade_config
+        from llama_stack.distribution.utils.exec import formulate_run_args, run_command
+
+        if args.enable_ui:
+            self._start_ui_development_server(args.port)
+        image_type, image_name = self._get_image_type_and_name(args)
+
+        # Resolve config file and template name first
+        config_file, template_name = self._resolve_config_and_template(args)
+
+        # Check if config is required based on image type
+        if (image_type in [ImageType.CONDA.value, ImageType.VENV.value]) and not config_file:
+            self.parser.error("Config file is required for venv and conda environments")
+
+        if config_file:
             logger.info(f"Using run configuration: {config_file}")
 
             try:
@@ -138,8 +149,6 @@ class StackRun(Subcommand):
                 self.parser.error(f"failed to parse config file '{config_file}':\n {e}")
         else:
             config = None
-            config_file = None
-            template_name = None
 
         # If neither image type nor image name is provided, assume the server should be run directly
         # using the current environment packages.
@@ -172,9 +181,6 @@ class StackRun(Subcommand):
             run_args.extend([str(args.port)])
 
             if config_file:
-                if template_name:
-                    run_args.extend(["--template", str(template_name)])
-                else:
                 run_args.extend(["--config", str(config_file)])
 
             if args.env:

llama_stack/distribution/access_control/access_control.py

@@ -81,7 +81,7 @@ def is_action_allowed(
     if not len(policy):
         policy = default_policy()
 
-    qualified_resource_id = resource.type + "::" + resource.identifier
+    qualified_resource_id = f"{resource.type}::{resource.identifier}"
     for rule in policy:
         if rule.forbid and matches_scope(rule.forbid, action, qualified_resource_id, user.principal):
             if rule.when:

llama_stack/distribution/build_container.sh

@@ -96,7 +96,7 @@ FROM $container_base
 WORKDIR /app
 
 # We install the Python 3.12 dev headers and build tools so that any
-# C‑extension wheels (e.g. polyleven, faiss‑cpu) can compile successfully.
+# C-extension wheels (e.g. polyleven, faiss-cpu) can compile successfully.
 
 RUN dnf -y update && dnf install -y iputils git net-tools wget \
     vim-minimal python3.12 python3.12-pip python3.12-wheel \
@@ -169,7 +169,7 @@ if [ -n "$run_config" ]; then
     echo "Copying external providers directory: $external_providers_dir"
     cp -r "$external_providers_dir" "$BUILD_CONTEXT_DIR/providers.d"
     add_to_container << EOF
-COPY --chmod=g+w providers.d /.llama/providers.d
+COPY providers.d /.llama/providers.d
 EOF
     fi
 

llama_stack/distribution/server/server.py

@@ -445,7 +445,7 @@ def main(args: argparse.Namespace | None = None):
     logger.info(log_line)
 
     logger.info("Run configuration:")
-    safe_config = redact_sensitive_fields(config.model_dump())
+    safe_config = redact_sensitive_fields(config.model_dump(mode="json"))
     logger.info(yaml.dump(safe_config, indent=2))
 
     app = FastAPI(

llama_stack/distribution/stack.py

@@ -98,6 +98,7 @@ async def register_resources(run_config: StackRunConfig, impls: dict[Api, Any]):
 
         method = getattr(impls[api], register_method)
         for obj in objects:
+            logger.debug(f"registering {rsrc.capitalize()} {obj} for provider {obj.provider_id}")
             # Do not register models on disabled providers
             if hasattr(obj, "provider_id") and obj.provider_id is not None and obj.provider_id == "__disabled__":
                 logger.debug(f"Skipping {rsrc.capitalize()} registration for disabled provider.")
@@ -112,6 +113,11 @@ async def register_resources(run_config: StackRunConfig, impls: dict[Api, Any]):
             ):
                 logger.debug(f"Skipping {rsrc.capitalize()} registration for disabled model.")
                 continue
+
+            if hasattr(obj, "shield_id") and obj.shield_id is not None and obj.shield_id == "__disabled__":
+                logger.debug(f"Skipping {rsrc.capitalize()} registration for disabled shield.")
+                continue
+
             # we want to maintain the type information in arguments to method.
             # instead of method(**obj.model_dump()), which may convert a typed attr to a dict,
             # we use model_dump() to find all the attrs and then getattr to get the still typed value.

llama_stack/distribution/utils/context.py

@@ -6,12 +6,9 @@
 
 from collections.abc import AsyncGenerator
 from contextvars import ContextVar
-from typing import TypeVar
-
-T = TypeVar("T")
 
 
-def preserve_contexts_async_generator(
+def preserve_contexts_async_generator[T](
     gen: AsyncGenerator[T, None], context_vars: list[ContextVar]
 ) -> AsyncGenerator[T, None]:
     """

llama_stack/providers/inline/post_training/torchtune/post_training.py

@@ -123,7 +123,8 @@ class TorchtunePostTrainingImpl:
         training_config: TrainingConfig,
         hyperparam_search_config: dict[str, Any],
         logger_config: dict[str, Any],
-    ) -> PostTrainingJob: ...
+    ) -> PostTrainingJob:
+        raise NotImplementedError()
 
     async def get_training_jobs(self) -> ListPostTrainingJobsResponse:
         return ListPostTrainingJobsResponse(

llama_stack/providers/inline/safety/llama_guard/llama_guard.py

@@ -146,10 +146,9 @@ class LlamaGuardSafetyImpl(Safety, ShieldsProtocolPrivate):
         pass
 
     async def register_shield(self, shield: Shield) -> None:
-        if shield.provider_resource_id not in LLAMA_GUARD_MODEL_IDS:
+        # Allow any model to be registered as a shield
-            raise ValueError(
+        # The model will be validated during runtime when making inference calls
-                f"Unsupported Llama Guard type: {shield.provider_resource_id}. Allowed types: {LLAMA_GUARD_MODEL_IDS}"
+        pass
-            )
 
     async def run_shield(
         self,
@@ -167,11 +166,25 @@ class LlamaGuardSafetyImpl(Safety, ShieldsProtocolPrivate):
         if len(messages) > 0 and messages[0].role != Role.user.value:
             messages[0] = UserMessage(content=messages[0].content)
 
-        model = LLAMA_GUARD_MODEL_IDS[shield.provider_resource_id]
+        # Use the inference API's model resolution instead of hardcoded mappings
+        # This allows the shield to work with any registered model
+        model_id = shield.provider_resource_id
+
+        # Determine safety categories based on the model type
+        # For known Llama Guard models, use specific categories
+        if model_id in LLAMA_GUARD_MODEL_IDS:
+            # Use the mapped model for categories but the original model_id for inference
+            mapped_model = LLAMA_GUARD_MODEL_IDS[model_id]
+            safety_categories = MODEL_TO_SAFETY_CATEGORIES_MAP.get(mapped_model, DEFAULT_LG_V3_SAFETY_CATEGORIES)
+        else:
+            # For unknown models, use default Llama Guard 3 8B categories
+            safety_categories = DEFAULT_LG_V3_SAFETY_CATEGORIES + [CAT_CODE_INTERPRETER_ABUSE]
+
         impl = LlamaGuardShield(
-            model=model,
+            model=model_id,
             inference_api=self.inference_api,
             excluded_categories=self.config.excluded_categories,
+            safety_categories=safety_categories,
         )
 
         return await impl.run(messages)
@@ -183,20 +196,21 @@ class LlamaGuardShield:
         model: str,
         inference_api: Inference,
         excluded_categories: list[str] | None = None,
+        safety_categories: list[str] | None = None,
     ):
         if excluded_categories is None:
             excluded_categories = []
+        if safety_categories is None:
+            safety_categories = []
 
         assert len(excluded_categories) == 0 or all(
             x in SAFETY_CATEGORIES_TO_CODE_MAP.values() for x in excluded_categories
         ), "Invalid categories in excluded categories. Expected format is ['S1', 'S2', ..]"
 
-        if model not in MODEL_TO_SAFETY_CATEGORIES_MAP:
-            raise ValueError(f"Unsupported model: {model}")
-
         self.model = model
         self.inference_api = inference_api
         self.excluded_categories = excluded_categories
+        self.safety_categories = safety_categories
 
     def check_unsafe_response(self, response: str) -> str | None:
         match = re.match(r"^unsafe\n(.*)$", response)
@@ -214,7 +228,7 @@ class LlamaGuardShield:
 
         final_categories = []
 
-        all_categories = MODEL_TO_SAFETY_CATEGORIES_MAP[self.model]
+        all_categories = self.safety_categories
         for cat in all_categories:
             cat_code = SAFETY_CATEGORIES_TO_CODE_MAP[cat]
             if cat_code in excluded_categories:

llama_stack/providers/inline/vector_io/faiss/faiss.py

@@ -181,8 +181,8 @@ class FaissVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolPr
             )
             self.cache[vector_db.identifier] = index
 
-        # Load existing OpenAI vector stores using the mixin method
+        # Load existing OpenAI vector stores into the in-memory cache
-        self.openai_vector_stores = await self._load_openai_vector_stores()
+        await self.initialize_openai_vector_stores()
 
     async def shutdown(self) -> None:
         # Cleanup if needed
@@ -261,42 +261,10 @@ class FaissVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolPr
 
         return await index.query_chunks(query, params)
 
-    # OpenAI Vector Store Mixin abstract method implementations
-    async def _save_openai_vector_store(self, store_id: str, store_info: dict[str, Any]) -> None:
-        """Save vector store metadata to kvstore."""
-        assert self.kvstore is not None
-        key = f"{OPENAI_VECTOR_STORES_PREFIX}{store_id}"
-        await self.kvstore.set(key=key, value=json.dumps(store_info))
-
-    async def _load_openai_vector_stores(self) -> dict[str, dict[str, Any]]:
-        """Load all vector store metadata from kvstore."""
-        assert self.kvstore is not None
-        start_key = OPENAI_VECTOR_STORES_PREFIX
-        end_key = f"{OPENAI_VECTOR_STORES_PREFIX}\xff"
-        stored_openai_stores = await self.kvstore.values_in_range(start_key, end_key)
-
-        stores = {}
-        for store_data in stored_openai_stores:
-            store_info = json.loads(store_data)
-            stores[store_info["id"]] = store_info
-        return stores
-
-    async def _update_openai_vector_store(self, store_id: str, store_info: dict[str, Any]) -> None:
-        """Update vector store metadata in kvstore."""
-        assert self.kvstore is not None
-        key = f"{OPENAI_VECTOR_STORES_PREFIX}{store_id}"
-        await self.kvstore.set(key=key, value=json.dumps(store_info))
-
-    async def _delete_openai_vector_store_from_storage(self, store_id: str) -> None:
-        """Delete vector store metadata from kvstore."""
-        assert self.kvstore is not None
-        key = f"{OPENAI_VECTOR_STORES_PREFIX}{store_id}"
-        await self.kvstore.delete(key)
-
     async def _save_openai_vector_store_file(
         self, store_id: str, file_id: str, file_info: dict[str, Any], file_contents: list[dict[str, Any]]
     ) -> None:
-        """Save vector store file metadata to kvstore."""
+        """Save vector store file data to kvstore."""
         assert self.kvstore is not None
         key = f"{OPENAI_VECTOR_STORES_FILES_PREFIX}{store_id}:{file_id}"
         await self.kvstore.set(key=key, value=json.dumps(file_info))
@@ -324,7 +292,16 @@ class FaissVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolPr
         await self.kvstore.set(key=key, value=json.dumps(file_info))
 
     async def _delete_openai_vector_store_file_from_storage(self, store_id: str, file_id: str) -> None:
-        """Delete vector store file metadata from kvstore."""
+        """Delete vector store data from kvstore."""
         assert self.kvstore is not None
-        key = f"{OPENAI_VECTOR_STORES_FILES_PREFIX}{store_id}:{file_id}"
+
+        keys_to_delete = [
+            f"{OPENAI_VECTOR_STORES_FILES_PREFIX}{store_id}:{file_id}",
+            f"{OPENAI_VECTOR_STORES_FILES_CONTENTS_PREFIX}{store_id}:{file_id}",
+        ]
+        for key in keys_to_delete:
+            try:
                 await self.kvstore.delete(key)
+            except Exception as e:
+                logger.warning(f"Failed to delete key {key}: {e}")
+                continue

llama_stack/providers/inline/vector_io/milvus/config.py

@@ -18,7 +18,7 @@ from llama_stack.schema_utils import json_schema_type
 @json_schema_type
 class MilvusVectorIOConfig(BaseModel):
     db_path: str
-    kvstore: KVStoreConfig
+    kvstore: KVStoreConfig = Field(description="Config for KV store backend (SQLite only for now)")
     consistency_level: str = Field(description="The consistency level of the Milvus server", default="Strong")
 
     @classmethod

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

@@ -6,14 +6,24 @@
 
 from typing import Any
 
-from pydantic import BaseModel
+from pydantic import BaseModel, Field
+
+from llama_stack.providers.utils.kvstore.config import (
+    KVStoreConfig,
+    SqliteKVStoreConfig,
+)
 
 
 class SQLiteVectorIOConfig(BaseModel):
-    db_path: str
+    db_path: str = Field(description="Path to the SQLite database file")
+    kvstore: KVStoreConfig = Field(description="Config for KV store backend (SQLite only for now)")
 
     @classmethod
     def sample_run_config(cls, __distro_dir__: str) -> dict[str, Any]:
         return {
             "db_path": "${env.SQLITE_STORE_DIR:=" + __distro_dir__ + "}/" + "sqlite_vec.db",
+            "kvstore": SqliteKVStoreConfig.sample_run_config(
+                __distro_dir__=__distro_dir__,
+                db_name="sqlite_vec_registry.db",
+            ),
         }

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

@@ -24,6 +24,8 @@ from llama_stack.apis.vector_io import (
     VectorIO,
 )
 from llama_stack.providers.datatypes import VectorDBsProtocolPrivate
+from llama_stack.providers.utils.kvstore import kvstore_impl
+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,
@@ -40,6 +42,13 @@ KEYWORD_SEARCH = "keyword"
 HYBRID_SEARCH = "hybrid"
 SEARCH_MODES = {VECTOR_SEARCH, KEYWORD_SEARCH, HYBRID_SEARCH}
 
+VERSION = "v3"
+VECTOR_DBS_PREFIX = f"vector_dbs:sqlite_vec:{VERSION}::"
+VECTOR_INDEX_PREFIX = f"vector_index:sqlite_vec:{VERSION}::"
+OPENAI_VECTOR_STORES_PREFIX = f"openai_vector_stores:sqlite_vec:{VERSION}::"
+OPENAI_VECTOR_STORES_FILES_PREFIX = f"openai_vector_stores_files:sqlite_vec:{VERSION}::"
+OPENAI_VECTOR_STORES_FILES_CONTENTS_PREFIX = f"openai_vector_stores_files_contents:sqlite_vec:{VERSION}::"
+
 
 def serialize_vector(vector: list[float]) -> bytes:
     """Serialize a list of floats into a compact binary representation."""
@@ -117,13 +126,14 @@ class SQLiteVecIndex(EmbeddingIndex):
       - An FTS5 table (fts_chunks_{bank_id}) for full-text keyword search.
     """
 
-    def __init__(self, dimension: int, db_path: str, bank_id: str):
+    def __init__(self, dimension: int, db_path: str, bank_id: str, kvstore: KVStore | None = None):
         self.dimension = dimension
         self.db_path = db_path
         self.bank_id = bank_id
         self.metadata_table = f"chunks_{bank_id}".replace("-", "_")
         self.vector_table = f"vec_chunks_{bank_id}".replace("-", "_")
         self.fts_table = f"fts_chunks_{bank_id}".replace("-", "_")
+        self.kvstore = kvstore
 
     @classmethod
     async def create(cls, dimension: int, db_path: str, bank_id: str):
@@ -425,27 +435,81 @@ class SQLiteVecVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtoc
         self.files_api = files_api
         self.cache: dict[str, VectorDBWithIndex] = {}
         self.openai_vector_stores: dict[str, dict[str, Any]] = {}
+        self.kvstore: KVStore | None = None
 
     async def initialize(self) -> None:
-        def _setup_connection():
+        self.kvstore = await kvstore_impl(self.config.kvstore)
-            # Open a connection to the SQLite database (the file is specified in the config).
+
+        start_key = VECTOR_DBS_PREFIX
+        end_key = f"{VECTOR_DBS_PREFIX}\xff"
+        stored_vector_dbs = await self.kvstore.values_in_range(start_key, end_key)
+        for db_json in stored_vector_dbs:
+            vector_db = VectorDB.model_validate_json(db_json)
+            index = await SQLiteVecIndex.create(
+                vector_db.embedding_dimension,
+                self.config.db_path,
+                vector_db.identifier,
+            )
+            self.cache[vector_db.identifier] = VectorDBWithIndex(vector_db, index, self.inference_api)
+
+        # Load existing OpenAI vector stores into the in-memory cache
+        await self.initialize_openai_vector_stores()
+
+    async def shutdown(self) -> None:
+        # nothing to do since we don't maintain a persistent connection
+        pass
+
+    async def list_vector_dbs(self) -> list[VectorDB]:
+        return [v.vector_db for v in self.cache.values()]
+
+    async def register_vector_db(self, vector_db: VectorDB) -> None:
+        index = await SQLiteVecIndex.create(
+            vector_db.embedding_dimension,
+            self.config.db_path,
+            vector_db.identifier,
+        )
+        self.cache[vector_db.identifier] = VectorDBWithIndex(vector_db, index, self.inference_api)
+
+    async def _get_and_cache_vector_db_index(self, vector_db_id: str) -> VectorDBWithIndex | None:
+        if vector_db_id in self.cache:
+            return self.cache[vector_db_id]
+
+        if self.vector_db_store is None:
+            raise ValueError(f"Vector DB {vector_db_id} not found")
+
+        vector_db = self.vector_db_store.get_vector_db(vector_db_id)
+        if not vector_db:
+            raise ValueError(f"Vector DB {vector_db_id} not found")
+
+        index = VectorDBWithIndex(
+            vector_db=vector_db,
+            index=SQLiteVecIndex(
+                dimension=vector_db.embedding_dimension,
+                db_path=self.config.db_path,
+                bank_id=vector_db.identifier,
+                kvstore=self.kvstore,
+            ),
+            inference_api=self.inference_api,
+        )
+        self.cache[vector_db_id] = index
+        return index
+
+    async def unregister_vector_db(self, vector_db_id: str) -> None:
+        if vector_db_id not in self.cache:
+            logger.warning(f"Vector DB {vector_db_id} not found")
+            return
+        await self.cache[vector_db_id].index.delete()
+        del self.cache[vector_db_id]
+
+    async def _save_openai_vector_store_file(
+        self, store_id: str, file_id: str, file_info: dict[str, Any], file_contents: list[dict[str, Any]]
+    ) -> None:
+        """Save vector store file metadata to SQLite database."""
+
+        def _create_or_store():
             connection = _create_sqlite_connection(self.config.db_path)
             cur = connection.cursor()
             try:
-                # Create a table to persist vector DB registrations.
-                cur.execute("""
-                    CREATE TABLE IF NOT EXISTS vector_dbs (
-                        id TEXT PRIMARY KEY,
-                        metadata TEXT
-                    );
-                """)
-                # Create a table to persist OpenAI vector stores.
-                cur.execute("""
-                    CREATE TABLE IF NOT EXISTS openai_vector_stores (
-                        id TEXT PRIMARY KEY,
-                        metadata TEXT
-                    );
-                """)
                 # Create a table to persist OpenAI vector store files.
                 cur.execute("""
                     CREATE TABLE IF NOT EXISTS openai_vector_store_files (
@@ -464,168 +528,6 @@ class SQLiteVecVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtoc
                     );
                 """)
                 connection.commit()
-                # Load any existing vector DB registrations.
-                cur.execute("SELECT metadata FROM vector_dbs")
-                vector_db_rows = cur.fetchall()
-                return vector_db_rows
-            finally:
-                cur.close()
-                connection.close()
-
-        vector_db_rows = await asyncio.to_thread(_setup_connection)
-
-        # Load existing vector DBs
-        for row in vector_db_rows:
-            vector_db_data = row[0]
-            vector_db = VectorDB.model_validate_json(vector_db_data)
-            index = await SQLiteVecIndex.create(
-                vector_db.embedding_dimension,
-                self.config.db_path,
-                vector_db.identifier,
-            )
-            self.cache[vector_db.identifier] = VectorDBWithIndex(vector_db, index, self.inference_api)
-
-        # Load existing OpenAI vector stores using the mixin method
-        self.openai_vector_stores = await self._load_openai_vector_stores()
-
-    async def shutdown(self) -> None:
-        # nothing to do since we don't maintain a persistent connection
-        pass
-
-    async def register_vector_db(self, vector_db: VectorDB) -> None:
-        def _register_db():
-            connection = _create_sqlite_connection(self.config.db_path)
-            cur = connection.cursor()
-            try:
-                cur.execute(
-                    "INSERT OR REPLACE INTO vector_dbs (id, metadata) VALUES (?, ?)",
-                    (vector_db.identifier, vector_db.model_dump_json()),
-                )
-                connection.commit()
-            finally:
-                cur.close()
-                connection.close()
-
-        await asyncio.to_thread(_register_db)
-        index = await SQLiteVecIndex.create(
-            vector_db.embedding_dimension,
-            self.config.db_path,
-            vector_db.identifier,
-        )
-        self.cache[vector_db.identifier] = VectorDBWithIndex(vector_db, index, self.inference_api)
-
-    async def list_vector_dbs(self) -> list[VectorDB]:
-        return [v.vector_db for v in self.cache.values()]
-
-    async def unregister_vector_db(self, vector_db_id: str) -> None:
-        if vector_db_id not in self.cache:
-            logger.warning(f"Vector DB {vector_db_id} not found")
-            return
-        await self.cache[vector_db_id].index.delete()
-        del self.cache[vector_db_id]
-
-        def _delete_vector_db_from_registry():
-            connection = _create_sqlite_connection(self.config.db_path)
-            cur = connection.cursor()
-            try:
-                cur.execute("DELETE FROM vector_dbs WHERE id = ?", (vector_db_id,))
-                connection.commit()
-            finally:
-                cur.close()
-                connection.close()
-
-        await asyncio.to_thread(_delete_vector_db_from_registry)
-
-    # OpenAI Vector Store Mixin abstract method implementations
-    async def _save_openai_vector_store(self, store_id: str, store_info: dict[str, Any]) -> None:
-        """Save vector store metadata to SQLite database."""
-
-        def _store():
-            connection = _create_sqlite_connection(self.config.db_path)
-            cur = connection.cursor()
-            try:
-                cur.execute(
-                    "INSERT OR REPLACE INTO openai_vector_stores (id, metadata) VALUES (?, ?)",
-                    (store_id, json.dumps(store_info)),
-                )
-                connection.commit()
-            except Exception as e:
-                logger.error(f"Error saving openai vector store {store_id}: {e}")
-                raise
-            finally:
-                cur.close()
-                connection.close()
-
-        try:
-            await asyncio.to_thread(_store)
-        except Exception as e:
-            logger.error(f"Error saving openai vector store {store_id}: {e}")
-            raise
-
-    async def _load_openai_vector_stores(self) -> dict[str, dict[str, Any]]:
-        """Load all vector store metadata from SQLite database."""
-
-        def _load():
-            connection = _create_sqlite_connection(self.config.db_path)
-            cur = connection.cursor()
-            try:
-                cur.execute("SELECT metadata FROM openai_vector_stores")
-                rows = cur.fetchall()
-                return rows
-            finally:
-                cur.close()
-                connection.close()
-
-        rows = await asyncio.to_thread(_load)
-        stores = {}
-        for row in rows:
-            store_data = row[0]
-            store_info = json.loads(store_data)
-            stores[store_info["id"]] = store_info
-        return stores
-
-    async def _update_openai_vector_store(self, store_id: str, store_info: dict[str, Any]) -> None:
-        """Update vector store metadata in SQLite database."""
-
-        def _update():
-            connection = _create_sqlite_connection(self.config.db_path)
-            cur = connection.cursor()
-            try:
-                cur.execute(
-                    "UPDATE openai_vector_stores SET metadata = ? WHERE id = ?",
-                    (json.dumps(store_info), store_id),
-                )
-                connection.commit()
-            finally:
-                cur.close()
-                connection.close()
-
-        await asyncio.to_thread(_update)
-
-    async def _delete_openai_vector_store_from_storage(self, store_id: str) -> None:
-        """Delete vector store metadata from SQLite database."""
-
-        def _delete():
-            connection = _create_sqlite_connection(self.config.db_path)
-            cur = connection.cursor()
-            try:
-                cur.execute("DELETE FROM openai_vector_stores WHERE id = ?", (store_id,))
-                connection.commit()
-            finally:
-                cur.close()
-                connection.close()
-
-        await asyncio.to_thread(_delete)
-
-    async def _save_openai_vector_store_file(
-        self, store_id: str, file_id: str, file_info: dict[str, Any], file_contents: list[dict[str, Any]]
-    ) -> None:
-        """Save vector store file metadata to SQLite database."""
-
-        def _store():
-            connection = _create_sqlite_connection(self.config.db_path)
-            cur = connection.cursor()
-            try:
                 cur.execute(
                     "INSERT OR REPLACE INTO openai_vector_store_files (store_id, file_id, metadata) VALUES (?, ?, ?)",
                     (store_id, file_id, json.dumps(file_info)),
@@ -643,7 +545,7 @@ class SQLiteVecVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtoc
                 connection.close()
 
         try:
-            await asyncio.to_thread(_store)
+            await asyncio.to_thread(_create_or_store)
         except Exception as e:
             logger.error(f"Error saving openai vector store file {store_id} {file_id}: {e}")
             raise
@@ -722,6 +624,10 @@ class SQLiteVecVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtoc
                 cur.execute(
                     "DELETE FROM openai_vector_store_files WHERE store_id = ? AND file_id = ?", (store_id, file_id)
                 )
+                cur.execute(
+                    "DELETE FROM openai_vector_store_files_contents WHERE store_id = ? AND file_id = ?",
+                    (store_id, file_id),
+                )
                 connection.commit()
             finally:
                 cur.close()
@@ -730,15 +636,17 @@ class SQLiteVecVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtoc
         await asyncio.to_thread(_delete)
 
     async def insert_chunks(self, vector_db_id: str, chunks: list[Chunk], ttl_seconds: int | None = None) -> None:
-        if vector_db_id not in self.cache:
+        index = await self._get_and_cache_vector_db_index(vector_db_id)
-            raise ValueError(f"Vector DB {vector_db_id} not found. Found: {list(self.cache.keys())}")
+        if not index:
+            raise ValueError(f"Vector DB {vector_db_id} not found")
         # The VectorDBWithIndex helper is expected to compute embeddings via the inference_api
         # and then call our index's add_chunks.
-        await self.cache[vector_db_id].insert_chunks(chunks)
+        await index.insert_chunks(chunks)
 
     async def query_chunks(
         self, vector_db_id: str, query: Any, params: dict[str, Any] | None = None
     ) -> QueryChunksResponse:
-        if vector_db_id not in self.cache:
+        index = await self._get_and_cache_vector_db_index(vector_db_id)
+        if not index:
             raise ValueError(f"Vector DB {vector_db_id} not found")
-        return await self.cache[vector_db_id].query_chunks(query, params)
+        return await index.query_chunks(query, params)

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

@@ -15,8 +15,11 @@ LLM_MODEL_IDS = [
     "anthropic/claude-3-5-haiku-latest",
 ]
 
+SAFETY_MODELS_ENTRIES = []
 
-MODEL_ENTRIES = [ProviderModelEntry(provider_model_id=m) for m in LLM_MODEL_IDS] + [
+MODEL_ENTRIES = (
+    [ProviderModelEntry(provider_model_id=m) for m in LLM_MODEL_IDS]
+    + [
         ProviderModelEntry(
             provider_model_id="anthropic/voyage-3",
             model_type=ModelType.embedding,
@@ -32,4 +35,6 @@ MODEL_ENTRIES = [ProviderModelEntry(provider_model_id=m) for m in LLM_MODEL_IDS]
             model_type=ModelType.embedding,
             metadata={"embedding_dimension": 1024, "context_length": 32000},
         ),
-]
+    ]
+    + SAFETY_MODELS_ENTRIES
+)

llama_stack/providers/remote/inference/bedrock/models.py

@@ -9,6 +9,10 @@ from llama_stack.providers.utils.inference.model_registry import (
     build_hf_repo_model_entry,
 )
 
+SAFETY_MODELS_ENTRIES = []
+
+
+# https://docs.aws.amazon.com/bedrock/latest/userguide/models-supported.html
 MODEL_ENTRIES = [
     build_hf_repo_model_entry(
         "meta.llama3-1-8b-instruct-v1:0",
@@ -22,4 +26,4 @@ MODEL_ENTRIES = [
         "meta.llama3-1-405b-instruct-v1:0",
         CoreModelId.llama3_1_405b_instruct.value,
     ),
-]
+] + SAFETY_MODELS_ENTRIES

llama_stack/providers/remote/inference/cerebras/models.py

@@ -9,6 +9,9 @@ from llama_stack.providers.utils.inference.model_registry import (
     build_hf_repo_model_entry,
 )
 
+SAFETY_MODELS_ENTRIES = []
+
+# https://inference-docs.cerebras.ai/models
 MODEL_ENTRIES = [
     build_hf_repo_model_entry(
         "llama3.1-8b",
@@ -18,4 +21,8 @@ MODEL_ENTRIES = [
         "llama-3.3-70b",
         CoreModelId.llama3_3_70b_instruct.value,
     ),
-]
+    build_hf_repo_model_entry(
+        "llama-4-scout-17b-16e-instruct",
+        CoreModelId.llama4_scout_17b_16e_instruct.value,
+    ),
+] + SAFETY_MODELS_ENTRIES

llama_stack/providers/remote/inference/databricks/databricks.py

@@ -47,7 +47,10 @@ from llama_stack.providers.utils.inference.prompt_adapter import (
 
 from .config import DatabricksImplConfig
 
-model_entries = [
+SAFETY_MODELS_ENTRIES = []
+
+# https://docs.databricks.com/aws/en/machine-learning/model-serving/foundation-model-overview
+MODEL_ENTRIES = [
     build_hf_repo_model_entry(
         "databricks-meta-llama-3-1-70b-instruct",
         CoreModelId.llama3_1_70b_instruct.value,
@@ -56,7 +59,7 @@ model_entries = [
         "databricks-meta-llama-3-1-405b-instruct",
         CoreModelId.llama3_1_405b_instruct.value,
     ),
-]
+] + SAFETY_MODELS_ENTRIES
 
 
 class DatabricksInferenceAdapter(
@@ -66,7 +69,7 @@ class DatabricksInferenceAdapter(
     OpenAICompletionToLlamaStackMixin,
 ):
     def __init__(self, config: DatabricksImplConfig) -> None:
-        ModelRegistryHelper.__init__(self, model_entries=model_entries)
+        ModelRegistryHelper.__init__(self, model_entries=MODEL_ENTRIES)
         self.config = config
 
     async def initialize(self) -> None:

llama_stack/providers/remote/inference/fireworks/models.py

@@ -11,6 +11,17 @@ from llama_stack.providers.utils.inference.model_registry import (
     build_hf_repo_model_entry,
 )
 
+SAFETY_MODELS_ENTRIES = [
+    build_hf_repo_model_entry(
+        "accounts/fireworks/models/llama-guard-3-8b",
+        CoreModelId.llama_guard_3_8b.value,
+    ),
+    build_hf_repo_model_entry(
+        "accounts/fireworks/models/llama-guard-3-11b-vision",
+        CoreModelId.llama_guard_3_11b_vision.value,
+    ),
+]
+
 MODEL_ENTRIES = [
     build_hf_repo_model_entry(
         "accounts/fireworks/models/llama-v3p1-8b-instruct",
@@ -40,14 +51,6 @@ MODEL_ENTRIES = [
         "accounts/fireworks/models/llama-v3p3-70b-instruct",
         CoreModelId.llama3_3_70b_instruct.value,
     ),
-    build_hf_repo_model_entry(
-        "accounts/fireworks/models/llama-guard-3-8b",
-        CoreModelId.llama_guard_3_8b.value,
-    ),
-    build_hf_repo_model_entry(
-        "accounts/fireworks/models/llama-guard-3-11b-vision",
-        CoreModelId.llama_guard_3_11b_vision.value,
-    ),
     build_hf_repo_model_entry(
         "accounts/fireworks/models/llama4-scout-instruct-basic",
         CoreModelId.llama4_scout_17b_16e_instruct.value,
@@ -64,4 +67,4 @@ MODEL_ENTRIES = [
             "context_length": 8192,
         },
     ),
-]
+] + SAFETY_MODELS_ENTRIES

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

@@ -17,11 +17,16 @@ LLM_MODEL_IDS = [
     "gemini/gemini-2.5-pro",
 ]
 
+SAFETY_MODELS_ENTRIES = []
 
-MODEL_ENTRIES = [ProviderModelEntry(provider_model_id=m) for m in LLM_MODEL_IDS] + [
+MODEL_ENTRIES = (
+    [ProviderModelEntry(provider_model_id=m) for m in LLM_MODEL_IDS]
+    + [
         ProviderModelEntry(
             provider_model_id="gemini/text-embedding-004",
             model_type=ModelType.embedding,
             metadata={"embedding_dimension": 768, "context_length": 2048},
         ),
-]
+    ]
+    + SAFETY_MODELS_ENTRIES
+)

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

@@ -38,24 +38,18 @@ class GroqInferenceAdapter(LiteLLMOpenAIMixin):
             provider_data_api_key_field="groq_api_key",
         )
         self.config = config
-        self._openai_client = None
 
     async def initialize(self):
         await super().initialize()
 
     async def shutdown(self):
         await super().shutdown()
-        if self._openai_client:
-            await self._openai_client.close()
-            self._openai_client = None
 
     def _get_openai_client(self) -> AsyncOpenAI:
-        if not self._openai_client:
+        return AsyncOpenAI(
-            self._openai_client = AsyncOpenAI(
             base_url=f"{self.config.url}/openai/v1",
-                api_key=self.config.api_key,
+            api_key=self.get_api_key(),
         )
-        return self._openai_client
 
     async def openai_chat_completion(
         self,

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

@@ -10,6 +10,8 @@ from llama_stack.providers.utils.inference.model_registry import (
     build_model_entry,
 )
 
+SAFETY_MODELS_ENTRIES = []
+
 MODEL_ENTRIES = [
     build_hf_repo_model_entry(
         "groq/llama3-8b-8192",
@@ -51,4 +53,4 @@ MODEL_ENTRIES = [
         "groq/meta-llama/llama-4-maverick-17b-128e-instruct",
         CoreModelId.llama4_maverick_17b_128e_instruct.value,
     ),
-]
+] + SAFETY_MODELS_ENTRIES

llama_stack/providers/remote/inference/nvidia/models.py

@@ -11,6 +11,9 @@ from llama_stack.providers.utils.inference.model_registry import (
     build_hf_repo_model_entry,
 )
 
+SAFETY_MODELS_ENTRIES = []
+
+# https://docs.nvidia.com/nim/large-language-models/latest/supported-llm-agnostic-architectures.html
 MODEL_ENTRIES = [
     build_hf_repo_model_entry(
         "meta/llama3-8b-instruct",
@@ -99,4 +102,4 @@ MODEL_ENTRIES = [
     ),
     # TODO(mf): how do we handle Nemotron models?
     # "Llama3.1-Nemotron-51B-Instruct" -> "meta/llama-3.1-nemotron-51b-instruct",
-]
+] + SAFETY_MODELS_ENTRIES

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

@@ -12,6 +12,19 @@ from llama_stack.providers.utils.inference.model_registry import (
     build_model_entry,
 )
 
+SAFETY_MODELS_ENTRIES = [
+    # The Llama Guard models don't have their full fp16 versions
+    # so we are going to alias their default version to the canonical SKU
+    build_hf_repo_model_entry(
+        "llama-guard3:8b",
+        CoreModelId.llama_guard_3_8b.value,
+    ),
+    build_hf_repo_model_entry(
+        "llama-guard3:1b",
+        CoreModelId.llama_guard_3_1b.value,
+    ),
+]
+
 MODEL_ENTRIES = [
     build_hf_repo_model_entry(
         "llama3.1:8b-instruct-fp16",
@@ -73,16 +86,6 @@ MODEL_ENTRIES = [
         "llama3.3:70b",
         CoreModelId.llama3_3_70b_instruct.value,
     ),
-    # The Llama Guard models don't have their full fp16 versions
-    # so we are going to alias their default version to the canonical SKU
-    build_hf_repo_model_entry(
-        "llama-guard3:8b",
-        CoreModelId.llama_guard_3_8b.value,
-    ),
-    build_hf_repo_model_entry(
-        "llama-guard3:1b",
-        CoreModelId.llama_guard_3_1b.value,
-    ),
     ProviderModelEntry(
         provider_model_id="all-minilm:l6-v2",
         aliases=["all-minilm"],
@@ -100,4 +103,4 @@ MODEL_ENTRIES = [
             "context_length": 8192,
         },
     ),
-]
+] + SAFETY_MODELS_ENTRIES

llama_stack/providers/remote/inference/openai/models.py

@@ -48,9 +48,11 @@ EMBEDDING_MODEL_IDS: dict[str, EmbeddingModelInfo] = {
     "text-embedding-3-small": EmbeddingModelInfo(1536, 8192),
     "text-embedding-3-large": EmbeddingModelInfo(3072, 8192),
 }
+SAFETY_MODELS_ENTRIES = []
 
-
+MODEL_ENTRIES = (
-MODEL_ENTRIES = [ProviderModelEntry(provider_model_id=m) for m in LLM_MODEL_IDS] + [
+    [ProviderModelEntry(provider_model_id=m) for m in LLM_MODEL_IDS]
+    + [
         ProviderModelEntry(
             provider_model_id=model_id,
             model_type=ModelType.embedding,
@@ -60,4 +62,6 @@ MODEL_ENTRIES = [ProviderModelEntry(provider_model_id=m) for m in LLM_MODEL_IDS]
             },
         )
         for model_id, model_info in EMBEDDING_MODEL_IDS.items()
-]
+    ]
+    + SAFETY_MODELS_ENTRIES
+)

llama_stack/providers/remote/inference/openai/openai.py

@@ -59,9 +59,6 @@ class OpenAIInferenceAdapter(LiteLLMOpenAIMixin):
         # if we do not set this, users will be exposed to the
         # litellm specific model names, an abstraction leak.
         self.is_openai_compat = True
-        self._openai_client = AsyncOpenAI(
-            api_key=self.config.api_key,
-        )
 
     async def initialize(self) -> None:
         await super().initialize()
@@ -69,6 +66,11 @@ class OpenAIInferenceAdapter(LiteLLMOpenAIMixin):
     async def shutdown(self) -> None:
         await super().shutdown()
 
+    def _get_openai_client(self) -> AsyncOpenAI:
+        return AsyncOpenAI(
+            api_key=self.get_api_key(),
+        )
+
     async def openai_completion(
         self,
         model: str,
@@ -120,7 +122,7 @@ class OpenAIInferenceAdapter(LiteLLMOpenAIMixin):
             user=user,
             suffix=suffix,
         )
-        return await self._openai_client.completions.create(**params)
+        return await self._get_openai_client().completions.create(**params)
 
     async def openai_chat_completion(
         self,
@@ -176,7 +178,7 @@ class OpenAIInferenceAdapter(LiteLLMOpenAIMixin):
             top_p=top_p,
             user=user,
         )
-        return await self._openai_client.chat.completions.create(**params)
+        return await self._get_openai_client().chat.completions.create(**params)
 
     async def openai_embeddings(
         self,
@@ -204,7 +206,7 @@ class OpenAIInferenceAdapter(LiteLLMOpenAIMixin):
             params["user"] = user
 
         # Call OpenAI embeddings API
-        response = await self._openai_client.embeddings.create(**params)
+        response = await self._get_openai_client().embeddings.create(**params)
 
         data = []
         for i, embedding_data in enumerate(response.data):

llama_stack/providers/remote/inference/runpod/runpod.py

@@ -11,7 +11,7 @@ from llama_stack.apis.inference import *  # noqa: F403
 from llama_stack.apis.inference import OpenAIEmbeddingsResponse
 
 # from llama_stack.providers.datatypes import ModelsProtocolPrivate
-from llama_stack.providers.utils.inference.model_registry import ModelRegistryHelper
+from llama_stack.providers.utils.inference.model_registry import ModelRegistryHelper, build_hf_repo_model_entry
 from llama_stack.providers.utils.inference.openai_compat import (
     OpenAIChatCompletionToLlamaStackMixin,
     OpenAICompletionToLlamaStackMixin,
@@ -25,6 +25,8 @@ from llama_stack.providers.utils.inference.prompt_adapter import (
 
 from .config import RunpodImplConfig
 
+# https://docs.runpod.io/serverless/vllm/overview#compatible-models
+# https://github.com/runpod-workers/worker-vllm/blob/main/README.md#compatible-model-architectures
 RUNPOD_SUPPORTED_MODELS = {
     "Llama3.1-8B": "meta-llama/Llama-3.1-8B",
     "Llama3.1-70B": "meta-llama/Llama-3.1-70B",
@@ -40,6 +42,14 @@ RUNPOD_SUPPORTED_MODELS = {
     "Llama3.2-3B": "meta-llama/Llama-3.2-3B",
 }
 
+SAFETY_MODELS_ENTRIES = []
+
+# Create MODEL_ENTRIES from RUNPOD_SUPPORTED_MODELS for compatibility with starter template
+MODEL_ENTRIES = [
+    build_hf_repo_model_entry(provider_model_id, model_descriptor)
+    for provider_model_id, model_descriptor in RUNPOD_SUPPORTED_MODELS.items()
+] + SAFETY_MODELS_ENTRIES
+
 
 class RunpodInferenceAdapter(
     ModelRegistryHelper,

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

@@ -9,6 +9,14 @@ from llama_stack.providers.utils.inference.model_registry import (
     build_hf_repo_model_entry,
 )
 
+SAFETY_MODELS_ENTRIES = [
+    build_hf_repo_model_entry(
+        "sambanova/Meta-Llama-Guard-3-8B",
+        CoreModelId.llama_guard_3_8b.value,
+    ),
+]
+
+
 MODEL_ENTRIES = [
     build_hf_repo_model_entry(
         "sambanova/Meta-Llama-3.1-8B-Instruct",
@@ -46,8 +54,4 @@ MODEL_ENTRIES = [
         "sambanova/Llama-4-Maverick-17B-128E-Instruct",
         CoreModelId.llama4_maverick_17b_128e_instruct.value,
     ),
-    build_hf_repo_model_entry(
+] + SAFETY_MODELS_ENTRIES
-        "sambanova/Meta-Llama-Guard-3-8B",
-        CoreModelId.llama_guard_3_8b.value,
-    ),
-]

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

@@ -7,6 +7,7 @@
 import json
 from collections.abc import Iterable
 
+import requests
 from openai.types.chat import (
     ChatCompletionAssistantMessageParam as OpenAIChatCompletionAssistantMessage,
 )
@@ -56,6 +57,7 @@ from llama_stack.apis.inference import (
     ToolResponseMessage,
     UserMessage,
 )
+from llama_stack.apis.models import Model
 from llama_stack.log import get_logger
 from llama_stack.models.llama.datatypes import BuiltinTool
 from llama_stack.providers.utils.inference.litellm_openai_mixin import LiteLLMOpenAIMixin
@@ -176,10 +178,11 @@ class SambaNovaInferenceAdapter(LiteLLMOpenAIMixin):
 
     def __init__(self, config: SambaNovaImplConfig):
         self.config = config
+        self.environment_available_models = []
         LiteLLMOpenAIMixin.__init__(
             self,
             model_entries=MODEL_ENTRIES,
-            api_key_from_config=self.config.api_key,
+            api_key_from_config=self.config.api_key.get_secret_value() if self.config.api_key else None,
             provider_data_api_key_field="sambanova_api_key",
         )
 
@@ -246,6 +249,22 @@ class SambaNovaInferenceAdapter(LiteLLMOpenAIMixin):
             **get_sampling_options(request.sampling_params),
         }
 
+    async def register_model(self, model: Model) -> Model:
+        model_id = self.get_provider_model_id(model.provider_resource_id)
+
+        list_models_url = self.config.url + "/models"
+        if len(self.environment_available_models) == 0:
+            try:
+                response = requests.get(list_models_url)
+                response.raise_for_status()
+            except requests.exceptions.RequestException as e:
+                raise RuntimeError(f"Request to {list_models_url} failed") from e
+            self.environment_available_models = [model.get("id") for model in response.json().get("data", {})]
+
+        if model_id.split("sambanova/")[-1] not in self.environment_available_models:
+            logger.warning(f"Model {model_id} not available in {list_models_url}")
+        return model
+
     async def initialize(self):
         await super().initialize()
 

llama_stack/providers/remote/inference/together/models.py

@@ -11,6 +11,16 @@ from llama_stack.providers.utils.inference.model_registry import (
     build_hf_repo_model_entry,
 )
 
+SAFETY_MODELS_ENTRIES = [
+    build_hf_repo_model_entry(
+        "meta-llama/Llama-Guard-3-8B",
+        CoreModelId.llama_guard_3_8b.value,
+    ),
+    build_hf_repo_model_entry(
+        "meta-llama/Llama-Guard-3-11B-Vision-Turbo",
+        CoreModelId.llama_guard_3_11b_vision.value,
+    ),
+]
 MODEL_ENTRIES = [
     build_hf_repo_model_entry(
         "meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo",
@@ -40,14 +50,6 @@ MODEL_ENTRIES = [
         "meta-llama/Llama-3.3-70B-Instruct-Turbo",
         CoreModelId.llama3_3_70b_instruct.value,
     ),
-    build_hf_repo_model_entry(
-        "meta-llama/Meta-Llama-Guard-3-8B",
-        CoreModelId.llama_guard_3_8b.value,
-    ),
-    build_hf_repo_model_entry(
-        "meta-llama/Llama-Guard-3-11B-Vision-Turbo",
-        CoreModelId.llama_guard_3_11b_vision.value,
-    ),
     ProviderModelEntry(
         provider_model_id="togethercomputer/m2-bert-80M-8k-retrieval",
         model_type=ModelType.embedding,
@@ -78,4 +80,4 @@ MODEL_ENTRIES = [
             "together/meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8",
         ],
     ),
-]
+] + SAFETY_MODELS_ENTRIES

llama_stack/providers/remote/inference/together/together.py

@@ -68,19 +68,12 @@ class TogetherInferenceAdapter(ModelRegistryHelper, Inference, NeedsRequestProvi
     def __init__(self, config: TogetherImplConfig) -> None:
         ModelRegistryHelper.__init__(self, MODEL_ENTRIES)
         self.config = config
-        self._client = None
-        self._openai_client = None
 
     async def initialize(self) -> None:
         pass
 
     async def shutdown(self) -> None:
-        if self._client:
+        pass
-            # Together client has no close method, so just set to None
-            self._client = None
-        if self._openai_client:
-            await self._openai_client.close()
-            self._openai_client = None
 
     async def completion(
         self,
@@ -108,7 +101,6 @@ class TogetherInferenceAdapter(ModelRegistryHelper, Inference, NeedsRequestProvi
             return await self._nonstream_completion(request)
 
     def _get_client(self) -> AsyncTogether:
-        if not self._client:
         together_api_key = None
         config_api_key = self.config.api_key.get_secret_value() if self.config.api_key else None
         if config_api_key:
@@ -120,17 +112,14 @@ class TogetherInferenceAdapter(ModelRegistryHelper, Inference, NeedsRequestProvi
                     'Pass Together API Key in the header X-LlamaStack-Provider-Data as { "together_api_key": <your api key>}'
                 )
             together_api_key = provider_data.together_api_key
-            self._client = AsyncTogether(api_key=together_api_key)
+        return AsyncTogether(api_key=together_api_key)
-        return self._client
 
     def _get_openai_client(self) -> AsyncOpenAI:
-        if not self._openai_client:
         together_client = self._get_client().client
-            self._openai_client = AsyncOpenAI(
+        return AsyncOpenAI(
             base_url=together_client.base_url,
             api_key=together_client.api_key,
         )
-        return self._openai_client
 
     async def _nonstream_completion(self, request: CompletionRequest) -> ChatCompletionResponse:
         params = await self._get_params(request)

llama_stack/providers/remote/safety/sambanova/sambanova.py

@@ -33,6 +33,7 @@ CANNED_RESPONSE_TEXT = "I can't answer that. Can I help with something else?"
 class SambaNovaSafetyAdapter(Safety, ShieldsProtocolPrivate, NeedsRequestProviderData):
     def __init__(self, config: SambaNovaSafetyConfig) -> None:
         self.config = config
+        self.environment_available_models = []
 
     async def initialize(self) -> None:
         pass
@@ -54,18 +55,18 @@ class SambaNovaSafetyAdapter(Safety, ShieldsProtocolPrivate, NeedsRequestProvide
 
     async def register_shield(self, shield: Shield) -> None:
         list_models_url = self.config.url + "/models"
+        if len(self.environment_available_models) == 0:
             try:
                 response = requests.get(list_models_url)
                 response.raise_for_status()
             except requests.exceptions.RequestException as e:
                 raise RuntimeError(f"Request to {list_models_url} failed") from e
-        available_models = [model.get("id") for model in response.json().get("data", {})]
+            self.environment_available_models = [model.get("id") for model in response.json().get("data", {})]
         if (
-            len(available_models) == 0
+            "guard" not in shield.provider_resource_id.lower()
-            or "guard" not in shield.provider_resource_id.lower()
+            or shield.provider_resource_id.split("sambanova/")[-1] not in self.environment_available_models
-            or shield.provider_resource_id.split("sambanova/")[-1] not in available_models
         ):
-            raise ValueError(f"Shield {shield.provider_resource_id} not found in SambaNova")
+            logger.warning(f"Shield {shield.provider_resource_id} not available in {list_models_url}")
 
     async def run_shield(
         self, shield_id: str, messages: list[Message], params: dict[str, Any] | None = None

llama_stack/providers/remote/vector_io/milvus/milvus.py

@@ -61,6 +61,11 @@ class MilvusIndex(EmbeddingIndex):
         self.consistency_level = consistency_level
         self.kvstore = kvstore
 
+    async def initialize(self):
+        # MilvusIndex does not require explicit initialization
+        # TODO: could move collection creation into initialization but it is not really necessary
+        pass
+
     async def delete(self):
         if await asyncio.to_thread(self.client.has_collection, self.collection_name):
             await asyncio.to_thread(self.client.drop_collection, collection_name=self.collection_name)
@@ -174,7 +179,8 @@ class MilvusVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolP
             uri = os.path.expanduser(self.config.db_path)
             self.client = MilvusClient(uri=uri)
 
-        self.openai_vector_stores = await self._load_openai_vector_stores()
+        # Load existing OpenAI vector stores into the in-memory cache
+        await self.initialize_openai_vector_stores()
 
     async def shutdown(self) -> None:
         self.client.close()
@@ -199,6 +205,9 @@ class MilvusVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolP
         if vector_db_id in self.cache:
             return self.cache[vector_db_id]
 
+        if self.vector_db_store is None:
+            raise ValueError(f"Vector DB {vector_db_id} not found")
+
         vector_db = await self.vector_db_store.get_vector_db(vector_db_id)
         if not vector_db:
             raise ValueError(f"Vector DB {vector_db_id} not found")
@@ -240,36 +249,6 @@ class MilvusVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolP
 
         return await index.query_chunks(query, params)
 
-    async def _save_openai_vector_store(self, store_id: str, store_info: dict[str, Any]) -> None:
-        """Save vector store metadata to persistent storage."""
-        assert self.kvstore is not None
-        key = f"{OPENAI_VECTOR_STORES_PREFIX}{store_id}"
-        await self.kvstore.set(key=key, value=json.dumps(store_info))
-        self.openai_vector_stores[store_id] = store_info
-
-    async def _update_openai_vector_store(self, store_id: str, store_info: dict[str, Any]) -> None:
-        """Update vector store metadata in persistent storage."""
-        assert self.kvstore is not None
-        key = f"{OPENAI_VECTOR_STORES_PREFIX}{store_id}"
-        await self.kvstore.set(key=key, value=json.dumps(store_info))
-        self.openai_vector_stores[store_id] = store_info
-
-    async def _delete_openai_vector_store_from_storage(self, store_id: str) -> None:
-        """Delete vector store metadata from persistent storage."""
-        assert self.kvstore is not None
-        key = f"{OPENAI_VECTOR_STORES_PREFIX}{store_id}"
-        await self.kvstore.delete(key)
-        if store_id in self.openai_vector_stores:
-            del self.openai_vector_stores[store_id]
-
-    async def _load_openai_vector_stores(self) -> dict[str, dict[str, Any]]:
-        """Load all vector store metadata from persistent storage."""
-        assert self.kvstore is not None
-        start_key = OPENAI_VECTOR_STORES_PREFIX
-        end_key = f"{OPENAI_VECTOR_STORES_PREFIX}\xff"
-        stored = await self.kvstore.values_in_range(start_key, end_key)
-        return {json.loads(s)["id"]: json.loads(s) for s in stored}
-
     async def _save_openai_vector_store_file(
         self, store_id: str, file_id: str, file_info: dict[str, Any], file_contents: list[dict[str, Any]]
     ) -> None:

llama_stack/providers/remote/vector_io/pgvector/config.py

@@ -8,6 +8,10 @@ from typing import Any
 
 from pydantic import BaseModel, Field
 
+from llama_stack.providers.utils.kvstore.config import (
+    KVStoreConfig,
+    SqliteKVStoreConfig,
+)
 from llama_stack.schema_utils import json_schema_type
 
 
@@ -18,10 +22,12 @@ class PGVectorVectorIOConfig(BaseModel):
     db: str | None = Field(default="postgres")
     user: str | None = Field(default="postgres")
     password: str | None = Field(default="mysecretpassword")
+    kvstore: KVStoreConfig | None = Field(description="Config for KV store backend (SQLite only for now)", default=None)
 
     @classmethod
     def sample_run_config(
         cls,
+        __distro_dir__: str,
         host: str = "${env.PGVECTOR_HOST:=localhost}",
         port: int = "${env.PGVECTOR_PORT:=5432}",
         db: str = "${env.PGVECTOR_DB}",
@@ -29,4 +35,14 @@ class PGVectorVectorIOConfig(BaseModel):
         password: str = "${env.PGVECTOR_PASSWORD}",
         **kwargs: Any,
     ) -> dict[str, Any]:
-        return {"host": host, "port": port, "db": db, "user": user, "password": password}
+        return {
+            "host": host,
+            "port": port,
+            "db": db,
+            "user": user,
+            "password": password,
+            "kvstore": SqliteKVStoreConfig.sample_run_config(
+                __distro_dir__=__distro_dir__,
+                db_name="pgvector_registry.db",
+            ),
+        }

llama_stack/providers/remote/vector_io/pgvector/pgvector.py

@@ -13,24 +13,18 @@ from psycopg2 import sql
 from psycopg2.extras import Json, execute_values
 from pydantic import BaseModel, TypeAdapter
 
+from llama_stack.apis.files.files import Files
 from llama_stack.apis.inference import InterleavedContent
 from llama_stack.apis.vector_dbs import VectorDB
 from llama_stack.apis.vector_io import (
     Chunk,
     QueryChunksResponse,
-    SearchRankingOptions,
     VectorIO,
-    VectorStoreChunkingStrategy,
-    VectorStoreDeleteResponse,
-    VectorStoreFileContentsResponse,
-    VectorStoreFileObject,
-    VectorStoreFileStatus,
-    VectorStoreListFilesResponse,
-    VectorStoreListResponse,
-    VectorStoreObject,
-    VectorStoreSearchResponsePage,
 )
 from llama_stack.providers.datatypes import Api, VectorDBsProtocolPrivate
+from llama_stack.providers.utils.kvstore import kvstore_impl
+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 (
     EmbeddingIndex,
     VectorDBWithIndex,
@@ -40,6 +34,13 @@ from .config import PGVectorVectorIOConfig
 
 log = logging.getLogger(__name__)
 
+VERSION = "v3"
+VECTOR_DBS_PREFIX = f"vector_dbs:pgvector:{VERSION}::"
+VECTOR_INDEX_PREFIX = f"vector_index:pgvector:{VERSION}::"
+OPENAI_VECTOR_STORES_PREFIX = f"openai_vector_stores:pgvector:{VERSION}::"
+OPENAI_VECTOR_STORES_FILES_PREFIX = f"openai_vector_stores_files:pgvector:{VERSION}::"
+OPENAI_VECTOR_STORES_FILES_CONTENTS_PREFIX = f"openai_vector_stores_files_contents:pgvector:{VERSION}::"
+
 
 def check_extension_version(cur):
     cur.execute("SELECT extversion FROM pg_extension WHERE extname = 'vector'")
@@ -69,7 +70,7 @@ def load_models(cur, cls):
 
 
 class PGVectorIndex(EmbeddingIndex):
-    def __init__(self, vector_db: VectorDB, dimension: int, conn):
+    def __init__(self, vector_db: VectorDB, dimension: int, conn, kvstore: KVStore | None = None):
         self.conn = conn
         with conn.cursor(cursor_factory=psycopg2.extras.DictCursor) as cur:
             # Sanitize the table name by replacing hyphens with underscores
@@ -77,6 +78,7 @@ class PGVectorIndex(EmbeddingIndex):
             # when created with patterns like "test-vector-db-{uuid4()}"
             sanitized_identifier = vector_db.identifier.replace("-", "_")
             self.table_name = f"vector_store_{sanitized_identifier}"
+            self.kvstore = kvstore
 
             cur.execute(
                 f"""
@@ -158,15 +160,28 @@ class PGVectorIndex(EmbeddingIndex):
             cur.execute(f"DROP TABLE IF EXISTS {self.table_name}")
 
 
-class PGVectorVectorIOAdapter(VectorIO, VectorDBsProtocolPrivate):
+class PGVectorVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolPrivate):
-    def __init__(self, config: PGVectorVectorIOConfig, inference_api: Api.inference) -> None:
+    def __init__(
+        self,
+        config: PGVectorVectorIOConfig,
+        inference_api: Api.inference,
+        files_api: Files | None = None,
+    ) -> None:
         self.config = config
         self.inference_api = inference_api
         self.conn = None
         self.cache = {}
+        self.files_api = files_api
+        self.kvstore: KVStore | None = None
+        self.vector_db_store = None
+        self.openai_vector_store: dict[str, dict[str, Any]] = {}
+        self.metadatadata_collection_name = "openai_vector_stores_metadata"
 
     async def initialize(self) -> None:
         log.info(f"Initializing PGVector memory adapter with config: {self.config}")
+        self.kvstore = await kvstore_impl(self.config.kvstore)
+        await self.initialize_openai_vector_stores()
+
         try:
             self.conn = psycopg2.connect(
                 host=self.config.host,
@@ -201,15 +216,32 @@ class PGVectorVectorIOAdapter(VectorIO, VectorDBsProtocolPrivate):
             log.info("Connection to PGVector database server closed")
 
     async def register_vector_db(self, vector_db: VectorDB) -> None:
+        # Persist vector DB metadata in the KV store
+        assert self.kvstore is not None
+        key = f"{VECTOR_DBS_PREFIX}{vector_db.identifier}"
+        await self.kvstore.set(key=key, value=vector_db.model_dump_json())
+
+        # Upsert model metadata in Postgres
         upsert_models(self.conn, [(vector_db.identifier, vector_db)])
 
-        index = PGVectorIndex(vector_db, vector_db.embedding_dimension, self.conn)
+        # Create and cache the PGVector index table for the vector DB
-        self.cache[vector_db.identifier] = VectorDBWithIndex(vector_db, index, self.inference_api)
+        index = VectorDBWithIndex(
+            vector_db,
+            index=PGVectorIndex(vector_db, vector_db.embedding_dimension, self.conn, kvstore=self.kvstore),
+            inference_api=self.inference_api,
+        )
+        self.cache[vector_db.identifier] = index
 
     async def unregister_vector_db(self, vector_db_id: str) -> None:
+        # Remove provider index and cache
+        if vector_db_id in self.cache:
             await self.cache[vector_db_id].index.delete()
             del self.cache[vector_db_id]
 
+        # Delete vector DB metadata from KV store
+        assert self.kvstore is not None
+        await self.kvstore.delete(key=f"{VECTOR_DBS_PREFIX}{vector_db_id}")
+
     async def insert_chunks(
         self,
         vector_db_id: str,
@@ -237,107 +269,20 @@ class PGVectorVectorIOAdapter(VectorIO, VectorDBsProtocolPrivate):
         self.cache[vector_db_id] = VectorDBWithIndex(vector_db, index, self.inference_api)
         return self.cache[vector_db_id]
 
-    async def openai_create_vector_store(
+    # OpenAI Vector Stores File operations are not supported in PGVector
-        self,
+    async def _save_openai_vector_store_file(
-        name: str,
+        self, store_id: str, file_id: str, file_info: dict[str, Any], file_contents: list[dict[str, Any]]
-        file_ids: list[str] | None = None,
+    ) -> None:
-        expires_after: dict[str, Any] | None = None,
-        chunking_strategy: dict[str, Any] | None = None,
-        metadata: dict[str, Any] | None = None,
-        embedding_model: str | None = None,
-        embedding_dimension: int | None = 384,
-        provider_id: str | None = None,
-        provider_vector_db_id: str | None = None,
-    ) -> VectorStoreObject:
         raise NotImplementedError("OpenAI Vector Stores API is not supported in PGVector")
 
-    async def openai_list_vector_stores(
+    async def _load_openai_vector_store_file(self, store_id: str, file_id: str) -> dict[str, Any]:
-        self,
-        limit: int | None = 20,
-        order: str | None = "desc",
-        after: str | None = None,
-        before: str | None = None,
-    ) -> VectorStoreListResponse:
         raise NotImplementedError("OpenAI Vector Stores API is not supported in PGVector")
 
-    async def openai_retrieve_vector_store(
+    async def _load_openai_vector_store_file_contents(self, store_id: str, file_id: str) -> list[dict[str, Any]]:
-        self,
-        vector_store_id: str,
-    ) -> VectorStoreObject:
         raise NotImplementedError("OpenAI Vector Stores API is not supported in PGVector")
 
-    async def openai_update_vector_store(
+    async def _update_openai_vector_store_file(self, store_id: str, file_id: str, file_info: dict[str, Any]) -> None:
-        self,
-        vector_store_id: str,
-        name: str | None = None,
-        expires_after: dict[str, Any] | None = None,
-        metadata: dict[str, Any] | None = None,
-    ) -> VectorStoreObject:
         raise NotImplementedError("OpenAI Vector Stores API is not supported in PGVector")
 
-    async def openai_delete_vector_store(
+    async def _delete_openai_vector_store_file_from_storage(self, store_id: str, file_id: str) -> None:
-        self,
-        vector_store_id: str,
-    ) -> VectorStoreDeleteResponse:
-        raise NotImplementedError("OpenAI Vector Stores API is not supported in PGVector")
-
-    async def openai_search_vector_store(
-        self,
-        vector_store_id: str,
-        query: str | list[str],
-        filters: dict[str, Any] | None = None,
-        max_num_results: int | None = 10,
-        ranking_options: SearchRankingOptions | None = None,
-        rewrite_query: bool | None = False,
-        search_mode: str | None = "vector",
-    ) -> VectorStoreSearchResponsePage:
-        raise NotImplementedError("OpenAI Vector Stores API is not supported in PGVector")
-
-    async def openai_attach_file_to_vector_store(
-        self,
-        vector_store_id: str,
-        file_id: str,
-        attributes: dict[str, Any] | None = None,
-        chunking_strategy: VectorStoreChunkingStrategy | None = None,
-    ) -> VectorStoreFileObject:
-        raise NotImplementedError("OpenAI Vector Stores API is not supported in PGVector")
-
-    async def openai_list_files_in_vector_store(
-        self,
-        vector_store_id: str,
-        limit: int | None = 20,
-        order: str | None = "desc",
-        after: str | None = None,
-        before: str | None = None,
-        filter: VectorStoreFileStatus | None = None,
-    ) -> VectorStoreListFilesResponse:
-        raise NotImplementedError("OpenAI Vector Stores API is not supported in PGVector")
-
-    async def openai_retrieve_vector_store_file(
-        self,
-        vector_store_id: str,
-        file_id: str,
-    ) -> VectorStoreFileObject:
-        raise NotImplementedError("OpenAI Vector Stores API is not supported in PGVector")
-
-    async def openai_retrieve_vector_store_file_contents(
-        self,
-        vector_store_id: str,
-        file_id: str,
-    ) -> VectorStoreFileContentsResponse:
-        raise NotImplementedError("OpenAI Vector Stores API is not supported in PGVector")
-
-    async def openai_update_vector_store_file(
-        self,
-        vector_store_id: str,
-        file_id: str,
-        attributes: dict[str, Any] | None = None,
-    ) -> VectorStoreFileObject:
-        raise NotImplementedError("OpenAI Vector Stores API is not supported in PGVector")
-
-    async def openai_delete_vector_store_file(
-        self,
-        vector_store_id: str,
-        file_id: str,
-    ) -> VectorStoreFileObject:
         raise NotImplementedError("OpenAI Vector Stores API is not supported in PGVector")

llama_stack/providers/remote/vector_io/weaviate/config.py

@@ -6,15 +6,26 @@
 
 from typing import Any
 
-from pydantic import BaseModel
+from pydantic import BaseModel, Field
+
+from llama_stack.providers.utils.kvstore.config import (
+    KVStoreConfig,
+    SqliteKVStoreConfig,
+)
 
 
 class WeaviateRequestProviderData(BaseModel):
     weaviate_api_key: str
     weaviate_cluster_url: str
+    kvstore: KVStoreConfig | None = Field(description="Config for KV store backend (SQLite only for now)", default=None)
 
 
 class WeaviateVectorIOConfig(BaseModel):
     @classmethod
-    def sample_run_config(cls, **kwargs: Any) -> dict[str, Any]:
+    def sample_run_config(cls, __distro_dir__: str, **kwargs: Any) -> dict[str, Any]:
-        return {}
+        return {
+            "kvstore": SqliteKVStoreConfig.sample_run_config(
+                __distro_dir__=__distro_dir__,
+                db_name="weaviate_registry.db",
+            ),
+        }

llama_stack/providers/remote/vector_io/weaviate/weaviate.py

@@ -14,10 +14,13 @@ from weaviate.classes.init import Auth
 from weaviate.classes.query import Filter
 
 from llama_stack.apis.common.content_types import InterleavedContent
+from llama_stack.apis.files.files import Files
 from llama_stack.apis.vector_dbs import VectorDB
 from llama_stack.apis.vector_io import Chunk, QueryChunksResponse, VectorIO
 from llama_stack.distribution.request_headers import NeedsRequestProviderData
 from llama_stack.providers.datatypes import Api, VectorDBsProtocolPrivate
+from llama_stack.providers.utils.kvstore import kvstore_impl
+from llama_stack.providers.utils.kvstore.api import KVStore
 from llama_stack.providers.utils.memory.vector_store import (
     EmbeddingIndex,
     VectorDBWithIndex,
@@ -27,11 +30,19 @@ from .config import WeaviateRequestProviderData, WeaviateVectorIOConfig
 
 log = logging.getLogger(__name__)
 
+VERSION = "v3"
+VECTOR_DBS_PREFIX = f"vector_dbs:weaviate:{VERSION}::"
+VECTOR_INDEX_PREFIX = f"vector_index:weaviate:{VERSION}::"
+OPENAI_VECTOR_STORES_PREFIX = f"openai_vector_stores:weaviate:{VERSION}::"
+OPENAI_VECTOR_STORES_FILES_PREFIX = f"openai_vector_stores_files:weaviate:{VERSION}::"
+OPENAI_VECTOR_STORES_FILES_CONTENTS_PREFIX = f"openai_vector_stores_files_contents:weaviate:{VERSION}::"
+
 
 class WeaviateIndex(EmbeddingIndex):
-    def __init__(self, client: weaviate.Client, collection_name: str):
+    def __init__(self, client: weaviate.Client, collection_name: str, kvstore: KVStore | None = None):
         self.client = client
         self.collection_name = collection_name
+        self.kvstore = kvstore
 
     async def add_chunks(self, chunks: list[Chunk], embeddings: NDArray):
         assert len(chunks) == len(embeddings), (
@@ -109,11 +120,21 @@ class WeaviateVectorIOAdapter(
     NeedsRequestProviderData,
     VectorDBsProtocolPrivate,
 ):
-    def __init__(self, config: WeaviateVectorIOConfig, inference_api: Api.inference) -> None:
+    def __init__(
+        self,
+        config: WeaviateVectorIOConfig,
+        inference_api: Api.inference,
+        files_api: Files | None,
+    ) -> None:
         self.config = config
         self.inference_api = inference_api
         self.client_cache = {}
         self.cache = {}
+        self.files_api = files_api
+        self.kvstore: KVStore | None = None
+        self.vector_db_store = None
+        self.openai_vector_stores: dict[str, dict[str, Any]] = {}
+        self.metadata_collection_name = "openai_vector_stores_metadata"
 
     def _get_client(self) -> weaviate.Client:
         provider_data = self.get_request_provider_data()
@@ -132,7 +153,26 @@ class WeaviateVectorIOAdapter(
         return client
 
     async def initialize(self) -> None:
-        pass
+        """Set up KV store and load existing vector DBs and OpenAI vector stores."""
+        # Initialize KV store for metadata
+        self.kvstore = await kvstore_impl(self.config.kvstore)
+
+        # Load existing vector DB definitions
+        start_key = VECTOR_DBS_PREFIX
+        end_key = f"{VECTOR_DBS_PREFIX}\xff"
+        stored = await self.kvstore.values_in_range(start_key, end_key)
+        for raw in stored:
+            vector_db = VectorDB.model_validate_json(raw)
+            client = self._get_client()
+            idx = WeaviateIndex(client=client, collection_name=vector_db.identifier, kvstore=self.kvstore)
+            self.cache[vector_db.identifier] = VectorDBWithIndex(
+                vector_db=vector_db,
+                index=idx,
+                inference_api=self.inference_api,
+            )
+
+        # Load OpenAI vector stores metadata into cache
+        await self.initialize_openai_vector_stores()
 
     async def shutdown(self) -> None:
         for client in self.client_cache.values():
@@ -206,3 +246,21 @@ class WeaviateVectorIOAdapter(
             raise ValueError(f"Vector DB {vector_db_id} not found")
 
         return await index.query_chunks(query, params)
+
+    # OpenAI Vector Stores File operations are not supported in Weaviate
+    async def _save_openai_vector_store_file(
+        self, store_id: str, file_id: str, file_info: dict[str, Any], file_contents: list[dict[str, Any]]
+    ) -> None:
+        raise NotImplementedError("OpenAI Vector Stores API is not supported in Weaviate")
+
+    async def _load_openai_vector_store_file(self, store_id: str, file_id: str) -> dict[str, Any]:
+        raise NotImplementedError("OpenAI Vector Stores API is not supported in Weaviate")
+
+    async def _load_openai_vector_store_file_contents(self, store_id: str, file_id: str) -> list[dict[str, Any]]:
+        raise NotImplementedError("OpenAI Vector Stores API is not supported in Weaviate")
+
+    async def _update_openai_vector_store_file(self, store_id: str, file_id: str, file_info: dict[str, Any]) -> None:
+        raise NotImplementedError("OpenAI Vector Stores API is not supported in Weaviate")
+
+    async def _delete_openai_vector_store_file_from_storage(self, store_id: str, file_id: str) -> None:
+        raise NotImplementedError("OpenAI Vector Stores API is not supported in Weaviate")

llama_stack/providers/utils/inference/model_registry.py

@@ -44,6 +44,7 @@ def build_hf_repo_model_entry(
     ]
     if additional_aliases:
         aliases.extend(additional_aliases)
+    aliases = [alias for alias in aliases if alias is not None]
     return ProviderModelEntry(
         provider_model_id=provider_model_id,
         aliases=aliases,
@@ -82,35 +83,35 @@ class ModelRegistryHelper(ModelsProtocolPrivate):
     def get_llama_model(self, provider_model_id: str) -> str | None:
         return self.provider_id_to_llama_model_map.get(provider_model_id, None)
 
-    async def query_available_models(self) -> list[str]:
+    async def check_model_availability(self, model: str) -> bool:
         """
-        Return a list of available models.
+        Check if a specific model is available from the provider (non-static check).
 
-        This is for subclassing purposes, so providers can lookup a list of
+        This is for subclassing purposes, so providers can check if a specific
-        of currently available models.
+        model is currently available for use through dynamic means (e.g., API calls).
 
-        This is combined with the statically configured model entries in
+        This method should NOT check statically configured model entries in
-        `self.alias_to_provider_id_map` to determine which models are
+        `self.alias_to_provider_id_map` - that is handled separately in register_model.
-        available for registration.
 
-        Default implementation returns no models.
+        Default implementation returns False (no dynamic models available).
 
-        :return: A list of model identifiers (provider_model_ids).
+        :param model: The model identifier to check.
+        :return: True if the model is available dynamically, False otherwise.
         """
-        return []
+        return False
 
     async def register_model(self, model: Model) -> Model:
         # Check if model is supported in static configuration
         supported_model_id = self.get_provider_model_id(model.provider_resource_id)
 
-        # If not found in static config, check if it's available from provider
+        # If not found in static config, check if it's available dynamically from provider
         if not supported_model_id:
-            available_models = await self.query_available_models()
+            if await self.check_model_availability(model.provider_resource_id):
-            if model.provider_resource_id in available_models:
                 supported_model_id = model.provider_resource_id
             else:
-                # Combine static and dynamic models for error message
+                # note: we cannot provide a complete list of supported models without
-                all_supported_models = list(self.alias_to_provider_id_map.keys()) + available_models
+                #       getting a complete list from the provider, so we return "..."
+                all_supported_models = [*self.alias_to_provider_id_map.keys(), "..."]
                 raise UnsupportedModelError(model.provider_resource_id, all_supported_models)
 
         provider_resource_id = self.get_provider_model_id(model.model_id)
@@ -118,7 +119,7 @@ class ModelRegistryHelper(ModelsProtocolPrivate):
             # embedding models are always registered by their provider model id and does not need to be mapped to a llama model
             provider_resource_id = model.provider_resource_id
         if provider_resource_id:
-            if provider_resource_id != supported_model_id:  # be idemopotent, only reject differences
+            if provider_resource_id != supported_model_id:  # be idempotent, only reject differences
                 raise ValueError(
                     f"Model id '{model.model_id}' is already registered. Please use a different id or unregister it first."
                 )

llama_stack/providers/utils/memory/openai_vector_store_mixin.py

@@ -5,6 +5,7 @@
 # the root directory of this source tree.
 
 import asyncio
+import json
 import logging
 import mimetypes
 import time
@@ -35,6 +36,7 @@ from llama_stack.apis.vector_io import (
     VectorStoreSearchResponse,
     VectorStoreSearchResponsePage,
 )
+from llama_stack.providers.utils.kvstore.api import KVStore
 from llama_stack.providers.utils.memory.vector_store import content_from_data_and_mime_type, make_overlapped_chunks
 
 logger = logging.getLogger(__name__)
@@ -59,26 +61,45 @@ class OpenAIVectorStoreMixin(ABC):
     # These should be provided by the implementing class
     openai_vector_stores: dict[str, dict[str, Any]]
     files_api: Files | None
+    # KV store for persisting OpenAI vector store metadata
+    kvstore: KVStore | None
 
-    @abstractmethod
     async def _save_openai_vector_store(self, store_id: str, store_info: dict[str, Any]) -> None:
         """Save vector store metadata to persistent storage."""
-        pass
+        assert self.kvstore is not None
+        key = f"{OPENAI_VECTOR_STORES_PREFIX}{store_id}"
+        await self.kvstore.set(key=key, value=json.dumps(store_info))
+        # update in-memory cache
+        self.openai_vector_stores[store_id] = store_info
 
-    @abstractmethod
     async def _load_openai_vector_stores(self) -> dict[str, dict[str, Any]]:
         """Load all vector store metadata from persistent storage."""
-        pass
+        assert self.kvstore is not None
+        start_key = OPENAI_VECTOR_STORES_PREFIX
+        end_key = f"{OPENAI_VECTOR_STORES_PREFIX}\xff"
+        stored_data = await self.kvstore.values_in_range(start_key, end_key)
+
+        stores: dict[str, dict[str, Any]] = {}
+        for item in stored_data:
+            info = json.loads(item)
+            stores[info["id"]] = info
+        return stores
 
-    @abstractmethod
     async def _update_openai_vector_store(self, store_id: str, store_info: dict[str, Any]) -> None:
         """Update vector store metadata in persistent storage."""
-        pass
+        assert self.kvstore is not None
+        key = f"{OPENAI_VECTOR_STORES_PREFIX}{store_id}"
+        await self.kvstore.set(key=key, value=json.dumps(store_info))
+        # update in-memory cache
+        self.openai_vector_stores[store_id] = store_info
 
-    @abstractmethod
     async def _delete_openai_vector_store_from_storage(self, store_id: str) -> None:
         """Delete vector store metadata from persistent storage."""
-        pass
+        assert self.kvstore is not None
+        key = f"{OPENAI_VECTOR_STORES_PREFIX}{store_id}"
+        await self.kvstore.delete(key)
+        # remove from in-memory cache
+        self.openai_vector_stores.pop(store_id, None)
 
     @abstractmethod
     async def _save_openai_vector_store_file(
@@ -117,6 +138,10 @@ class OpenAIVectorStoreMixin(ABC):
         """Unregister a vector database (provider-specific implementation)."""
         pass
 
+    async def initialize_openai_vector_stores(self) -> None:
+        """Load existing OpenAI vector stores into the in-memory cache."""
+        self.openai_vector_stores = await self._load_openai_vector_stores()
+
     @abstractmethod
     async def insert_chunks(
         self,

llama_stack/providers/utils/sqlstore/authorized_sqlstore.py

@@ -39,22 +39,10 @@ SQL_OPTIMIZED_POLICY = [
 
 
 class SqlRecord(ProtectedResource):
-    """Simple ProtectedResource implementation for SQL records."""
+    def __init__(self, record_id: str, table_name: str, owner: User):
-
-    def __init__(self, record_id: str, table_name: str, access_attributes: dict[str, list[str]] | None = None):
         self.type = f"sql_record::{table_name}"
         self.identifier = record_id
-
+        self.owner = owner
-        if access_attributes:
-            self.owner = User(
-                principal="system",
-                attributes=access_attributes,
-            )
-        else:
-            self.owner = User(
-                principal="system_public",
-                attributes=None,
-            )
 
 
 class AuthorizedSqlStore:
@@ -101,22 +89,27 @@ class AuthorizedSqlStore:
 
     async def create_table(self, table: str, schema: Mapping[str, ColumnType | ColumnDefinition]) -> None:
         """Create a table with built-in access control support."""
-        await self.sql_store.add_column_if_not_exists(table, "access_attributes", ColumnType.JSON)
 
         enhanced_schema = dict(schema)
         if "access_attributes" not in enhanced_schema:
             enhanced_schema["access_attributes"] = ColumnType.JSON
+        if "owner_principal" not in enhanced_schema:
+            enhanced_schema["owner_principal"] = ColumnType.STRING
 
         await self.sql_store.create_table(table, enhanced_schema)
+        await self.sql_store.add_column_if_not_exists(table, "access_attributes", ColumnType.JSON)
+        await self.sql_store.add_column_if_not_exists(table, "owner_principal", ColumnType.STRING)
 
     async def insert(self, table: str, data: Mapping[str, Any]) -> None:
         """Insert a row with automatic access control attribute capture."""
         enhanced_data = dict(data)
 
         current_user = get_authenticated_user()
-        if current_user and current_user.attributes:
+        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.insert(table, enhanced_data)
@@ -146,9 +139,12 @@ class AuthorizedSqlStore:
 
         for row in rows.data:
             stored_access_attrs = row.get("access_attributes")
+            stored_owner_principal = row.get("owner_principal") or ""
 
             record_id = row.get("id", "unknown")
-            sql_record = SqlRecord(str(record_id), table, stored_access_attrs)
+            sql_record = SqlRecord(
+                str(record_id), table, User(principal=stored_owner_principal, attributes=stored_access_attrs)
+            )
 
             if is_action_allowed(policy, Action.READ, sql_record, current_user):
                 filtered_rows.append(row)
@@ -186,8 +182,10 @@ class AuthorizedSqlStore:
         Only applies SQL filtering for the default policy to ensure correctness.
         For custom policies, uses conservative filtering to avoid blocking legitimate access.
         """
+        current_user = get_authenticated_user()
+
         if not policy or policy == SQL_OPTIMIZED_POLICY:
-            return self._build_default_policy_where_clause()
+            return self._build_default_policy_where_clause(current_user)
         else:
             return self._build_conservative_where_clause()
 
@@ -227,29 +225,27 @@ class AuthorizedSqlStore:
 
     def _get_public_access_conditions(self) -> list[str]:
         """Get the SQL conditions for public access."""
+        # Public records are records that have no owner_principal or access_attributes
+        conditions = ["owner_principal = ''"]
         if self.database_type == SqlStoreType.postgres:
             # Postgres stores JSON null as 'null'
-            return ["access_attributes::text = 'null'"]
+            conditions.append("access_attributes::text = 'null'")
         elif self.database_type == SqlStoreType.sqlite:
-            return ["access_attributes = 'null'"]
+            conditions.append("access_attributes = 'null'")
         else:
             raise ValueError(f"Unsupported database type: {self.database_type}")
+        return conditions
 
-    def _build_default_policy_where_clause(self) -> str:
+    def _build_default_policy_where_clause(self, current_user: User | None) -> str:
         """Build SQL WHERE clause for the default policy.
 
         Default policy: permit all actions when user in owners [roles, teams, projects, namespaces]
         This means user must match ALL attribute categories that exist in the resource.
         """
-        current_user = get_authenticated_user()
-
         base_conditions = self._get_public_access_conditions()
-        if not current_user or not current_user.attributes:
-            # Only allow public records
-            return f"({' OR '.join(base_conditions)})"
-        else:
         user_attr_conditions = []
 
+        if current_user and current_user.attributes:
             for attr_key, user_values in current_user.attributes.items():
                 if user_values:
                     value_conditions = []

llama_stack/providers/utils/sqlstore/sqlalchemy_sqlstore.py

@@ -244,16 +244,22 @@ class SqlAlchemySqlStoreImpl(SqlStore):
         engine = create_async_engine(self.config.engine_str)
 
         try:
-            inspector = inspect(engine)
+            async with engine.begin() as conn:
+
+                def check_column_exists(sync_conn):
+                    inspector = inspect(sync_conn)
 
                     table_names = inspector.get_table_names()
                     if table not in table_names:
-                return
+                        return False, False  # table doesn't exist, column doesn't exist
 
                     existing_columns = inspector.get_columns(table)
                     column_names = [col["name"] for col in existing_columns]
 
-            if column_name in column_names:
+                    return True, column_name in column_names  # table exists, column exists or not
+
+                table_exists, column_exists = await conn.run_sync(check_column_exists)
+                if not table_exists or column_exists:
                     return
 
                 sqlalchemy_type = TYPE_MAPPING.get(column_type)
@@ -269,10 +275,10 @@ class SqlAlchemySqlStoreImpl(SqlStore):
                 nullable_clause = "" if nullable else " NOT NULL"
                 add_column_sql = text(f"ALTER TABLE {table} ADD COLUMN {column_name} {compiled_type}{nullable_clause}")
 
-            async with engine.begin() as conn:
                 await conn.execute(add_column_sql)
 
-        except Exception:
+        except Exception as e:
             # If any error occurs during migration, log it but don't fail
             # The table creation will handle adding the column
+            logger.error(f"Error adding column {column_name} to table {table}: {e}")
             pass

llama_stack/providers/utils/telemetry/trace_protocol.py

@@ -9,14 +9,12 @@ import inspect
 import json
 from collections.abc import AsyncGenerator, Callable
 from functools import wraps
-from typing import Any, TypeVar
+from typing import Any
 
 from pydantic import BaseModel
 
 from llama_stack.models.llama.datatypes import Primitive
 
-T = TypeVar("T")
-
 
 def serialize_value(value: Any) -> Primitive:
     return str(_prepare_for_json(value))
@@ -44,7 +42,7 @@ def _prepare_for_json(value: Any) -> str:
             return str(value)
 
 
-def trace_protocol(cls: type[T]) -> type[T]:
+def trace_protocol[T](cls: type[T]) -> type[T]:
     """
     A class decorator that automatically traces all methods in a protocol/base class
     and its inheriting classes.

llama_stack/templates/nvidia/nvidia.py

@@ -68,7 +68,7 @@ def get_distribution_template() -> DistributionTemplate:
         ),
     ]
 
-    default_models = get_model_registry(available_models)
+    default_models, _ = get_model_registry(available_models)
     return DistributionTemplate(
         name="nvidia",
         distro_type="self_hosted",

llama_stack/templates/open-benchmark/open_benchmark.py

@@ -128,6 +128,7 @@ def get_distribution_template() -> DistributionTemplate:
             provider_id="${env.ENABLE_PGVECTOR:+pgvector}",
             provider_type="remote::pgvector",
             config=PGVectorVectorIOConfig.sample_run_config(
+                f"~/.llama/distributions/{name}",
                 db="${env.PGVECTOR_DB:=}",
                 user="${env.PGVECTOR_USER:=}",
                 password="${env.PGVECTOR_PASSWORD:=}",
@@ -146,7 +147,8 @@ def get_distribution_template() -> DistributionTemplate:
         ),
     ]
 
-    default_models = get_model_registry(available_models) + [
+    models, _ = get_model_registry(available_models)
+    default_models = models + [
         ModelInput(
             model_id="meta-llama/Llama-3.3-70B-Instruct",
             provider_id="groq",

llama_stack/templates/open-benchmark/run.yaml

@@ -39,6 +39,9 @@ providers:
     provider_type: inline::sqlite-vec
     config:
       db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/open-benchmark}/sqlite_vec.db
+      kvstore:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/open-benchmark}/sqlite_vec_registry.db
   - provider_id: ${env.ENABLE_CHROMADB:+chromadb}
     provider_type: remote::chromadb
     config:
@@ -51,6 +54,9 @@ providers:
       db: ${env.PGVECTOR_DB:=}
       user: ${env.PGVECTOR_USER:=}
       password: ${env.PGVECTOR_PASSWORD:=}
+      kvstore:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/open-benchmark}/pgvector_registry.db
   safety:
   - provider_id: llama-guard
     provider_type: inline::llama-guard

llama_stack/templates/starter/run.yaml

@@ -144,6 +144,9 @@ providers:
     provider_type: inline::sqlite-vec
     config:
       db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/sqlite_vec.db
+      kvstore:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/sqlite_vec_registry.db
   - provider_id: ${env.ENABLE_MILVUS:=__disabled__}
     provider_type: inline::milvus
     config:
@@ -163,6 +166,9 @@ providers:
       db: ${env.PGVECTOR_DB:=}
       user: ${env.PGVECTOR_USER:=}
       password: ${env.PGVECTOR_PASSWORD:=}
+      kvstore:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/pgvector_registry.db
   files:
   - provider_id: meta-reference-files
     provider_type: inline::localfs
@@ -256,11 +262,46 @@ inference_store:
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/inference_store.db
 models:
+- metadata: {}
+  model_id: ${env.ENABLE_CEREBRAS:=__disabled__}/llama3.1-8b
+  provider_id: ${env.ENABLE_CEREBRAS:=__disabled__}
+  provider_model_id: llama3.1-8b
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_CEREBRAS:=__disabled__}/meta-llama/Llama-3.1-8B-Instruct
+  provider_id: ${env.ENABLE_CEREBRAS:=__disabled__}
+  provider_model_id: llama3.1-8b
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_CEREBRAS:=__disabled__}/llama-3.3-70b
+  provider_id: ${env.ENABLE_CEREBRAS:=__disabled__}
+  provider_model_id: llama-3.3-70b
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_CEREBRAS:=__disabled__}/meta-llama/Llama-3.3-70B-Instruct
+  provider_id: ${env.ENABLE_CEREBRAS:=__disabled__}
+  provider_model_id: llama-3.3-70b
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_CEREBRAS:=__disabled__}/llama-4-scout-17b-16e-instruct
+  provider_id: ${env.ENABLE_CEREBRAS:=__disabled__}
+  provider_model_id: llama-4-scout-17b-16e-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_CEREBRAS:=__disabled__}/meta-llama/Llama-4-Scout-17B-16E-Instruct
+  provider_id: ${env.ENABLE_CEREBRAS:=__disabled__}
+  provider_model_id: llama-4-scout-17b-16e-instruct
+  model_type: llm
 - metadata: {}
   model_id: ${env.ENABLE_OLLAMA:=__disabled__}/${env.OLLAMA_INFERENCE_MODEL:=__disabled__}
   provider_id: ${env.ENABLE_OLLAMA:=__disabled__}
   provider_model_id: ${env.OLLAMA_INFERENCE_MODEL:=__disabled__}
   model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_OLLAMA:=__disabled__}/${env.SAFETY_MODEL:=__disabled__}
+  provider_id: ${env.ENABLE_OLLAMA:=__disabled__}
+  provider_model_id: ${env.SAFETY_MODEL:=__disabled__}
+  model_type: llm
 - metadata:
     embedding_dimension: ${env.OLLAMA_EMBEDDING_DIMENSION:=384}
   model_id: ${env.ENABLE_OLLAMA:=__disabled__}/${env.OLLAMA_EMBEDDING_MODEL:=__disabled__}
@@ -342,26 +383,6 @@ models:
   provider_id: ${env.ENABLE_FIREWORKS:=__disabled__}
   provider_model_id: accounts/fireworks/models/llama-v3p3-70b-instruct
   model_type: llm
-- metadata: {}
-  model_id: ${env.ENABLE_FIREWORKS:=__disabled__}/accounts/fireworks/models/llama-guard-3-8b
-  provider_id: ${env.ENABLE_FIREWORKS:=__disabled__}
-  provider_model_id: accounts/fireworks/models/llama-guard-3-8b
-  model_type: llm
-- metadata: {}
-  model_id: ${env.ENABLE_FIREWORKS:=__disabled__}/meta-llama/Llama-Guard-3-8B
-  provider_id: ${env.ENABLE_FIREWORKS:=__disabled__}
-  provider_model_id: accounts/fireworks/models/llama-guard-3-8b
-  model_type: llm
-- metadata: {}
-  model_id: ${env.ENABLE_FIREWORKS:=__disabled__}/accounts/fireworks/models/llama-guard-3-11b-vision
-  provider_id: ${env.ENABLE_FIREWORKS:=__disabled__}
-  provider_model_id: accounts/fireworks/models/llama-guard-3-11b-vision
-  model_type: llm
-- metadata: {}
-  model_id: ${env.ENABLE_FIREWORKS:=__disabled__}/meta-llama/Llama-Guard-3-11B-Vision
-  provider_id: ${env.ENABLE_FIREWORKS:=__disabled__}
-  provider_model_id: accounts/fireworks/models/llama-guard-3-11b-vision
-  model_type: llm
 - metadata: {}
   model_id: ${env.ENABLE_FIREWORKS:=__disabled__}/accounts/fireworks/models/llama4-scout-instruct-basic
   provider_id: ${env.ENABLE_FIREWORKS:=__disabled__}
@@ -389,6 +410,26 @@ models:
   provider_id: ${env.ENABLE_FIREWORKS:=__disabled__}
   provider_model_id: nomic-ai/nomic-embed-text-v1.5
   model_type: embedding
+- metadata: {}
+  model_id: ${env.ENABLE_FIREWORKS:=__disabled__}/accounts/fireworks/models/llama-guard-3-8b
+  provider_id: ${env.ENABLE_FIREWORKS:=__disabled__}
+  provider_model_id: accounts/fireworks/models/llama-guard-3-8b
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_FIREWORKS:=__disabled__}/meta-llama/Llama-Guard-3-8B
+  provider_id: ${env.ENABLE_FIREWORKS:=__disabled__}
+  provider_model_id: accounts/fireworks/models/llama-guard-3-8b
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_FIREWORKS:=__disabled__}/accounts/fireworks/models/llama-guard-3-11b-vision
+  provider_id: ${env.ENABLE_FIREWORKS:=__disabled__}
+  provider_model_id: accounts/fireworks/models/llama-guard-3-11b-vision
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_FIREWORKS:=__disabled__}/meta-llama/Llama-Guard-3-11B-Vision
+  provider_id: ${env.ENABLE_FIREWORKS:=__disabled__}
+  provider_model_id: accounts/fireworks/models/llama-guard-3-11b-vision
+  model_type: llm
 - metadata: {}
   model_id: ${env.ENABLE_TOGETHER:=__disabled__}/meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo
   provider_id: ${env.ENABLE_TOGETHER:=__disabled__}
@@ -459,26 +500,6 @@ models:
   provider_id: ${env.ENABLE_TOGETHER:=__disabled__}
   provider_model_id: meta-llama/Llama-3.3-70B-Instruct-Turbo
   model_type: llm
-- metadata: {}
-  model_id: ${env.ENABLE_TOGETHER:=__disabled__}/meta-llama/Meta-Llama-Guard-3-8B
-  provider_id: ${env.ENABLE_TOGETHER:=__disabled__}
-  provider_model_id: meta-llama/Meta-Llama-Guard-3-8B
-  model_type: llm
-- metadata: {}
-  model_id: ${env.ENABLE_TOGETHER:=__disabled__}/meta-llama/Llama-Guard-3-8B
-  provider_id: ${env.ENABLE_TOGETHER:=__disabled__}
-  provider_model_id: meta-llama/Meta-Llama-Guard-3-8B
-  model_type: llm
-- metadata: {}
-  model_id: ${env.ENABLE_TOGETHER:=__disabled__}/meta-llama/Llama-Guard-3-11B-Vision-Turbo
-  provider_id: ${env.ENABLE_TOGETHER:=__disabled__}
-  provider_model_id: meta-llama/Llama-Guard-3-11B-Vision-Turbo
-  model_type: llm
-- metadata: {}
-  model_id: ${env.ENABLE_TOGETHER:=__disabled__}/meta-llama/Llama-Guard-3-11B-Vision
-  provider_id: ${env.ENABLE_TOGETHER:=__disabled__}
-  provider_model_id: meta-llama/Llama-Guard-3-11B-Vision-Turbo
-  model_type: llm
 - metadata:
     embedding_dimension: 768
     context_length: 8192
@@ -523,6 +544,264 @@ models:
   provider_id: ${env.ENABLE_TOGETHER:=__disabled__}
   provider_model_id: meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8
   model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_TOGETHER:=__disabled__}/meta-llama/Llama-Guard-3-8B
+  provider_id: ${env.ENABLE_TOGETHER:=__disabled__}
+  provider_model_id: meta-llama/Llama-Guard-3-8B
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_TOGETHER:=__disabled__}/meta-llama/Llama-Guard-3-8B
+  provider_id: ${env.ENABLE_TOGETHER:=__disabled__}
+  provider_model_id: meta-llama/Llama-Guard-3-8B
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_TOGETHER:=__disabled__}/meta-llama/Llama-Guard-3-11B-Vision-Turbo
+  provider_id: ${env.ENABLE_TOGETHER:=__disabled__}
+  provider_model_id: meta-llama/Llama-Guard-3-11B-Vision-Turbo
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_TOGETHER:=__disabled__}/meta-llama/Llama-Guard-3-11B-Vision
+  provider_id: ${env.ENABLE_TOGETHER:=__disabled__}
+  provider_model_id: meta-llama/Llama-Guard-3-11B-Vision-Turbo
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_BEDROCK:=__disabled__}/meta.llama3-1-8b-instruct-v1:0
+  provider_id: ${env.ENABLE_BEDROCK:=__disabled__}
+  provider_model_id: meta.llama3-1-8b-instruct-v1:0
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_BEDROCK:=__disabled__}/meta-llama/Llama-3.1-8B-Instruct
+  provider_id: ${env.ENABLE_BEDROCK:=__disabled__}
+  provider_model_id: meta.llama3-1-8b-instruct-v1:0
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_BEDROCK:=__disabled__}/meta.llama3-1-70b-instruct-v1:0
+  provider_id: ${env.ENABLE_BEDROCK:=__disabled__}
+  provider_model_id: meta.llama3-1-70b-instruct-v1:0
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_BEDROCK:=__disabled__}/meta-llama/Llama-3.1-70B-Instruct
+  provider_id: ${env.ENABLE_BEDROCK:=__disabled__}
+  provider_model_id: meta.llama3-1-70b-instruct-v1:0
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_BEDROCK:=__disabled__}/meta.llama3-1-405b-instruct-v1:0
+  provider_id: ${env.ENABLE_BEDROCK:=__disabled__}
+  provider_model_id: meta.llama3-1-405b-instruct-v1:0
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_BEDROCK:=__disabled__}/meta-llama/Llama-3.1-405B-Instruct-FP8
+  provider_id: ${env.ENABLE_BEDROCK:=__disabled__}
+  provider_model_id: meta.llama3-1-405b-instruct-v1:0
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_DATABRICKS:=__disabled__}/databricks-meta-llama-3-1-70b-instruct
+  provider_id: ${env.ENABLE_DATABRICKS:=__disabled__}
+  provider_model_id: databricks-meta-llama-3-1-70b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_DATABRICKS:=__disabled__}/meta-llama/Llama-3.1-70B-Instruct
+  provider_id: ${env.ENABLE_DATABRICKS:=__disabled__}
+  provider_model_id: databricks-meta-llama-3-1-70b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_DATABRICKS:=__disabled__}/databricks-meta-llama-3-1-405b-instruct
+  provider_id: ${env.ENABLE_DATABRICKS:=__disabled__}
+  provider_model_id: databricks-meta-llama-3-1-405b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_DATABRICKS:=__disabled__}/meta-llama/Llama-3.1-405B-Instruct-FP8
+  provider_id: ${env.ENABLE_DATABRICKS:=__disabled__}
+  provider_model_id: databricks-meta-llama-3-1-405b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta/llama3-8b-instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama3-8b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta-llama/Llama-3-8B-Instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama3-8b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta/llama3-70b-instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama3-70b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta-llama/Llama-3-70B-Instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama3-70b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta/llama-3.1-8b-instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama-3.1-8b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta-llama/Llama-3.1-8B-Instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama-3.1-8b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta/llama-3.1-70b-instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama-3.1-70b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta-llama/Llama-3.1-70B-Instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama-3.1-70b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta/llama-3.1-405b-instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama-3.1-405b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta-llama/Llama-3.1-405B-Instruct-FP8
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama-3.1-405b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta/llama-3.2-1b-instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama-3.2-1b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta-llama/Llama-3.2-1B-Instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama-3.2-1b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta/llama-3.2-3b-instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama-3.2-3b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta-llama/Llama-3.2-3B-Instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama-3.2-3b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta/llama-3.2-11b-vision-instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama-3.2-11b-vision-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta-llama/Llama-3.2-11B-Vision-Instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama-3.2-11b-vision-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta/llama-3.2-90b-vision-instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama-3.2-90b-vision-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta-llama/Llama-3.2-90B-Vision-Instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama-3.2-90b-vision-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta/llama-3.3-70b-instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama-3.3-70b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/meta-llama/Llama-3.3-70B-Instruct
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: meta/llama-3.3-70b-instruct
+  model_type: llm
+- metadata:
+    embedding_dimension: 2048
+    context_length: 8192
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/nvidia/llama-3.2-nv-embedqa-1b-v2
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: nvidia/llama-3.2-nv-embedqa-1b-v2
+  model_type: embedding
+- metadata:
+    embedding_dimension: 1024
+    context_length: 512
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/nvidia/nv-embedqa-e5-v5
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: nvidia/nv-embedqa-e5-v5
+  model_type: embedding
+- metadata:
+    embedding_dimension: 4096
+    context_length: 512
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/nvidia/nv-embedqa-mistral-7b-v2
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: nvidia/nv-embedqa-mistral-7b-v2
+  model_type: embedding
+- metadata:
+    embedding_dimension: 1024
+    context_length: 512
+  model_id: ${env.ENABLE_NVIDIA:=__disabled__}/snowflake/arctic-embed-l
+  provider_id: ${env.ENABLE_NVIDIA:=__disabled__}
+  provider_model_id: snowflake/arctic-embed-l
+  model_type: embedding
+- metadata: {}
+  model_id: ${env.ENABLE_RUNPOD:=__disabled__}/Llama3.1-8B
+  provider_id: ${env.ENABLE_RUNPOD:=__disabled__}
+  provider_model_id: Llama3.1-8B
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_RUNPOD:=__disabled__}/Llama3.1-70B
+  provider_id: ${env.ENABLE_RUNPOD:=__disabled__}
+  provider_model_id: Llama3.1-70B
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_RUNPOD:=__disabled__}/Llama3.1-405B:bf16-mp8
+  provider_id: ${env.ENABLE_RUNPOD:=__disabled__}
+  provider_model_id: Llama3.1-405B:bf16-mp8
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_RUNPOD:=__disabled__}/Llama3.1-405B
+  provider_id: ${env.ENABLE_RUNPOD:=__disabled__}
+  provider_model_id: Llama3.1-405B
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_RUNPOD:=__disabled__}/Llama3.1-405B:bf16-mp16
+  provider_id: ${env.ENABLE_RUNPOD:=__disabled__}
+  provider_model_id: Llama3.1-405B:bf16-mp16
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_RUNPOD:=__disabled__}/Llama3.1-8B-Instruct
+  provider_id: ${env.ENABLE_RUNPOD:=__disabled__}
+  provider_model_id: Llama3.1-8B-Instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_RUNPOD:=__disabled__}/Llama3.1-70B-Instruct
+  provider_id: ${env.ENABLE_RUNPOD:=__disabled__}
+  provider_model_id: Llama3.1-70B-Instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_RUNPOD:=__disabled__}/Llama3.1-405B-Instruct:bf16-mp8
+  provider_id: ${env.ENABLE_RUNPOD:=__disabled__}
+  provider_model_id: Llama3.1-405B-Instruct:bf16-mp8
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_RUNPOD:=__disabled__}/Llama3.1-405B-Instruct
+  provider_id: ${env.ENABLE_RUNPOD:=__disabled__}
+  provider_model_id: Llama3.1-405B-Instruct
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_RUNPOD:=__disabled__}/Llama3.1-405B-Instruct:bf16-mp16
+  provider_id: ${env.ENABLE_RUNPOD:=__disabled__}
+  provider_model_id: Llama3.1-405B-Instruct:bf16-mp16
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_RUNPOD:=__disabled__}/Llama3.2-1B
+  provider_id: ${env.ENABLE_RUNPOD:=__disabled__}
+  provider_model_id: Llama3.2-1B
+  model_type: llm
+- metadata: {}
+  model_id: ${env.ENABLE_RUNPOD:=__disabled__}/Llama3.2-3B
+  provider_id: ${env.ENABLE_RUNPOD:=__disabled__}
+  provider_model_id: Llama3.2-3B
+  model_type: llm
 - metadata: {}
   model_id: ${env.ENABLE_OPENAI:=__disabled__}/openai/gpt-4o
   provider_id: ${env.ENABLE_OPENAI:=__disabled__}
@@ -894,7 +1173,9 @@ models:
   model_id: all-MiniLM-L6-v2
   provider_id: ${env.ENABLE_SENTENCE_TRANSFORMERS:=sentence-transformers}
   model_type: embedding
-shields: []
+shields:
+- shield_id: ${env.SAFETY_MODEL:=__disabled__}
+  provider_shield_id: ${env.ENABLE_OLLAMA:=__disabled__}/${env.SAFETY_MODEL:=__disabled__}
 vector_dbs: []
 datasets: []
 scoring_fns: []

llama_stack/templates/starter/starter.py

@@ -31,6 +31,15 @@ from llama_stack.providers.registry.inference import available_providers
 from llama_stack.providers.remote.inference.anthropic.models import (
     MODEL_ENTRIES as ANTHROPIC_MODEL_ENTRIES,
 )
+from llama_stack.providers.remote.inference.bedrock.models import (
+    MODEL_ENTRIES as BEDROCK_MODEL_ENTRIES,
+)
+from llama_stack.providers.remote.inference.cerebras.models import (
+    MODEL_ENTRIES as CEREBRAS_MODEL_ENTRIES,
+)
+from llama_stack.providers.remote.inference.databricks.databricks import (
+    MODEL_ENTRIES as DATABRICKS_MODEL_ENTRIES,
+)
 from llama_stack.providers.remote.inference.fireworks.models import (
     MODEL_ENTRIES as FIREWORKS_MODEL_ENTRIES,
 )
@@ -40,9 +49,15 @@ from llama_stack.providers.remote.inference.gemini.models import (
 from llama_stack.providers.remote.inference.groq.models import (
     MODEL_ENTRIES as GROQ_MODEL_ENTRIES,
 )
+from llama_stack.providers.remote.inference.nvidia.models import (
+    MODEL_ENTRIES as NVIDIA_MODEL_ENTRIES,
+)
 from llama_stack.providers.remote.inference.openai.models import (
     MODEL_ENTRIES as OPENAI_MODEL_ENTRIES,
 )
+from llama_stack.providers.remote.inference.runpod.runpod import (
+    MODEL_ENTRIES as RUNPOD_MODEL_ENTRIES,
+)
 from llama_stack.providers.remote.inference.sambanova.models import (
     MODEL_ENTRIES as SAMBANOVA_MODEL_ENTRIES,
 )
@@ -59,6 +74,7 @@ from llama_stack.templates.template import (
     DistributionTemplate,
     RunConfigSettings,
     get_model_registry,
+    get_shield_registry,
 )
 
 
@@ -72,6 +88,11 @@ def _get_model_entries_for_provider(provider_type: str) -> list[ProviderModelEnt
         "gemini": GEMINI_MODEL_ENTRIES,
         "groq": GROQ_MODEL_ENTRIES,
         "sambanova": SAMBANOVA_MODEL_ENTRIES,
+        "cerebras": CEREBRAS_MODEL_ENTRIES,
+        "bedrock": BEDROCK_MODEL_ENTRIES,
+        "databricks": DATABRICKS_MODEL_ENTRIES,
+        "nvidia": NVIDIA_MODEL_ENTRIES,
+        "runpod": RUNPOD_MODEL_ENTRIES,
     }
 
     # Special handling for providers with dynamic model entries
@@ -81,6 +102,10 @@ def _get_model_entries_for_provider(provider_type: str) -> list[ProviderModelEnt
                 provider_model_id="${env.OLLAMA_INFERENCE_MODEL:=__disabled__}",
                 model_type=ModelType.llm,
             ),
+            ProviderModelEntry(
+                provider_model_id="${env.SAFETY_MODEL:=__disabled__}",
+                model_type=ModelType.llm,
+            ),
             ProviderModelEntry(
                 provider_model_id="${env.OLLAMA_EMBEDDING_MODEL:=__disabled__}",
                 model_type=ModelType.embedding,
@@ -100,6 +125,20 @@ def _get_model_entries_for_provider(provider_type: str) -> list[ProviderModelEnt
     return model_entries_map.get(provider_type, [])
 
 
+def _get_model_safety_entries_for_provider(provider_type: str) -> list[ProviderModelEntry]:
+    """Get model entries for a specific provider type."""
+    safety_model_entries_map = {
+        "ollama": [
+            ProviderModelEntry(
+                provider_model_id="${env.SAFETY_MODEL:=__disabled__}",
+                model_type=ModelType.llm,
+            ),
+        ],
+    }
+
+    return safety_model_entries_map.get(provider_type, [])
+
+
 def _get_config_for_provider(provider_spec: ProviderSpec) -> dict[str, Any]:
     """Get configuration for a provider using its adapter's config class."""
     config_class = instantiate_class_type(provider_spec.config_class)
@@ -155,6 +194,23 @@ def get_remote_inference_providers() -> tuple[list[Provider], dict[str, list[Pro
     return inference_providers, available_models
 
 
+# build a list of shields for all possible providers
+def get_safety_models_for_providers(providers: list[Provider]) -> dict[str, list[ProviderModelEntry]]:
+    available_models = {}
+    for provider in providers:
+        provider_type = provider.provider_type.split("::")[1]
+        safety_model_entries = _get_model_safety_entries_for_provider(provider_type)
+        if len(safety_model_entries) == 0:
+            continue
+
+        env_var = f"ENABLE_{provider_type.upper().replace('-', '_').replace('::', '_')}"
+        provider_id = f"${{env.{env_var}:=__disabled__}}"
+
+        available_models[provider_id] = safety_model_entries
+
+    return available_models
+
+
 def get_distribution_template() -> DistributionTemplate:
     remote_inference_providers, available_models = get_remote_inference_providers()
 
@@ -185,6 +241,7 @@ def get_distribution_template() -> DistributionTemplate:
             provider_id="${env.ENABLE_PGVECTOR:=__disabled__}",
             provider_type="remote::pgvector",
             config=PGVectorVectorIOConfig.sample_run_config(
+                f"~/.llama/distributions/{name}",
                 db="${env.PGVECTOR_DB:=}",
                 user="${env.PGVECTOR_USER:=}",
                 password="${env.PGVECTOR_PASSWORD:=}",
@@ -244,7 +301,10 @@ def get_distribution_template() -> DistributionTemplate:
         },
     )
 
-    default_models = get_model_registry(available_models)
+    default_models, ids_conflict_in_models = get_model_registry(available_models)
+
+    available_safety_models = get_safety_models_for_providers(remote_inference_providers)
+    shields = get_shield_registry(available_safety_models, ids_conflict_in_models)
 
     return DistributionTemplate(
         name=name,
@@ -266,9 +326,7 @@ def get_distribution_template() -> DistributionTemplate:
                 default_models=default_models + [embedding_model],
                 default_tool_groups=default_tool_groups,
                 # TODO: add a way to enable/disable shields on the fly
-                # default_shields=[
+                default_shields=shields,
-                #     ShieldInput(provider_id="llama-guard", shield_id="${env.SAFETY_MODEL:=meta-llama/Llama-Guard-3-8B}")
-                # ],
             ),
         },
         run_config_env_vars={

llama_stack/templates/template.py

@@ -37,7 +37,7 @@ from llama_stack.providers.utils.sqlstore.sqlstore import get_pip_packages as ge
 
 def get_model_registry(
     available_models: dict[str, list[ProviderModelEntry]],
-) -> list[ModelInput]:
+) -> tuple[list[ModelInput], bool]:
     models = []
 
     # check for conflicts in model ids
@@ -74,7 +74,50 @@ def get_model_registry(
                         metadata=entry.metadata,
                     )
                 )
-    return models
+    return models, ids_conflict
+
+
+def get_shield_registry(
+    available_safety_models: dict[str, list[ProviderModelEntry]],
+    ids_conflict_in_models: bool,
+) -> list[ShieldInput]:
+    shields = []
+
+    # check for conflicts in shield ids
+    all_ids = set()
+    ids_conflict = False
+
+    for _, entries in available_safety_models.items():
+        for entry in entries:
+            ids = [entry.provider_model_id] + entry.aliases
+            for model_id in ids:
+                if model_id in all_ids:
+                    ids_conflict = True
+                    rich.print(
+                        f"[yellow]Shield id {model_id} conflicts; all shield ids will be prefixed with provider id[/yellow]"
+                    )
+                    break
+            all_ids.update(ids)
+            if ids_conflict:
+                break
+        if ids_conflict:
+            break
+
+    for provider_id, entries in available_safety_models.items():
+        for entry in entries:
+            ids = [entry.provider_model_id] + entry.aliases
+            for model_id in ids:
+                identifier = f"{provider_id}/{model_id}" if ids_conflict and provider_id not in model_id else model_id
+                shields.append(
+                    ShieldInput(
+                        shield_id=identifier,
+                        provider_shield_id=f"{provider_id}/{entry.provider_model_id}"
+                        if ids_conflict_in_models
+                        else entry.provider_model_id,
+                    )
+                )
+
+    return shields
 
 
 class DefaultModel(BaseModel):

llama_stack/templates/watsonx/watsonx.py

@@ -69,7 +69,7 @@ def get_distribution_template() -> DistributionTemplate:
         },
     )
 
-    default_models = get_model_registry(available_models)
+    default_models, _ = get_model_registry(available_models)
     return DistributionTemplate(
         name="watsonx",
         distro_type="remote_hosted",

llama_stack/ui/app/logs/vector-stores/[id]/page.tsx

@@ -0,0 +1,82 @@
+"use client";
+
+import { useEffect, useState } from "react";
+import { useParams, useRouter } from "next/navigation";
+import { useAuthClient } from "@/hooks/use-auth-client";
+import type { VectorStore } from "llama-stack-client/resources/vector-stores/vector-stores";
+import type { VectorStoreFile } from "llama-stack-client/resources/vector-stores/files";
+import { VectorStoreDetailView } from "@/components/vector-stores/vector-store-detail";
+
+export default function VectorStoreDetailPage() {
+  const params = useParams();
+  const id = params.id as string;
+  const client = useAuthClient();
+  const router = useRouter();
+
+  const [store, setStore] = useState<VectorStore | null>(null);
+  const [files, setFiles] = useState<VectorStoreFile[]>([]);
+  const [isLoadingStore, setIsLoadingStore] = useState(true);
+  const [isLoadingFiles, setIsLoadingFiles] = useState(true);
+  const [errorStore, setErrorStore] = useState<Error | null>(null);
+  const [errorFiles, setErrorFiles] = useState<Error | null>(null);
+
+  useEffect(() => {
+    if (!id) {
+      setErrorStore(new Error("Vector Store ID is missing."));
+      setIsLoadingStore(false);
+      return;
+    }
+    const fetchStore = async () => {
+      setIsLoadingStore(true);
+      setErrorStore(null);
+      try {
+        const response = await client.vectorStores.retrieve(id);
+        setStore(response as VectorStore);
+      } catch (err) {
+        setErrorStore(
+          err instanceof Error
+            ? err
+            : new Error("Failed to load vector store."),
+        );
+      } finally {
+        setIsLoadingStore(false);
+      }
+    };
+    fetchStore();
+  }, [id, client]);
+
+  useEffect(() => {
+    if (!id) {
+      setErrorFiles(new Error("Vector Store ID is missing."));
+      setIsLoadingFiles(false);
+      return;
+    }
+    const fetchFiles = async () => {
+      setIsLoadingFiles(true);
+      setErrorFiles(null);
+      try {
+        const result = await client.vectorStores.files.list(id as any);
+        setFiles((result as any).data);
+      } catch (err) {
+        setErrorFiles(
+          err instanceof Error ? err : new Error("Failed to load files."),
+        );
+      } finally {
+        setIsLoadingFiles(false);
+      }
+    };
+    fetchFiles();
+  }, [id]);
+
+  return (
+    <VectorStoreDetailView
+      store={store}
+      files={files}
+      isLoadingStore={isLoadingStore}
+      isLoadingFiles={isLoadingFiles}
+      errorStore={errorStore}
+      errorFiles={errorFiles}
+      id={id}
+    />
+  );
+}

llama_stack/ui/app/logs/vector-stores/layout.tsx

@@ -0,0 +1,16 @@
+"use client";
+
+import React from "react";
+import LogsLayout from "@/components/layout/logs-layout";
+
+export default function VectorStoresLayout({
+  children,
+}: {
+  children: React.ReactNode;
+}) {
+  return (
+    <LogsLayout sectionLabel="Vector Stores" basePath="/logs/vector-stores">
+      {children}
+    </LogsLayout>
+  );
+}

llama_stack/ui/app/logs/vector-stores/page.tsx

@@ -0,0 +1,121 @@
+"use client";
+
+import React from "react";
+import { useAuthClient } from "@/hooks/use-auth-client";
+import type {
+  ListVectorStoresResponse,
+  VectorStore,
+} from "llama-stack-client/resources/vector-stores/vector-stores";
+import { useRouter } from "next/navigation";
+import { usePagination } from "@/hooks/use-pagination";
+import {
+  Table,
+  TableBody,
+  TableCaption,
+  TableCell,
+  TableHead,
+  TableHeader,
+  TableRow,
+} from "@/components/ui/table";
+import { Skeleton } from "@/components/ui/skeleton";
+
+export default function VectorStoresPage() {
+  const client = useAuthClient();
+  const router = useRouter();
+  const {
+    data: stores,
+    status,
+    hasMore,
+    error,
+    loadMore,
+  } = usePagination<VectorStore>({
+    limit: 20,
+    order: "desc",
+    fetchFunction: async (client, params) => {
+      const response = await client.vectorStores.list({
+        after: params.after,
+        limit: params.limit,
+        order: params.order,
+      } as any);
+      return response as ListVectorStoresResponse;
+    },
+    errorMessagePrefix: "vector stores",
+  });
+
+  // Auto-load all pages for infinite scroll behavior (like Responses)
+  React.useEffect(() => {
+    if (status === "idle" && hasMore) {
+      loadMore();
+    }
+  }, [status, hasMore, loadMore]);
+
+  if (status === "loading") {
+    return (
+      <div className="space-y-2">
+        <Skeleton className="h-8 w-full" />
+        <Skeleton className="h-4 w-full" />
+        <Skeleton className="h-4 w-full" />
+      </div>
+    );
+  }
+
+  if (status === "error") {
+    return <div className="text-destructive">Error: {error?.message}</div>;
+  }
+
+  if (!stores || stores.length === 0) {
+    return <p>No vector stores found.</p>;
+  }
+
+  return (
+    <div className="overflow-auto flex-1 min-h-0">
+      <Table>
+        <TableHeader>
+          <TableRow>
+            <TableHead>ID</TableHead>
+            <TableHead>Name</TableHead>
+            <TableHead>Created</TableHead>
+            <TableHead>Completed</TableHead>
+            <TableHead>Cancelled</TableHead>
+            <TableHead>Failed</TableHead>
+            <TableHead>In Progress</TableHead>
+            <TableHead>Total</TableHead>
+            <TableHead>Usage Bytes</TableHead>
+            <TableHead>Provider ID</TableHead>
+            <TableHead>Provider Vector DB ID</TableHead>
+          </TableRow>
+        </TableHeader>
+        <TableBody>
+          {stores.map((store) => {
+            const fileCounts = store.file_counts;
+            const metadata = store.metadata || {};
+            const providerId = metadata.provider_id ?? "";
+            const providerDbId = metadata.provider_vector_db_id ?? "";
+
+            return (
+              <TableRow
+                key={store.id}
+                onClick={() => router.push(`/logs/vector-stores/${store.id}`)}
+                className="cursor-pointer hover:bg-muted/50"
+              >
+                <TableCell>{store.id}</TableCell>
+                <TableCell>{store.name}</TableCell>
+                <TableCell>
+                  {new Date(store.created_at * 1000).toLocaleString()}
+                </TableCell>
+                <TableCell>{fileCounts.completed}</TableCell>
+                <TableCell>{fileCounts.cancelled}</TableCell>
+                <TableCell>{fileCounts.failed}</TableCell>
+                <TableCell>{fileCounts.in_progress}</TableCell>
+                <TableCell>{fileCounts.total}</TableCell>
+                <TableCell>{store.usage_bytes}</TableCell>
+                <TableCell>{providerId}</TableCell>
+                <TableCell>{providerDbId}</TableCell>
+              </TableRow>
+            );
+          })}
+        </TableBody>
+      </Table>
+    </div>
+  );
+}

llama_stack/ui/components/layout/app-sidebar.tsx

@@ -1,6 +1,11 @@
 "use client";
 
-import { MessageSquareText, MessagesSquare, MoveUpRight } from "lucide-react";
+import {
+  MessageSquareText,
+  MessagesSquare,
+  MoveUpRight,
+  Database,
+} from "lucide-react";
 import Link from "next/link";
 import { usePathname } from "next/navigation";
 import { cn } from "@/lib/utils";
@@ -28,6 +33,11 @@ const logItems = [
     url: "/logs/responses",
     icon: MessagesSquare,
   },
+  {
+    title: "Vector Stores",
+    url: "/logs/vector-stores",
+    icon: Database,
+  },
   {
     title: "Documentation",
     url: "https://llama-stack.readthedocs.io/en/latest/references/api_reference/index.html",
@@ -57,13 +67,13 @@ export function AppSidebar() {
                       className={cn(
                         "justify-start",
                         isActive &&
-                          "bg-gray-200 hover:bg-gray-200 text-primary hover:text-primary",
+                          "bg-gray-200 dark:bg-gray-700 hover:bg-gray-200 dark:hover:bg-gray-700 text-gray-900 dark:text-gray-100",
                       )}
                     >
                       <Link href={item.url}>
                         <item.icon
                           className={cn(
-                            isActive && "text-primary",
+                            isActive && "text-gray-900 dark:text-gray-100",
                             "mr-2 h-4 w-4",
                           )}
                         />

llama_stack/ui/components/layout/detail-layout.tsx

@@ -93,7 +93,9 @@ export function PropertyItem({
     >
       <strong>{label}:</strong>{" "}
       {typeof value === "string" || typeof value === "number" ? (
-        <span className="text-gray-900 font-medium">{value}</span>
+        <span className="text-gray-900 dark:text-gray-100 font-medium">
+          {value}
+        </span>
       ) : (
         value
       )}
@@ -112,7 +114,9 @@ export function PropertiesCard({ children }: PropertiesCardProps) {
         <CardTitle>Properties</CardTitle>
       </CardHeader>
       <CardContent>
-        <ul className="space-y-2 text-sm text-gray-600">{children}</ul>
+        <ul className="space-y-2 text-sm text-gray-600 dark:text-gray-400">
+          {children}
+        </ul>
       </CardContent>
     </Card>
   );

llama_stack/ui/components/ui/message-components.tsx

@@ -17,10 +17,10 @@ export const MessageBlock: React.FC<MessageBlockProps> = ({
 }) => {
   return (
     <div className={`mb-4 ${className}`}>
-      <p className="py-1 font-semibold text-gray-800 mb-1">
+      <p className="py-1 font-semibold text-muted-foreground mb-1">
         {label}
         {labelDetail && (
-          <span className="text-xs text-gray-500 font-normal ml-1">
+          <span className="text-xs text-muted-foreground font-normal ml-1">
             {labelDetail}
           </span>
         )}

llama_stack/ui/components/vector-stores/vector-store-detail.tsx

@@ -0,0 +1,128 @@
+"use client";
+
+import type { VectorStore } from "llama-stack-client/resources/vector-stores/vector-stores";
+import type { VectorStoreFile } from "llama-stack-client/resources/vector-stores/files";
+import { Card, CardContent, CardHeader, CardTitle } from "@/components/ui/card";
+import { Skeleton } from "@/components/ui/skeleton";
+import {
+  DetailLoadingView,
+  DetailErrorView,
+  DetailNotFoundView,
+  DetailLayout,
+  PropertiesCard,
+  PropertyItem,
+} from "@/components/layout/detail-layout";
+import {
+  Table,
+  TableBody,
+  TableCaption,
+  TableCell,
+  TableHead,
+  TableHeader,
+  TableRow,
+} from "@/components/ui/table";
+
+interface VectorStoreDetailViewProps {
+  store: VectorStore | null;
+  files: VectorStoreFile[];
+  isLoadingStore: boolean;
+  isLoadingFiles: boolean;
+  errorStore: Error | null;
+  errorFiles: Error | null;
+  id: string;
+}
+
+export function VectorStoreDetailView({
+  store,
+  files,
+  isLoadingStore,
+  isLoadingFiles,
+  errorStore,
+  errorFiles,
+  id,
+}: VectorStoreDetailViewProps) {
+  const title = "Vector Store Details";
+
+  if (errorStore) {
+    return <DetailErrorView title={title} id={id} error={errorStore} />;
+  }
+  if (isLoadingStore) {
+    return <DetailLoadingView title={title} />;
+  }
+  if (!store) {
+    return <DetailNotFoundView title={title} id={id} />;
+  }
+
+  const mainContent = (
+    <>
+      <Card>
+        <CardHeader>
+          <CardTitle>Files</CardTitle>
+        </CardHeader>
+        <CardContent>
+          {isLoadingFiles ? (
+            <Skeleton className="h-4 w-full" />
+          ) : errorFiles ? (
+            <div className="text-destructive text-sm">
+              Error loading files: {errorFiles.message}
+            </div>
+          ) : files.length > 0 ? (
+            <Table>
+              <TableCaption>Files in this vector store</TableCaption>
+              <TableHeader>
+                <TableRow>
+                  <TableHead>ID</TableHead>
+                  <TableHead>Status</TableHead>
+                  <TableHead>Created</TableHead>
+                  <TableHead>Usage Bytes</TableHead>
+                </TableRow>
+              </TableHeader>
+              <TableBody>
+                {files.map((file) => (
+                  <TableRow key={file.id}>
+                    <TableCell>{file.id}</TableCell>
+                    <TableCell>{file.status}</TableCell>
+                    <TableCell>
+                      {new Date(file.created_at * 1000).toLocaleString()}
+                    </TableCell>
+                    <TableCell>{file.usage_bytes}</TableCell>
+                  </TableRow>
+                ))}
+              </TableBody>
+            </Table>
+          ) : (
+            <p className="text-gray-500 italic text-sm">
+              No files in this vector store.
+            </p>
+          )}
+        </CardContent>
+      </Card>
+    </>
+  );
+
+  const sidebar = (
+    <PropertiesCard>
+      <PropertyItem label="ID" value={store.id} />
+      <PropertyItem label="Name" value={store.name || ""} />
+      <PropertyItem
+        label="Created"
+        value={new Date(store.created_at * 1000).toLocaleString()}
+      />
+      <PropertyItem label="Status" value={store.status} />
+      <PropertyItem label="Total Files" value={store.file_counts.total} />
+      <PropertyItem label="Usage Bytes" value={store.usage_bytes} />
+      <PropertyItem
+        label="Provider ID"
+        value={(store.metadata.provider_id as string) || ""}
+      />
+      <PropertyItem
+        label="Provider DB ID"
+        value={(store.metadata.provider_vector_db_id as string) || ""}
+      />
+    </PropertiesCard>
+  );
+
+  return (
+    <DetailLayout title={title} mainContent={mainContent} sidebar={sidebar} />
+  );
+}

llama_stack/ui/package-lock.json

@@ -15,7 +15,7 @@
         "@radix-ui/react-tooltip": "^1.2.6",
         "class-variance-authority": "^0.7.1",
         "clsx": "^2.1.1",
-        "llama-stack-client": "0.2.13",
+        "llama-stack-client": "^0.2.14",
         "lucide-react": "^0.510.0",
         "next": "15.3.3",
         "next-auth": "^4.24.11",
@@ -676,406 +676,6 @@
         "tslib": "^2.4.0"
       }
     },
-    "node_modules/@esbuild/aix-ppc64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/aix-ppc64/-/aix-ppc64-0.25.5.tgz",
-      "integrity": "sha512-9o3TMmpmftaCMepOdA5k/yDw8SfInyzWWTjYTFCX3kPSDJMROQTb8jg+h9Cnwnmm1vOzvxN7gIfB5V2ewpjtGA==",
-      "cpu": [
-        "ppc64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "aix"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/android-arm": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/android-arm/-/android-arm-0.25.5.tgz",
-      "integrity": "sha512-AdJKSPeEHgi7/ZhuIPtcQKr5RQdo6OO2IL87JkianiMYMPbCtot9fxPbrMiBADOWWm3T2si9stAiVsGbTQFkbA==",
-      "cpu": [
-        "arm"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "android"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/android-arm64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/android-arm64/-/android-arm64-0.25.5.tgz",
-      "integrity": "sha512-VGzGhj4lJO+TVGV1v8ntCZWJktV7SGCs3Pn1GRWI1SBFtRALoomm8k5E9Pmwg3HOAal2VDc2F9+PM/rEY6oIDg==",
-      "cpu": [
-        "arm64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "android"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/android-x64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/android-x64/-/android-x64-0.25.5.tgz",
-      "integrity": "sha512-D2GyJT1kjvO//drbRT3Hib9XPwQeWd9vZoBJn+bu/lVsOZ13cqNdDeqIF/xQ5/VmWvMduP6AmXvylO/PIc2isw==",
-      "cpu": [
-        "x64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "android"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/darwin-arm64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/darwin-arm64/-/darwin-arm64-0.25.5.tgz",
-      "integrity": "sha512-GtaBgammVvdF7aPIgH2jxMDdivezgFu6iKpmT+48+F8Hhg5J/sfnDieg0aeG/jfSvkYQU2/pceFPDKlqZzwnfQ==",
-      "cpu": [
-        "arm64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "darwin"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/darwin-x64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/darwin-x64/-/darwin-x64-0.25.5.tgz",
-      "integrity": "sha512-1iT4FVL0dJ76/q1wd7XDsXrSW+oLoquptvh4CLR4kITDtqi2e/xwXwdCVH8hVHU43wgJdsq7Gxuzcs6Iq/7bxQ==",
-      "cpu": [
-        "x64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "darwin"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/freebsd-arm64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/freebsd-arm64/-/freebsd-arm64-0.25.5.tgz",
-      "integrity": "sha512-nk4tGP3JThz4La38Uy/gzyXtpkPW8zSAmoUhK9xKKXdBCzKODMc2adkB2+8om9BDYugz+uGV7sLmpTYzvmz6Sw==",
-      "cpu": [
-        "arm64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "freebsd"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/freebsd-x64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/freebsd-x64/-/freebsd-x64-0.25.5.tgz",
-      "integrity": "sha512-PrikaNjiXdR2laW6OIjlbeuCPrPaAl0IwPIaRv+SMV8CiM8i2LqVUHFC1+8eORgWyY7yhQY+2U2fA55mBzReaw==",
-      "cpu": [
-        "x64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "freebsd"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/linux-arm": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/linux-arm/-/linux-arm-0.25.5.tgz",
-      "integrity": "sha512-cPzojwW2okgh7ZlRpcBEtsX7WBuqbLrNXqLU89GxWbNt6uIg78ET82qifUy3W6OVww6ZWobWub5oqZOVtwolfw==",
-      "cpu": [
-        "arm"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "linux"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/linux-arm64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/linux-arm64/-/linux-arm64-0.25.5.tgz",
-      "integrity": "sha512-Z9kfb1v6ZlGbWj8EJk9T6czVEjjq2ntSYLY2cw6pAZl4oKtfgQuS4HOq41M/BcoLPzrUbNd+R4BXFyH//nHxVg==",
-      "cpu": [
-        "arm64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "linux"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/linux-ia32": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/linux-ia32/-/linux-ia32-0.25.5.tgz",
-      "integrity": "sha512-sQ7l00M8bSv36GLV95BVAdhJ2QsIbCuCjh/uYrWiMQSUuV+LpXwIqhgJDcvMTj+VsQmqAHL2yYaasENvJ7CDKA==",
-      "cpu": [
-        "ia32"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "linux"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/linux-loong64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/linux-loong64/-/linux-loong64-0.25.5.tgz",
-      "integrity": "sha512-0ur7ae16hDUC4OL5iEnDb0tZHDxYmuQyhKhsPBV8f99f6Z9KQM02g33f93rNH5A30agMS46u2HP6qTdEt6Q1kg==",
-      "cpu": [
-        "loong64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "linux"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/linux-mips64el": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/linux-mips64el/-/linux-mips64el-0.25.5.tgz",
-      "integrity": "sha512-kB/66P1OsHO5zLz0i6X0RxlQ+3cu0mkxS3TKFvkb5lin6uwZ/ttOkP3Z8lfR9mJOBk14ZwZ9182SIIWFGNmqmg==",
-      "cpu": [
-        "mips64el"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "linux"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/linux-ppc64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/linux-ppc64/-/linux-ppc64-0.25.5.tgz",
-      "integrity": "sha512-UZCmJ7r9X2fe2D6jBmkLBMQetXPXIsZjQJCjgwpVDz+YMcS6oFR27alkgGv3Oqkv07bxdvw7fyB71/olceJhkQ==",
-      "cpu": [
-        "ppc64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "linux"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/linux-riscv64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/linux-riscv64/-/linux-riscv64-0.25.5.tgz",
-      "integrity": "sha512-kTxwu4mLyeOlsVIFPfQo+fQJAV9mh24xL+y+Bm6ej067sYANjyEw1dNHmvoqxJUCMnkBdKpvOn0Ahql6+4VyeA==",
-      "cpu": [
-        "riscv64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "linux"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/linux-s390x": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/linux-s390x/-/linux-s390x-0.25.5.tgz",
-      "integrity": "sha512-K2dSKTKfmdh78uJ3NcWFiqyRrimfdinS5ErLSn3vluHNeHVnBAFWC8a4X5N+7FgVE1EjXS1QDZbpqZBjfrqMTQ==",
-      "cpu": [
-        "s390x"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "linux"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/linux-x64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/linux-x64/-/linux-x64-0.25.5.tgz",
-      "integrity": "sha512-uhj8N2obKTE6pSZ+aMUbqq+1nXxNjZIIjCjGLfsWvVpy7gKCOL6rsY1MhRh9zLtUtAI7vpgLMK6DxjO8Qm9lJw==",
-      "cpu": [
-        "x64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "linux"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/netbsd-arm64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/netbsd-arm64/-/netbsd-arm64-0.25.5.tgz",
-      "integrity": "sha512-pwHtMP9viAy1oHPvgxtOv+OkduK5ugofNTVDilIzBLpoWAM16r7b/mxBvfpuQDpRQFMfuVr5aLcn4yveGvBZvw==",
-      "cpu": [
-        "arm64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "netbsd"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/netbsd-x64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/netbsd-x64/-/netbsd-x64-0.25.5.tgz",
-      "integrity": "sha512-WOb5fKrvVTRMfWFNCroYWWklbnXH0Q5rZppjq0vQIdlsQKuw6mdSihwSo4RV/YdQ5UCKKvBy7/0ZZYLBZKIbwQ==",
-      "cpu": [
-        "x64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "netbsd"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/openbsd-arm64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/openbsd-arm64/-/openbsd-arm64-0.25.5.tgz",
-      "integrity": "sha512-7A208+uQKgTxHd0G0uqZO8UjK2R0DDb4fDmERtARjSHWxqMTye4Erz4zZafx7Di9Cv+lNHYuncAkiGFySoD+Mw==",
-      "cpu": [
-        "arm64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "openbsd"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/openbsd-x64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/openbsd-x64/-/openbsd-x64-0.25.5.tgz",
-      "integrity": "sha512-G4hE405ErTWraiZ8UiSoesH8DaCsMm0Cay4fsFWOOUcz8b8rC6uCvnagr+gnioEjWn0wC+o1/TAHt+It+MpIMg==",
-      "cpu": [
-        "x64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "openbsd"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/sunos-x64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/sunos-x64/-/sunos-x64-0.25.5.tgz",
-      "integrity": "sha512-l+azKShMy7FxzY0Rj4RCt5VD/q8mG/e+mDivgspo+yL8zW7qEwctQ6YqKX34DTEleFAvCIUviCFX1SDZRSyMQA==",
-      "cpu": [
-        "x64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "sunos"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/win32-arm64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/win32-arm64/-/win32-arm64-0.25.5.tgz",
-      "integrity": "sha512-O2S7SNZzdcFG7eFKgvwUEZ2VG9D/sn/eIiz8XRZ1Q/DO5a3s76Xv0mdBzVM5j5R639lXQmPmSo0iRpHqUUrsxw==",
-      "cpu": [
-        "arm64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "win32"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/win32-ia32": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/win32-ia32/-/win32-ia32-0.25.5.tgz",
-      "integrity": "sha512-onOJ02pqs9h1iMJ1PQphR+VZv8qBMQ77Klcsqv9CNW2w6yLqoURLcgERAIurY6QE63bbLuqgP9ATqajFLK5AMQ==",
-      "cpu": [
-        "ia32"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "win32"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
-    "node_modules/@esbuild/win32-x64": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/@esbuild/win32-x64/-/win32-x64-0.25.5.tgz",
-      "integrity": "sha512-TXv6YnJ8ZMVdX+SXWVBo/0p8LTcrUYngpWjvm91TMjjBQii7Oz11Lw5lbDV5Y0TzuhSJHwiH4hEtC1I42mMS0g==",
-      "cpu": [
-        "x64"
-      ],
-      "license": "MIT",
-      "optional": true,
-      "os": [
-        "win32"
-      ],
-      "engines": {
-        "node": ">=18"
-      }
-    },
     "node_modules/@eslint-community/eslint-utils": {
       "version": "4.7.0",
       "resolved": "https://registry.npmjs.org/@eslint-community/eslint-utils/-/eslint-utils-4.7.0.tgz",
@@ -5999,46 +5599,6 @@
         "url": "https://github.com/sponsors/ljharb"
       }
     },
-    "node_modules/esbuild": {
-      "version": "0.25.5",
-      "resolved": "https://registry.npmjs.org/esbuild/-/esbuild-0.25.5.tgz",
-      "integrity": "sha512-P8OtKZRv/5J5hhz0cUAdu/cLuPIKXpQl1R9pZtvmHWQvrAUVd0UNIPT4IB4W3rNOqVO0rlqHmCIbSwxh/c9yUQ==",
-      "hasInstallScript": true,
-      "license": "MIT",
-      "bin": {
-        "esbuild": "bin/esbuild"
-      },
-      "engines": {
-        "node": ">=18"
-      },
-      "optionalDependencies": {
-        "@esbuild/aix-ppc64": "0.25.5",
-        "@esbuild/android-arm": "0.25.5",
-        "@esbuild/android-arm64": "0.25.5",
-        "@esbuild/android-x64": "0.25.5",
-        "@esbuild/darwin-arm64": "0.25.5",
-        "@esbuild/darwin-x64": "0.25.5",
-        "@esbuild/freebsd-arm64": "0.25.5",
-        "@esbuild/freebsd-x64": "0.25.5",
-        "@esbuild/linux-arm": "0.25.5",
-        "@esbuild/linux-arm64": "0.25.5",
-        "@esbuild/linux-ia32": "0.25.5",
-        "@esbuild/linux-loong64": "0.25.5",
-        "@esbuild/linux-mips64el": "0.25.5",
-        "@esbuild/linux-ppc64": "0.25.5",
-        "@esbuild/linux-riscv64": "0.25.5",
-        "@esbuild/linux-s390x": "0.25.5",
-        "@esbuild/linux-x64": "0.25.5",
-        "@esbuild/netbsd-arm64": "0.25.5",
-        "@esbuild/netbsd-x64": "0.25.5",
-        "@esbuild/openbsd-arm64": "0.25.5",
-        "@esbuild/openbsd-x64": "0.25.5",
-        "@esbuild/sunos-x64": "0.25.5",
-        "@esbuild/win32-arm64": "0.25.5",
-        "@esbuild/win32-ia32": "0.25.5",
-        "@esbuild/win32-x64": "0.25.5"
-      }
-    },
     "node_modules/escalade": {
       "version": "3.2.0",
       "resolved": "https://registry.npmjs.org/escalade/-/escalade-3.2.0.tgz",
@@ -6993,6 +6553,7 @@
       "version": "2.3.3",
       "resolved": "https://registry.npmjs.org/fsevents/-/fsevents-2.3.3.tgz",
       "integrity": "sha512-5xoDfX+fL7faATnagmWPpbFtwh/R77WmMMqqHGS65C3vvB0YHrgF+B1YmZ3441tMj5n63k0212XNoJwzlhffQw==",
+      "dev": true,
       "hasInstallScript": true,
       "license": "MIT",
       "optional": true,
@@ -7154,6 +6715,7 @@
       "version": "4.10.0",
       "resolved": "https://registry.npmjs.org/get-tsconfig/-/get-tsconfig-4.10.0.tgz",
       "integrity": "sha512-kGzZ3LWWQcGIAmg6iWvXn0ei6WDtV26wzHRMwDSzmAbcXrTEXxHy6IehI6/4eT6VRKyMP1eF1VqwrVUmE/LR7A==",
+      "dev": true,
       "license": "MIT",
       "dependencies": {
         "resolve-pkg-maps": "^1.0.0"
@@ -9537,9 +9099,10 @@
       "license": "MIT"
     },
     "node_modules/llama-stack-client": {
-      "version": "0.2.13",
+      "version": "0.2.14",
-      "resolved": "https://registry.npmjs.org/llama-stack-client/-/llama-stack-client-0.2.13.tgz",
+      "resolved": "https://registry.npmjs.org/llama-stack-client/-/llama-stack-client-0.2.14.tgz",
-      "integrity": "sha512-R1rTFLwgUimr+KjEUkzUvFL6vLASwS9qj3UDSVkJ5BmrKAs5GwVAMeL7yZaTBXGuPUVh124WSlC4d9H0FjWqLA==",
+      "integrity": "sha512-bVU3JHp+EPEKR0Vb9vcd9ZyQj/72jSDuptKLwOXET9WrkphIQ8xuW5ueecMTgq8UEls3lwB3HiZM2cDOR9eDsQ==",
+      "license": "Apache-2.0",
       "dependencies": {
         "@types/node": "^18.11.18",
         "@types/node-fetch": "^2.6.4",
@@ -9547,8 +9110,7 @@
         "agentkeepalive": "^4.2.1",
         "form-data-encoder": "1.7.2",
         "formdata-node": "^4.3.2",
-        "node-fetch": "^2.6.7",
+        "node-fetch": "^2.6.7"
-        "tsx": "^4.19.2"
       }
     },
     "node_modules/llama-stack-client/node_modules/@types/node": {
@@ -11148,6 +10710,7 @@
       "version": "1.0.0",
       "resolved": "https://registry.npmjs.org/resolve-pkg-maps/-/resolve-pkg-maps-1.0.0.tgz",
       "integrity": "sha512-seS2Tj26TBVOC2NIc2rOe2y2ZO7efxITtLZcGSOnHHNOQ7CkiUBfw0Iw2ck6xkIhPwLhKNLS8BO+hEpngQlqzw==",
+      "dev": true,
       "license": "MIT",
       "funding": {
         "url": "https://github.com/privatenumber/resolve-pkg-maps?sponsor=1"
@@ -12198,25 +11761,6 @@
       "integrity": "sha512-oJFu94HQb+KVduSUQL7wnpmqnfmLsOA/nAh6b6EH0wCEoK0/mPeXU6c3wKDV83MkOuHPRHtSXKKU99IBazS/2w==",
       "license": "0BSD"
     },
-    "node_modules/tsx": {
-      "version": "4.19.4",
-      "resolved": "https://registry.npmjs.org/tsx/-/tsx-4.19.4.tgz",
-      "integrity": "sha512-gK5GVzDkJK1SI1zwHf32Mqxf2tSJkNx+eYcNly5+nHvWqXUJYUkWBQtKauoESz3ymezAI++ZwT855x5p5eop+Q==",
-      "license": "MIT",
-      "dependencies": {
-        "esbuild": "~0.25.0",
-        "get-tsconfig": "^4.7.5"
-      },
-      "bin": {
-        "tsx": "dist/cli.mjs"
-      },
-      "engines": {
-        "node": ">=18.0.0"
-      },
-      "optionalDependencies": {
-        "fsevents": "~2.3.3"
-      }
-    },
     "node_modules/tw-animate-css": {
       "version": "1.2.9",
       "resolved": "https://registry.npmjs.org/tw-animate-css/-/tw-animate-css-1.2.9.tgz",

llama_stack/ui/package.json

@@ -20,7 +20,7 @@
     "@radix-ui/react-tooltip": "^1.2.6",
     "class-variance-authority": "^0.7.1",
     "clsx": "^2.1.1",
-    "llama-stack-client": "0.2.13",
+    "llama-stack-client": "^0.2.14",
     "lucide-react": "^0.510.0",
     "next": "15.3.3",
     "next-auth": "^4.24.11",

pyproject.toml

@@ -32,7 +32,7 @@ dependencies = [
     "openai>=1.66",
     "prompt-toolkit",
     "python-dotenv",
-    "python-jose",
+    "python-jose[cryptography]",
     "pydantic>=2",
     "rich",
     "starlette",
@@ -42,8 +42,8 @@ dependencies = [
     "h11>=0.16.0",
     "python-multipart>=0.0.20",               # For fastapi Form
     "uvicorn>=0.34.0",                        # server
-    "opentelemetry-sdk",                      # server
+    "opentelemetry-sdk>=1.30.0",              # server
-    "opentelemetry-exporter-otlp-proto-http", # server
+    "opentelemetry-exporter-otlp-proto-http>=1.30.0", # server
     "aiosqlite>=0.21.0",                      # server - for metadata store
     "asyncpg",                                # for metadata store
 ]
@@ -58,12 +58,13 @@ ui = [
 
 [dependency-groups]
 dev = [
-    "pytest",
+    "pytest>=8.4",
     "pytest-timeout",
-    "pytest-asyncio",
+    "pytest-asyncio>=1.0",
     "pytest-cov",
     "pytest-html",
     "pytest-json-report",
+    "pytest-socket",      # For blocking network access in unit tests
     "nbval",              # For notebook testing
     "black",
     "ruff",
@@ -87,6 +88,8 @@ unit = [
     "blobfile",
     "faiss-cpu",
     "pymilvus>=2.5.12",
+    "litellm",
+    "together",
 ]
 # These are the core dependencies required for running integration tests. They are shared across all
 # providers. If a provider requires additional dependencies, please add them to your environment
@@ -226,7 +229,6 @@ follow_imports = "silent"
 exclude = [
     # As we fix more and more of these, we should remove them from the list
     "^llama_stack/cli/download\\.py$",
-    "^llama_stack/cli/stack/_build\\.py$",
     "^llama_stack/distribution/build\\.py$",
     "^llama_stack/distribution/client\\.py$",
     "^llama_stack/distribution/request_headers\\.py$",
@@ -256,7 +258,6 @@ exclude = [
     "^llama_stack/providers/inline/inference/sentence_transformers/sentence_transformers\\.py$",
     "^llama_stack/providers/inline/inference/vllm/",
     "^llama_stack/providers/inline/post_training/common/validator\\.py$",
-    "^llama_stack/providers/inline/post_training/torchtune/post_training\\.py$",
     "^llama_stack/providers/inline/safety/code_scanner/",
     "^llama_stack/providers/inline/safety/llama_guard/",
     "^llama_stack/providers/inline/safety/prompt_guard/",
@@ -341,3 +342,9 @@ warn_required_dynamic_aliases = true
 
 [tool.ruff.lint.pep8-naming]
 classmethod-decorators = ["classmethod", "pydantic.field_validator"]
+
+[tool.pytest.ini_options]
+asyncio_mode = "auto"
+markers = [
+    "allow_network: Allow network access for specific unit tests",
+]

requirements.txt

@@ -28,6 +28,8 @@ certifi==2025.1.31
     #   httpcore
     #   httpx
     #   requests
+cffi==1.17.1 ; platform_python_implementation != 'PyPy'
+    # via cryptography
 charset-normalizer==3.4.1
     # via requests
 click==8.1.8
@@ -38,6 +40,8 @@ colorama==0.4.6 ; sys_platform == 'win32'
     # via
     #   click
     #   tqdm
+cryptography==45.0.5
+    # via python-jose
 deprecated==1.2.18
     # via
     #   opentelemetry-api
@@ -156,6 +160,8 @@ pyasn1==0.4.8
     # via
     #   python-jose
     #   rsa
+pycparser==2.22 ; platform_python_implementation != 'PyPy'
+    # via cffi
 pydantic==2.10.6
     # via
     #   fastapi

scripts/unit-tests.sh

@@ -16,4 +16,4 @@ if [ $FOUND_PYTHON -ne 0 ]; then
      uv python install "$PYTHON_VERSION"
 fi
 
-uv run --python "$PYTHON_VERSION" --with-editable . --group unit pytest --asyncio-mode=auto -s -v tests/unit/ $@
+uv run --python "$PYTHON_VERSION" --with-editable . --group unit pytest -s -v tests/unit/ $@

tests/Containerfile

@@ -7,7 +7,8 @@ FROM --platform=linux/amd64 ollama/ollama:latest
 RUN ollama serve & \
     sleep 5 && \
     ollama pull llama3.2:3b-instruct-fp16 && \
-    ollama pull all-minilm:l6-v2
+    ollama pull all-minilm:l6-v2 && \
+    ollama pull llama-guard3:1b
 
 # Set the entrypoint to start ollama serve
 ENTRYPOINT ["ollama", "serve"]

tests/integration/agents/test_agents.py

@@ -77,6 +77,24 @@ def agent_config(llama_stack_client, text_model_id):
     return agent_config
 
 
+@pytest.fixture(scope="session")
+def agent_config_without_safety(text_model_id):
+    agent_config = dict(
+        model=text_model_id,
+        instructions="You are a helpful assistant",
+        sampling_params={
+            "strategy": {
+                "type": "top_p",
+                "temperature": 0.0001,
+                "top_p": 0.9,
+            },
+        },
+        tools=[],
+        enable_session_persistence=False,
+    )
+    return agent_config
+
+
 def test_agent_simple(llama_stack_client, agent_config):
     agent = Agent(llama_stack_client, **agent_config)
     session_id = agent.create_session(f"test-session-{uuid4()}")
@@ -491,7 +509,7 @@ def test_rag_agent(llama_stack_client, agent_config, rag_tool_name):
             assert expected_kw in response.output_message.content.lower()
 
 
-def test_rag_agent_with_attachments(llama_stack_client, agent_config):
+def test_rag_agent_with_attachments(llama_stack_client, agent_config_without_safety):
     urls = ["llama3.rst", "lora_finetune.rst"]
     documents = [
         # passign as url
@@ -514,14 +532,8 @@ def test_rag_agent_with_attachments(llama_stack_client, agent_config):
             metadata={},
         ),
     ]
-    rag_agent = Agent(llama_stack_client, **agent_config)
+    rag_agent = Agent(llama_stack_client, **agent_config_without_safety)
     session_id = rag_agent.create_session(f"test-session-{uuid4()}")
-    user_prompts = [
-        (
-            "Instead of the standard multi-head attention, what attention type does Llama3-8B use?",
-            "grouped",
-        ),
-    ]
     user_prompts = [
         (
             "I am attaching some documentation for Torchtune. Help me answer questions I will ask next.",
@@ -549,82 +561,6 @@ def test_rag_agent_with_attachments(llama_stack_client, agent_config):
     assert "lora" in response.output_message.content.lower()
 
 
-@pytest.mark.skip(reason="Code interpreter is currently disabled in the Stack")
-def test_rag_and_code_agent(llama_stack_client, agent_config):
-    if "llama-4" in agent_config["model"].lower():
-        pytest.xfail("Not working for llama4")
-
-    documents = []
-    documents.append(
-        Document(
-            document_id="nba_wiki",
-            content="The NBA was created on August 3, 1949, with the merger of the Basketball Association of America (BAA) and the National Basketball League (NBL).",
-            metadata={},
-        )
-    )
-    documents.append(
-        Document(
-            document_id="perplexity_wiki",
-            content="""Perplexity the company was founded in 2022 by Aravind Srinivas, Andy Konwinski, Denis Yarats and Johnny Ho, engineers with backgrounds in back-end systems, artificial intelligence (AI) and machine learning:
-
-    Srinivas, the CEO, worked at OpenAI as an AI researcher.
-    Konwinski was among the founding team at Databricks.
-    Yarats, the CTO, was an AI research scientist at Meta.
-    Ho, the CSO, worked as an engineer at Quora, then as a quantitative trader on Wall Street.[5]""",
-            metadata={},
-        )
-    )
-    vector_db_id = f"test-vector-db-{uuid4()}"
-    llama_stack_client.vector_dbs.register(
-        vector_db_id=vector_db_id,
-        embedding_model="all-MiniLM-L6-v2",
-        embedding_dimension=384,
-    )
-    llama_stack_client.tool_runtime.rag_tool.insert(
-        documents=documents,
-        vector_db_id=vector_db_id,
-        chunk_size_in_tokens=128,
-    )
-    agent_config = {
-        **agent_config,
-        "tools": [
-            dict(
-                name="builtin::rag/knowledge_search",
-                args={"vector_db_ids": [vector_db_id]},
-            ),
-            "builtin::code_interpreter",
-        ],
-    }
-    agent = Agent(llama_stack_client, **agent_config)
-    user_prompts = [
-        (
-            "when was Perplexity the company founded?",
-            [],
-            "knowledge_search",
-            "2022",
-        ),
-        (
-            "when was the nba created?",
-            [],
-            "knowledge_search",
-            "1949",
-        ),
-    ]
-
-    for prompt, docs, tool_name, expected_kw in user_prompts:
-        session_id = agent.create_session(f"test-session-{uuid4()}")
-        response = agent.create_turn(
-            messages=[{"role": "user", "content": prompt}],
-            session_id=session_id,
-            documents=docs,
-            stream=False,
-        )
-        tool_execution_step = next(step for step in response.steps if step.step_type == "tool_execution")
-        assert tool_execution_step.tool_calls[0].tool_name == tool_name, f"Failed on {prompt}"
-        if expected_kw:
-            assert expected_kw in response.output_message.content.lower()
-
-
 @pytest.mark.parametrize(
     "client_tools",
     [(get_boiling_point, False), (get_boiling_point_with_metadata, True)],

tests/integration/agents/test_persistence.py

@@ -44,7 +44,6 @@ def common_params(inference_model):
     )
 
 
-@pytest.mark.asyncio
 @pytest.mark.skip(reason="This test needs to be migrated to api / client-sdk world")
 async def test_delete_agents_and_sessions(self, agents_stack, common_params):
     agents_impl = agents_stack.impls[Api.agents]
@@ -73,7 +72,6 @@ async def test_delete_agents_and_sessions(self, agents_stack, common_params):
     assert agent_response is None
 
 
-@pytest.mark.asyncio
 @pytest.mark.skip(reason="This test needs to be migrated to api / client-sdk world")
 async def test_get_agent_turns_and_steps(self, agents_stack, sample_messages, common_params):
     agents_impl = agents_stack.impls[Api.agents]

tests/integration/fixtures/common.py

@@ -6,6 +6,7 @@
 
 import inspect
 import os
+import signal
 import socket
 import subprocess
 import tempfile
@@ -45,6 +46,8 @@ def start_llama_stack_server(config_name: str) -> subprocess.Popen:
         stderr=subprocess.PIPE,  # keep stderr to see errors
         text=True,
         env={**os.environ, "LLAMA_STACK_LOG_FILE": "server.log"},
+        # Create new process group so we can kill all child processes
+        preexec_fn=os.setsid,
     )
     return process
 
@@ -197,7 +200,7 @@ def llama_stack_client(request, provider_data):
             server_process = start_llama_stack_server(config_name)
 
             # Wait for server to be ready
-            if not wait_for_server_ready(base_url, timeout=30, process=server_process):
+            if not wait_for_server_ready(base_url, timeout=120, process=server_process):
                 print("Server failed to start within timeout")
                 server_process.terminate()
                 raise RuntimeError(
@@ -215,6 +218,7 @@ def llama_stack_client(request, provider_data):
         return LlamaStackClient(
             base_url=base_url,
             provider_data=provider_data,
+            timeout=int(os.environ.get("LLAMA_STACK_CLIENT_TIMEOUT", "30")),
         )
 
     # check if this looks like a URL using proper URL parsing
@@ -267,14 +271,17 @@ def cleanup_server_process(request):
                 print(f"Server process already terminated with return code: {server_process.returncode}")
                 return
             try:
-                server_process.terminate()
+                print(f"Terminating process {server_process.pid} and its group...")
+                # Kill the entire process group
+                os.killpg(os.getpgid(server_process.pid), signal.SIGTERM)
                 server_process.wait(timeout=10)
-                print("Server process terminated gracefully")
+                print("Server process and children terminated gracefully")
             except subprocess.TimeoutExpired:
                 print("Server process did not terminate gracefully, killing it")
-                server_process.kill()
+                # Force kill the entire process group
+                os.killpg(os.getpgid(server_process.pid), signal.SIGKILL)
                 server_process.wait()
-                print("Server process killed")
+                print("Server process and children killed")
             except Exception as e:
                 print(f"Error during server cleanup: {e}")
         else:

tests/integration/inference/test_openai_completion.py

@@ -71,7 +71,6 @@ def skip_if_model_doesnt_support_openai_chat_completion(client_with_models, mode
         "remote::cerebras",
         "remote::databricks",
         "remote::runpod",
-        "remote::sambanova",
         "remote::tgi",
     ):
         pytest.skip(f"Model {model_id} hosted by {provider.provider_type} doesn't support OpenAI chat completions.")

tests/integration/inspect/test_inspect.py

@@ -4,20 +4,17 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
-import pytest
 from llama_stack_client import LlamaStackClient
 
 from llama_stack import LlamaStackAsLibraryClient
 
 
 class TestInspect:
-    @pytest.mark.asyncio
     def test_health(self, llama_stack_client: LlamaStackAsLibraryClient | LlamaStackClient):
         health = llama_stack_client.inspect.health()
         assert health is not None
         assert health.status == "OK"
 
-    @pytest.mark.asyncio
     def test_version(self, llama_stack_client: LlamaStackAsLibraryClient | LlamaStackClient):
         version = llama_stack_client.inspect.version()
         assert version is not None

tests/integration/providers/test_providers.py

@@ -4,14 +4,12 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
-import pytest
 from llama_stack_client import LlamaStackClient
 
 from llama_stack import LlamaStackAsLibraryClient
 
 
 class TestProviders:
-    @pytest.mark.asyncio
     def test_providers(self, llama_stack_client: LlamaStackAsLibraryClient | LlamaStackClient):
         provider_list = llama_stack_client.providers.list()
         assert provider_list is not None

tests/integration/providers/utils/sqlstore/test_authorized_sqlstore.py

@@ -14,8 +14,7 @@ from llama_stack.distribution.access_control.access_control import default_polic
 from llama_stack.distribution.datatypes import User
 from llama_stack.providers.utils.sqlstore.api import ColumnType
 from llama_stack.providers.utils.sqlstore.authorized_sqlstore import AuthorizedSqlStore
-from llama_stack.providers.utils.sqlstore.sqlalchemy_sqlstore import SqlAlchemySqlStoreImpl
+from llama_stack.providers.utils.sqlstore.sqlstore import PostgresSqlStoreConfig, SqliteSqlStoreConfig, sqlstore_impl
-from llama_stack.providers.utils.sqlstore.sqlstore import PostgresSqlStoreConfig, SqliteSqlStoreConfig
 
 
 def get_postgres_config():
@@ -30,45 +29,47 @@ def get_postgres_config():
 
 
 def get_sqlite_config():
-    """Get SQLite configuration with temporary database."""
+    """Get SQLite configuration with temporary file database."""
-    tmp_file = tempfile.NamedTemporaryFile(suffix=".db", delete=False)
+    temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".db")
-    tmp_file.close()
+    temp_file.close()
-    return SqliteSqlStoreConfig(db_path=tmp_file.name), tmp_file.name
+    return SqliteSqlStoreConfig(db_path=temp_file.name)
 
 
-@pytest.mark.asyncio
+# Backend configurations for parametrized tests
-@pytest.mark.parametrize(
+BACKEND_CONFIGS = [
-    "backend_config",
-    [
     pytest.param(
-            ("postgres", get_postgres_config),
+        get_postgres_config,
         marks=pytest.mark.skipif(
             not os.environ.get("ENABLE_POSTGRES_TESTS"),
             reason="PostgreSQL tests require ENABLE_POSTGRES_TESTS environment variable",
         ),
         id="postgres",
     ),
-        pytest.param(("sqlite", get_sqlite_config), id="sqlite"),
+    pytest.param(get_sqlite_config, id="sqlite"),
-    ],
+]
-)
+
-@patch("llama_stack.providers.utils.sqlstore.authorized_sqlstore.get_authenticated_user")
+
-async def test_json_comparison(mock_get_authenticated_user, backend_config):
+@pytest.fixture
-    """Test that JSON column comparisons work correctly for both PostgreSQL and SQLite"""
+def authorized_store(backend_config):
-    backend_name, config_func = backend_config
+    """Set up authorized store with proper cleanup."""
+    config_func = backend_config
 
-    # Handle different config types
-    if backend_name == "postgres":
     config = config_func()
-        cleanup_path = None
-    else:  # sqlite
-        config, cleanup_path = config_func()
 
-    try:
+    base_sqlstore = sqlstore_impl(config)
-        base_sqlstore = SqlAlchemySqlStoreImpl(config)
     authorized_store = AuthorizedSqlStore(base_sqlstore)
 
-        # Create test table
+    yield authorized_store
-        table_name = f"test_json_comparison_{backend_name}"
+
+    if hasattr(config, "db_path"):
+        try:
+            os.unlink(config.db_path)
+        except (OSError, FileNotFoundError):
+            pass
+
+
+async def create_test_table(authorized_store, table_name):
+    """Create a test table with standard schema."""
     await authorized_store.create_table(
         table=table_name,
         schema={
@@ -77,6 +78,26 @@ async def test_json_comparison(mock_get_authenticated_user, backend_config):
         },
     )
 
+
+async def cleanup_records(sql_store, table_name, record_ids):
+    """Clean up test records."""
+    for record_id in record_ids:
+        try:
+            await sql_store.delete(table_name, {"id": record_id})
+        except Exception:
+            pass
+
+
+@pytest.mark.parametrize("backend_config", BACKEND_CONFIGS)
+@patch("llama_stack.providers.utils.sqlstore.authorized_sqlstore.get_authenticated_user")
+async def test_authorized_store_attributes(mock_get_authenticated_user, authorized_store, request):
+    """Test that JSON column comparisons work correctly for both PostgreSQL and SQLite"""
+    backend_name = request.node.callspec.id
+
+    # Create test table
+    table_name = f"test_json_comparison_{backend_name}"
+    await create_test_table(authorized_store, table_name)
+
     try:
         # Test with no authenticated user (should handle JSON null comparison)
         mock_get_authenticated_user.return_value = None
@@ -158,16 +179,61 @@ async def test_json_comparison(mock_get_authenticated_user, backend_config):
 
     finally:
         # Clean up records
-            for record_id in ["1", "2", "3", "4", "5", "6"]:
+        await cleanup_records(authorized_store.sql_store, table_name, ["1", "2", "3", "4", "5", "6"])
+
+
+@pytest.mark.parametrize("backend_config", BACKEND_CONFIGS)
+@patch("llama_stack.providers.utils.sqlstore.authorized_sqlstore.get_authenticated_user")
+async def test_user_ownership_policy(mock_get_authenticated_user, authorized_store, request):
+    """Test that 'user is owner' policies work correctly with record ownership"""
+    from llama_stack.distribution.access_control.datatypes import AccessRule, Action, Scope
+
+    backend_name = request.node.callspec.id
+
+    # Create test table
+    table_name = f"test_ownership_{backend_name}"
+    await create_test_table(authorized_store, table_name)
+
     try:
-                    await base_sqlstore.delete(table_name, {"id": record_id})
+        # Test with first user who creates records
-                except Exception:
+        user1 = User("user1", {"roles": ["admin"]})
-                    pass
+        mock_get_authenticated_user.return_value = user1
+
+        # Insert a record owned by user1
+        await authorized_store.insert(table_name, {"id": "1", "data": "user1_data"})
+
+        # Test with second user
+        user2 = User("user2", {"roles": ["user"]})
+        mock_get_authenticated_user.return_value = user2
+
+        # Insert a record owned by user2
+        await authorized_store.insert(table_name, {"id": "2", "data": "user2_data"})
+
+        # Create a policy that only allows access when user is the owner
+        owner_only_policy = [
+            AccessRule(
+                permit=Scope(actions=[Action.READ]),
+                when=["user is owner"],
+            ),
+        ]
+
+        # Test user1 access - should only see their own record
+        mock_get_authenticated_user.return_value = user1
+        result = await authorized_store.fetch_all(table_name, policy=owner_only_policy)
+        assert len(result.data) == 1, f"Expected user1 to see 1 record, got {len(result.data)}"
+        assert result.data[0]["id"] == "1", f"Expected user1's record, got {result.data[0]['id']}"
+
+        # Test user2 access - should only see their own record
+        mock_get_authenticated_user.return_value = user2
+        result = await authorized_store.fetch_all(table_name, policy=owner_only_policy)
+        assert len(result.data) == 1, f"Expected user2 to see 1 record, got {len(result.data)}"
+        assert result.data[0]["id"] == "2", f"Expected user2's record, got {result.data[0]['id']}"
+
+        # Test with anonymous user - should see no records
+        mock_get_authenticated_user.return_value = None
+        result = await authorized_store.fetch_all(table_name, policy=owner_only_policy)
+        assert len(result.data) == 0, f"Expected anonymous user to see 0 records, got {len(result.data)}"
 
     finally:
-        # Clean up temporary SQLite database file if needed
+        # Clean up records
-        if cleanup_path:
+        await cleanup_records(authorized_store.sql_store, table_name, ["1", "2"])
-            try:
-                os.unlink(cleanup_path)
-            except OSError:
-                pass

tests/unit/conftest.py

@@ -4,6 +4,17 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
+import pytest_socket
+
 # We need to import the fixtures here so that pytest can find them
 # but ruff doesn't think they are used and removes the import. "noqa: F401" prevents them from being removed
 from .fixtures import cached_disk_dist_registry, disk_dist_registry, sqlite_kvstore  # noqa: F401
+
+
+def pytest_runtest_setup(item):
+    """Setup for each test - check if network access should be allowed."""
+    if "allow_network" in item.keywords:
+        pytest_socket.enable_socket()
+    else:
+        # Allowing Unix sockets is necessary for some tests that use local servers and mocks
+        pytest_socket.disable_socket(allow_unix_socket=True)

tests/unit/distribution/routers/test_routing_tables.py

@@ -8,8 +8,6 @@
 
 from unittest.mock import AsyncMock
 
-import pytest
-
 from llama_stack.apis.common.type_system import NumberType
 from llama_stack.apis.datasets.datasets import Dataset, DatasetPurpose, URIDataSource
 from llama_stack.apis.datatypes import Api
@@ -119,7 +117,6 @@ class ToolGroupsImpl(Impl):
         )
 
 
-@pytest.mark.asyncio
 async def test_models_routing_table(cached_disk_dist_registry):
     table = ModelsRoutingTable({"test_provider": InferenceImpl()}, cached_disk_dist_registry, {})
     await table.initialize()
@@ -161,7 +158,6 @@ async def test_models_routing_table(cached_disk_dist_registry):
     assert len(openai_models.data) == 0
 
 
-@pytest.mark.asyncio
 async def test_shields_routing_table(cached_disk_dist_registry):
     table = ShieldsRoutingTable({"test_provider": SafetyImpl()}, cached_disk_dist_registry, {})
     await table.initialize()
@@ -177,7 +173,6 @@ async def test_shields_routing_table(cached_disk_dist_registry):
     assert "test-shield-2" in shield_ids
 
 
-@pytest.mark.asyncio
 async def test_vectordbs_routing_table(cached_disk_dist_registry):
     table = VectorDBsRoutingTable({"test_provider": VectorDBImpl()}, cached_disk_dist_registry, {})
     await table.initialize()
@@ -233,7 +228,6 @@ async def test_datasets_routing_table(cached_disk_dist_registry):
     assert len(datasets.data) == 0
 
 
-@pytest.mark.asyncio
 async def test_scoring_functions_routing_table(cached_disk_dist_registry):
     table = ScoringFunctionsRoutingTable({"test_provider": ScoringFunctionsImpl()}, cached_disk_dist_registry, {})
     await table.initialize()
@@ -259,7 +253,6 @@ async def test_scoring_functions_routing_table(cached_disk_dist_registry):
     assert "test-scoring-fn-2" in scoring_fn_ids
 
 
-@pytest.mark.asyncio
 async def test_benchmarks_routing_table(cached_disk_dist_registry):
     table = BenchmarksRoutingTable({"test_provider": BenchmarksImpl()}, cached_disk_dist_registry, {})
     await table.initialize()
@@ -277,7 +270,6 @@ async def test_benchmarks_routing_table(cached_disk_dist_registry):
     assert "test-benchmark" in benchmark_ids
 
 
-@pytest.mark.asyncio
 async def test_tool_groups_routing_table(cached_disk_dist_registry):
     table = ToolGroupsRoutingTable({"test_provider": ToolGroupsImpl()}, cached_disk_dist_registry, {})
     await table.initialize()

tests/unit/distribution/test_context.py

@@ -13,7 +13,6 @@ import pytest
 from llama_stack.distribution.utils.context import preserve_contexts_async_generator
 
 
-@pytest.mark.asyncio
 async def test_preserve_contexts_with_exception():
     # Create context variable
     context_var = ContextVar("exception_var", default="initial")
@@ -41,7 +40,6 @@ async def test_preserve_contexts_with_exception():
     context_var.reset(token)
 
 
-@pytest.mark.asyncio
 async def test_preserve_contexts_empty_generator():
     # Create context variable
     context_var = ContextVar("empty_var", default="initial")
@@ -66,7 +64,6 @@ async def test_preserve_contexts_empty_generator():
     context_var.reset(token)
 
 
-@pytest.mark.asyncio
 async def test_preserve_contexts_across_event_loops():
     """
     Test that context variables are preserved across event loop boundaries with nested generators.

tests/unit/files/test_files.py

@@ -6,7 +6,6 @@
 
 
 import pytest
-import pytest_asyncio
 
 from llama_stack.apis.common.responses import Order
 from llama_stack.apis.files import OpenAIFilePurpose
@@ -29,7 +28,7 @@ class MockUploadFile:
         return self.content
 
 
-@pytest_asyncio.fixture
+@pytest.fixture
 async def files_provider(tmp_path):
     """Create a files provider with temporary storage for testing."""
     storage_dir = tmp_path / "files"
@@ -68,7 +67,6 @@ def large_file():
 class TestOpenAIFilesAPI:
     """Test suite for OpenAI Files API endpoints."""
 
-    @pytest.mark.asyncio
     async def test_upload_file_success(self, files_provider, sample_text_file):
         """Test successful file upload."""
         # Upload file
@@ -82,7 +80,6 @@ class TestOpenAIFilesAPI:
         assert result.created_at > 0
         assert result.expires_at > result.created_at
 
-    @pytest.mark.asyncio
     async def test_upload_different_purposes(self, files_provider, sample_text_file):
         """Test uploading files with different purposes."""
         purposes = list(OpenAIFilePurpose)
@@ -93,7 +90,6 @@ class TestOpenAIFilesAPI:
             uploaded_files.append(result)
             assert result.purpose == purpose
 
-    @pytest.mark.asyncio
     async def test_upload_different_file_types(self, files_provider, sample_text_file, sample_json_file, large_file):
         """Test uploading different types and sizes of files."""
         files_to_test = [
@@ -107,7 +103,6 @@ class TestOpenAIFilesAPI:
             assert result.filename == expected_filename
             assert result.bytes == len(file_obj.content)
 
-    @pytest.mark.asyncio
     async def test_list_files_empty(self, files_provider):
         """Test listing files when no files exist."""
         result = await files_provider.openai_list_files()
@@ -117,7 +112,6 @@ class TestOpenAIFilesAPI:
         assert result.first_id == ""
         assert result.last_id == ""
 
-    @pytest.mark.asyncio
     async def test_list_files_with_content(self, files_provider, sample_text_file, sample_json_file):
         """Test listing files when files exist."""
         # Upload multiple files
@@ -132,7 +126,6 @@ class TestOpenAIFilesAPI:
         assert file1.id in file_ids
         assert file2.id in file_ids
 
-    @pytest.mark.asyncio
     async def test_list_files_with_purpose_filter(self, files_provider, sample_text_file):
         """Test listing files with purpose filtering."""
         # Upload file with specific purpose
@@ -146,7 +139,6 @@ class TestOpenAIFilesAPI:
         assert result.data[0].id == uploaded_file.id
         assert result.data[0].purpose == OpenAIFilePurpose.ASSISTANTS
 
-    @pytest.mark.asyncio
     async def test_list_files_with_limit(self, files_provider, sample_text_file):
         """Test listing files with limit parameter."""
         # Upload multiple files
@@ -157,7 +149,6 @@ class TestOpenAIFilesAPI:
         result = await files_provider.openai_list_files(limit=3)
         assert len(result.data) == 3
 
-    @pytest.mark.asyncio
     async def test_list_files_with_order(self, files_provider, sample_text_file):
         """Test listing files with different order."""
         # Upload multiple files
@@ -178,7 +169,6 @@ class TestOpenAIFilesAPI:
         # Oldest should be first
         assert result_asc.data[0].created_at <= result_asc.data[1].created_at <= result_asc.data[2].created_at
 
-    @pytest.mark.asyncio
     async def test_retrieve_file_success(self, files_provider, sample_text_file):
         """Test successful file retrieval."""
         # Upload file
@@ -197,13 +187,11 @@ class TestOpenAIFilesAPI:
         assert retrieved_file.created_at == uploaded_file.created_at
         assert retrieved_file.expires_at == uploaded_file.expires_at
 
-    @pytest.mark.asyncio
     async def test_retrieve_file_not_found(self, files_provider):
         """Test retrieving a non-existent file."""
         with pytest.raises(ValueError, match="File with id file-nonexistent not found"):
             await files_provider.openai_retrieve_file("file-nonexistent")
 
-    @pytest.mark.asyncio
     async def test_retrieve_file_content_success(self, files_provider, sample_text_file):
         """Test successful file content retrieval."""
         # Upload file
@@ -217,13 +205,11 @@ class TestOpenAIFilesAPI:
         # Verify content
         assert content.body == sample_text_file.content
 
-    @pytest.mark.asyncio
     async def test_retrieve_file_content_not_found(self, files_provider):
         """Test retrieving content of a non-existent file."""
         with pytest.raises(ValueError, match="File with id file-nonexistent not found"):
             await files_provider.openai_retrieve_file_content("file-nonexistent")
 
-    @pytest.mark.asyncio
     async def test_delete_file_success(self, files_provider, sample_text_file):
         """Test successful file deletion."""
         # Upload file
@@ -245,13 +231,11 @@ class TestOpenAIFilesAPI:
         with pytest.raises(ValueError, match=f"File with id {uploaded_file.id} not found"):
             await files_provider.openai_retrieve_file(uploaded_file.id)
 
-    @pytest.mark.asyncio
     async def test_delete_file_not_found(self, files_provider):
         """Test deleting a non-existent file."""
         with pytest.raises(ValueError, match="File with id file-nonexistent not found"):
             await files_provider.openai_delete_file("file-nonexistent")
 
-    @pytest.mark.asyncio
     async def test_file_persistence_across_operations(self, files_provider, sample_text_file):
         """Test that files persist correctly across multiple operations."""
         # Upload file
@@ -279,7 +263,6 @@ class TestOpenAIFilesAPI:
         files_list = await files_provider.openai_list_files()
         assert len(files_list.data) == 0
 
-    @pytest.mark.asyncio
     async def test_multiple_files_operations(self, files_provider, sample_text_file, sample_json_file):
         """Test operations with multiple files."""
         # Upload multiple files
@@ -302,7 +285,6 @@ class TestOpenAIFilesAPI:
         content = await files_provider.openai_retrieve_file_content(file2.id)
         assert content.body == sample_json_file.content
 
-    @pytest.mark.asyncio
     async def test_file_id_uniqueness(self, files_provider, sample_text_file):
         """Test that each uploaded file gets a unique ID."""
         file_ids = set()
@@ -316,7 +298,6 @@ class TestOpenAIFilesAPI:
             file_ids.add(uploaded_file.id)
             assert uploaded_file.id.startswith("file-")
 
-    @pytest.mark.asyncio
     async def test_file_no_filename_handling(self, files_provider):
         """Test handling files with no filename."""
         file_without_name = MockUploadFile(b"content", None)  # No filename
@@ -327,7 +308,6 @@ class TestOpenAIFilesAPI:
 
         assert uploaded_file.filename == "uploaded_file"  # Default filename
 
-    @pytest.mark.asyncio
     async def test_after_pagination_works(self, files_provider, sample_text_file):
         """Test that 'after' pagination works correctly."""
         # Upload multiple files to test pagination

tests/unit/fixtures.py

@@ -4,14 +4,14 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
-import pytest_asyncio
+import pytest
 
 from llama_stack.distribution.store.registry import CachedDiskDistributionRegistry, DiskDistributionRegistry
 from llama_stack.providers.utils.kvstore.config import SqliteKVStoreConfig
 from llama_stack.providers.utils.kvstore.sqlite import SqliteKVStoreImpl
 
 
-@pytest_asyncio.fixture(scope="function")
+@pytest.fixture(scope="function")
 async def sqlite_kvstore(tmp_path):
     db_path = tmp_path / "test_kv.db"
     kvstore_config = SqliteKVStoreConfig(db_path=db_path.as_posix())
@@ -20,14 +20,14 @@ async def sqlite_kvstore(tmp_path):
     yield kvstore
 
 
-@pytest_asyncio.fixture(scope="function")
+@pytest.fixture(scope="function")
 async def disk_dist_registry(sqlite_kvstore):
     registry = DiskDistributionRegistry(sqlite_kvstore)
     await registry.initialize()
     yield registry
 
 
-@pytest_asyncio.fixture(scope="function")
+@pytest.fixture(scope="function")
 async def cached_disk_dist_registry(sqlite_kvstore):
     registry = CachedDiskDistributionRegistry(sqlite_kvstore)
     await registry.initialize()