# What does this PR do?
*Add dynamic authentication token forwarding support for vLLM provider*
This enables per-request authentication tokens for vLLM providers,
supporting use cases like RAG operations where different requests may
need different authentication tokens. The implementation follows the
same pattern as other providers like Together AI, Fireworks, and
Passthrough.
- Add LiteLLMOpenAIMixin that manages the vllm_api_token properly
Usage:
- Static: VLLM_API_TOKEN env var or config.api_token
- Dynamic: X-LlamaStack-Provider-Data header with vllm_api_token
All existing functionality is preserved while adding new dynamic
capabilities.
<!-- Provide a short summary of what this PR does and why. Link to
relevant issues if applicable. -->
<!-- If resolving an issue, uncomment and update the line below -->
<!-- Closes #[issue-number] -->
## Test Plan
<!-- Describe the tests you ran to verify your changes with result
summaries. *Provide clear instructions so the plan can be easily
re-executed.* -->
```
curl -X POST "http://localhost:8000/v1/chat/completions" -H "Authorization: Bearer my-dynamic-token" \
-H "X-LlamaStack-Provider-Data: {\"vllm_api_token\": \"Bearer my-dynamic-token\", \"vllm_url\": \"http://dynamic-server:8000\"}" \
-H "Content-Type: application/json" \
-d '{"model": "llama-3.1-8b", "messages": [{"role": "user", "content": "Hello!"}]}'
```
---------
Signed-off-by: Akram Ben Aissi <akram.benaissi@gmail.com>
# What does this PR do?
Fixes this warning in llama stack build:
```bash
WARNING 2025-09-15 15:29:02,197 llama_stack.core.distribution:149 core: Failed to import module prompts: No module named
'llama_stack.providers.registry.prompts'"
```
## Test Plan
Test added
---------
Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>
# What does this PR do?
Modified the code in registry.py.
The key changes are:
1. Removed the `return False` statement
2. Added a warning log message that includes the object type,
identifier, and provider_id for better debugging.
3. The method now continues with the registration process instead of
early returning.
---------
Co-authored-by: Omar Abdelwahab <omara@fb.com>
# What does this PR do?
<!-- Provide a short summary of what this PR does and why. Link to
relevant issues if applicable. -->
This PR provides functionality for users to unregister ScoringFn and
Benchmark resources for `scoring` and `eval` APIs.
<!-- If resolving an issue, uncomment and update the line below -->
<!-- Closes #[issue-number] -->
Closes#3051
## Test Plan
<!-- Describe the tests you ran to verify your changes with result
summaries. *Provide clear instructions so the plan can be easily
re-executed.* -->
Updated integration and unit tests via CI workflow
# What does this PR do?
the @required_args decorator in openai-python is masking the async
nature of the {AsyncCompletions,chat.AsyncCompletions}.create method.
see https://github.com/openai/openai-python/issues/996
this means two things -
0. we cannot use iscoroutine in the recorder to detect async vs non
1. our mocks are inappropriately introducing identifiable async
for (0), we update the iscoroutine check w/ detection of /v1/models,
which is the only non-async function we mock & record.
for (1), we could leave everything as is and assume (0) will catch
errors. to be defensive, we update the unit tests to mock below create
methods, allowing the true openai-python create() methods to be tested.
# What does this PR do?
the recorder mocks the openai-python interface. the openai-python
interface allows NOT_GIVEN as an input option. this change properly
handles NOT_GIVEN.
## Test Plan
ci (coverage for chat, completions, embeddings)
# What does this PR do?
Migrates MD5 and SHA-1 hash algorithms to SHA-256.
In particular, replaces:
- MD5 in chunk ID generation.
- MD5 in file verification.
- SHA-1 in model identifier digests.
And updates all related test expectations.
Original discussion:
https://github.com/llamastack/llama-stack/discussions/3413
<!-- If resolving an issue, uncomment and update the line below -->
Closes#3424.
## Test Plan
Unit tests from scripts/unit-tests.sh were updated to match the new hash
output, and ran to verify the tests pass.
Signed-off-by: Doug Edgar <dedgar@redhat.com>
# What does this PR do?
update the async detection test for vllm
- remove a network access from unit tests
- remove direct logging use
the idea behind the test is to mock inference w/ a sleep, initiate
concurrent inference calls, verify the total execution time is close to
the sleep time. in a non-async env the total time would be closer to
sleep * num concurrent calls.
## Test Plan
ci
# What does this PR do?
update vLLM inference provider to use OpenAIMixin for openai-compat
functions
inference recordings from Qwen3-0.6B and vLLM 0.8.3 -
```
docker run --gpus all -v ~/.cache/huggingface:/root/.cache/huggingface -p 8000:8000 --ipc=host \
vllm/vllm-openai:latest \
--model Qwen/Qwen3-0.6B --enable-auto-tool-choice --tool-call-parser hermes
```
## Test Plan
```
./scripts/integration-tests.sh --stack-config server:ci-tests --setup vllm --subdirs inference
```
# What does this PR do?
- Updating documentation on migration from RAG Tool to Vector Stores and
Files APIs
- Adding exception handling for Vector Stores in RAG Tool
- Add more tests on migration from RAG Tool to Vector Stores
- Migrate off of inference_api for context_retriever for RAG
<!-- If resolving an issue, uncomment and update the line below -->
<!-- Closes #[issue-number] -->
## Test Plan
Integration and unit tests added
Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>
Fixes#3370
AWS switched to requiring region-prefixed inference profile IDs instead
of foundation model IDs for on-demand throughput. This was causing
ValidationException errors.
Added auto-detection based on boto3 client region to convert model IDs
like meta.llama3-1-70b-instruct-v1:0 to
us.meta.llama3-1-70b-instruct-v1:0 depending on the detected region.
Also handles edge cases like ARNs, case insensitive regions, and None
regions.
Tested with this request.
```json
{
"model_id": "meta.llama3-1-8b-instruct-v1:0",
"messages": [
{
"role": "system",
"content": "You are a helpful assistant."
},
{
"role": "user",
"content": "tell me a riddle"
}
],
"sampling_params": {
"strategy": {
"type": "top_p",
"temperature": 0.7,
"top_p": 0.9
},
"max_tokens": 512
}
}
```
<img width="1488" height="878" alt="image"
src="https://github.com/user-attachments/assets/0d61beec-3869-4a31-8f37-9f554c280b88"
/>
# What does this PR do?
Adds a write worker queue for writes to inference store. This avoids
overwhelming request processing with slow inference writes.
## Test Plan
Benchmark:
```
cd /docs/source/distributions/k8s-benchmark
# start mock server
python openai-mock-server.py --port 8000
# start stack server
LLAMA_STACK_LOGGING="all=WARNING" uv run --with llama-stack python -m llama_stack.core.server.server docs/source/distributions/k8s-benchmark/stack_run_config.yaml
# run benchmark script
uv run python3 benchmark.py --duration 120 --concurrent 50 --base-url=http://localhost:8321/v1/openai/v1 --model=vllm-inference/meta-llama/Llama-3.2-3B-Instruct
```
## RPS from 21 -> 57
# What does this PR do?
<!-- Provide a short summary of what this PR does and why. Link to
relevant issues if applicable. -->
Fix pre-commit issues: non executable shebang file, @pytest.mark.asyncio
decorator
<!-- If resolving an issue, uncomment and update the line below -->
<!-- Closes #[issue-number] -->
## Test Plan
<!-- Describe the tests you ran to verify your changes with result
summaries. *Provide clear instructions so the plan can be easily
re-executed.* -->
# What does this PR do?
<!-- Provide a short summary of what this PR does and why. Link to
relevant issues if applicable. -->
The test_query_adds_vector_db_id_to_chunk_metadata test was failing
because MemoryToolRuntimeImpl.__init__() now requires a files_api
parameter.
Fixes failing unit tests for Python 3.12 and 3.13.
<!-- If resolving an issue, uncomment and update the line below -->
<!-- Closes #[issue-number] -->
## Test Plan
<!-- Describe the tests you ran to verify your changes with result
summaries. *Provide clear instructions so the plan can be easily
re-executed.* -->
# What does this PR do?
When running RAG in a multi vector DB setting, it can be difficult to
trace where retrieved chunks originate from. This PR adds the
`vector_db_id` into each chunk’s metadata, making it easier to
understand which database a given chunk came from. This is helpful for
debugging and for analyzing retrieval behavior of multiple DBs.
Relevant code:
```python
for vector_db_id, result in zip(vector_db_ids, results):
for chunk, score in zip(result.chunks, result.scores):
if not hasattr(chunk, "metadata") or chunk.metadata is None:
chunk.metadata = {}
chunk.metadata["vector_db_id"] = vector_db_id
chunks.append(chunk)
scores.append(score)
```
## Test Plan
* Ran Llama Stack in debug mode.
* Verified that `vector_db_id` was added to each chunk’s metadata.
* Confirmed that the metadata was printed in the console when using the
RAG tool.
---------
Co-authored-by: are-ces <cpompeia@redhat.com>
Co-authored-by: Francisco Arceo <arceofrancisco@gmail.com>
# What does this PR do?
This PR adds support for OpenAI Prompts API.
Note, OpenAI does not explicitly expose the Prompts API but instead
makes it available in the Responses API and in the [Prompts
Dashboard](https://platform.openai.com/docs/guides/prompting#create-a-prompt).
I have added the following APIs:
- CREATE
- GET
- LIST
- UPDATE
- Set Default Version
The Set Default Version API is made available only in the Prompts
Dashboard and configures which prompt version is returned in the GET
(the latest version is the default).
Overall, the expected functionality in Responses will look like this:
```python
from openai import OpenAI
client = OpenAI()
response = client.responses.create(
prompt={
"id": "pmpt_68b0c29740048196bd3a6e6ac3c4d0e20ed9a13f0d15bf5e",
"version": "2",
"variables": {
"city": "San Francisco",
"age": 30,
}
}
)
```
### Resolves https://github.com/llamastack/llama-stack/issues/3276
## Test Plan
Unit tests added. Integration tests can be added after client
generation.
## Next Steps
1. Update Responses API to support Prompt API
2. I'll enhance the UI to implement the Prompt Dashboard.
3. Add cache for lower latency
---------
Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>
# What does this PR do?
Add Kubernetes authentication provider support
- Add KubernetesAuthProvider class for token validation using Kubernetes
SelfSubjectReview API
- Add KubernetesAuthProviderConfig with configurable API server URL, TLS
settings, and claims mapping
- Implement authentication via POST requests to
/apis/authentication.k8s.io/v1/selfsubjectreviews endpoint
- Add support for parsing Kubernetes SelfSubjectReview response format
to extract user information
- Add KUBERNETES provider type to AuthProviderType enum
- Update create_auth_provider factory function to handle 'kubernetes'
provider type
- Add comprehensive unit tests for KubernetesAuthProvider functionality
- Add documentation with configuration examples and usage instructions
The provider validates tokens by sending SelfSubjectReview requests to
the Kubernetes API server and extracts user information from the
userInfo structure in the response.
<!-- If resolving an issue, uncomment and update the line below -->
<!-- Closes #[issue-number] -->
## Test Plan
<!-- Describe the tests you ran to verify your changes with result
summaries. *Provide clear instructions so the plan can be easily
re-executed.* -->
What This Verifies:
Authentication header validation
Token validation with Kubernetes SelfSubjectReview and kubernetes server
API endpoint
Error handling for invalid tokens and HTTP errors
Request payload structure and headers
```
python -m pytest tests/unit/server/test_auth.py -k "kubernetes" -v
```
Signed-off-by: Akram Ben Aissi <akram.benaissi@gmail.com>
# What does this PR do?
update Groq inference provider to use OpenAIMixin for openai-compat
endpoints
changes on api.groq.com -
- json_schema is now supported for specific models, see
https://console.groq.com/docs/structured-outputs#supported-models
- response_format with streaming is now supported for models that
support response_format
- groq no longer returns a 400 error if tools are provided and
tool_choice is not "required"
## Test Plan
```
$ GROQ_API_KEY=... uv run llama stack build --image-type venv --providers inference=remote::groq --run
...
$ LLAMA_STACK_CONFIG=http://localhost:8321 uv run --group test pytest -v -ra --text-model groq/llama-3.3-70b-versatile tests/integration/inference/test_openai_completion.py -k 'not store'
...
SKIPPED [3] tests/integration/inference/test_openai_completion.py:44: Model groq/llama-3.3-70b-versatile hosted by remote::groq doesn't support OpenAI completions.
SKIPPED [3] tests/integration/inference/test_openai_completion.py:94: Model groq/llama-3.3-70b-versatile hosted by remote::groq doesn't support vllm extra_body parameters.
SKIPPED [4] tests/integration/inference/test_openai_completion.py:73: Model groq/llama-3.3-70b-versatile hosted by remote::groq doesn't support n param.
SKIPPED [1] tests/integration/inference/test_openai_completion.py💯 Model groq/llama-3.3-70b-versatile hosted by remote::groq doesn't support chat completion calls with base64 encoded files.
======================= 8 passed, 11 skipped, 8 deselected, 2 warnings in 5.13s ========================
```
---------
Co-authored-by: raghotham <rsm@meta.com>
# What does this PR do?
This change migrates the VectorDB id generation to Vector Stores.
This is a breaking change for **_some users_** that may have application
code using the `vector_db_id` parameter in the request of the VectorDB
protocol instead of the `VectorDB.identifier` in the response.
By default we will now create a Vector Store every time we register a
VectorDB. The caveat with this approach is that this maps the
`vector_db_id` → `vector_store.name`. This is a reasonable tradeoff to
transition users towards OpenAI Vector Stores.
As an added benefit, registering VectorDBs will result in them appearing
in the VectorStores admin UI.
### Why?
This PR makes the `POST` API call to `/v1/vector-dbs` swap the
`vector_db_id` parameter in the **request body** into the VectorStore's
name field and sets the `vector_db_id` to the generated vector store id
(e.g., `vs_038247dd-4bbb-4dbb-a6be-d5ecfd46cfdb`).
That means that users would have to do something like follows in their
application code:
```python
res = client.vector_dbs.register(
vector_db_id='my-vector-db-id',
embedding_model='ollama/all-minilm:l6-v2',
embedding_dimension=384,
)
vector_db_id = res.identifier
```
And then the rest of their code would behave, including `VectorIO`'s
insert protocol using `vector_db_id` in the request.
An alternative implementation would be to just delete the `vector_db_id`
parameter in `VectorDB` but the end result would still require users
having to write `vector_db_id = res.identifier` since
`VectorStores.create()` generates the ID for you.
So this approach felt the easiest way to migrate users towards
VectorStores (subsequent PRs will be added to trigger `files.create()`
and `vector_stores.files.create()`).
## Test Plan
Unit tests and integration tests have been added.
Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>
# What does this PR do?
Improved bedrock provider config to read from environment variables like
AWS_ACCESS_KEY_ID. Updated all
fields to use default_factory with lambda patterns like the nvidia
provider does.
Now the environment variables work as documented.
Closes#3305
## Test Plan
Ran the new bedrock config tests:
```bash
python -m pytest tests/unit/providers/inference/bedrock/test_config.py
-v
Verified existing provider tests still work:
python -m pytest tests/unit/providers/test_configs.py -v
What does this PR do?
Fixes error handling when MCP server connections fail. Instead of
returning generic 500 errors, now provides
descriptive error messages with proper HTTP status codes.
Closes#3107
Test Plan
Before fix:
curl -X GET
"http://localhost:8321/v1/tool-runtime/list-tools?tool_group_id=bad-mcp-server"
Returns: {"detail": "Internal server error: An unexpected error
occurred."} (500)
After fix:
curl -X GET
"http://localhost:8321/v1/tool-runtime/list-tools?tool_group_id=bad-mcp-server"
Returns: {"error": {"detail": "Failed to connect to MCP server at
http://localhost:9999/sse: Connection
refused"}} (502)
Tests:
- Added unit test for ConnectionError → 502 translation
- Manually tested with unreachable MCP servers (connection refused)
One needed to specify record-replay related environment variables for
running integration tests. We could not use defaults because integration
tests could be run against Ollama instances which could be running
different models. For example, text vs vision tests needed separate
instances of Ollama because a single instance typically cannot serve
both of these models if you assume the standard CI worker configuration
on Github. As a result, `client.list()` as returned by the Ollama client
would be different between these runs and we'd end up overwriting
responses.
This PR "solves" it by adding a small amount of complexity -- we store
model list responses specially, keyed by the hashes of the models they
return. At replay time, we merge all of them and pretend that we have
the union of all models available.
## Test Plan
Re-recorded all the tests using `scripts/integration-tests.sh
--inference-mode record`, including the vision tests.
# What does this PR do?
add the ability to use inequalities in the where clause of the sqlstore.
this is infrastructure for files expiration.
## Test Plan
unit tests
# What does this PR do?
During env var replacement, we're implicitly converting all config types
to their apparent types (e.g., "true" to True, "123" to 123). This may
be arguably useful for when doing an env var substitution, as those are
always strings, but we should definitely avoid touching config values
that have explicit types and are uninvolved in env var substitution.
## Test Plan
Unit
Recording files use a predictable naming format, making the SQLite index
redundant. The binary SQLite file was causing frequent git conflicts.
Simplify by calculating file paths directly from request hashes.
Signed-off-by: Derek Higgins <derekh@redhat.com>
# What does this PR do?
As described in #3134 a langchain example works against openai's
responses impl, but not against llama stack's. This turned out to be due
to the order of the inputs. The langchain example has the two function
call outputs first, followed by each call result in turn. This seems to
be valid as it is accepted by openai's impl. However in llama stack,
these inputs are converted to chat completion inputs and the resulting
order for that api is not accpeted by openai.
This PR fixes the issue by ensuring that the converted chat completions
inputs are in the expected order.
Closes#3134
## Test Plan
Added unit and integration tests. Verified this fixes original issue as
reported.
---------
Signed-off-by: Gordon Sim <gsim@redhat.com>
# What does this PR do?
<!-- Provide a short summary of what this PR does and why. Link to
relevant issues if applicable. -->
This PR removes `init()` from `LlamaStackAsLibrary`
Currently client.initialize() had to be invoked by user.
To improve dev experience and to avoid runtime errors, this PR init
LlamaStackAsLibrary implicitly upon using the client.
It prevents also multiple init of the same client, while maintaining
backward ccompatibility.
This PR does the following
- Automatic Initialization: Constructor calls initialize_impl()
automatically.
- Client is fully initialized after __init__ completes.
- Prevents consecutive initialization after the client has been
successfully initialized.
- initialize() method still exists but is now a no-op.
<!-- If resolving an issue, uncomment and update the line below -->
<!-- Closes #[issue-number] -->
fixes https://github.com/meta-llama/llama-stack/issues/2946
---------
Signed-off-by: Mustafa Elbehery <melbeher@redhat.com>
Adds flexible CORS (Cross-Origin Resource Sharing) configuration support
to the FastAPI
server with both local development and explicit configuration modes:
- **Local development mode**: `cors: true` enables localhost-only access
with regex
pattern `https?://localhost:\d+`
- **Explicit configuration mode**: Specific origins configuration with
credential support
and validation
- Prevents insecure combinations (wildcards with credentials)
- FastAPI CORSMiddleware integration via `model_dump()`
Addresses the need for configurable CORS policies to support web
frontends and
cross-origin API access while maintaining security.
Closes#2119
## Test Plan
1. Ran Unit Tests.
2. Manual tests: FastAPI middleware integration with actual HTTP
requests
- Local development mode localhost access validation
- Explicit configuration mode origins validation
- Preflight OPTIONS request handling
Some screenshots of manual tests.
<img width="1920" height="927" alt="image"
src="https://github.com/user-attachments/assets/79322338-40c7-45c9-a9ea-e3e8d8e2f849"
/>
<img width="1911" height="1037" alt="image"
src="https://github.com/user-attachments/assets/1683524e-b0c9-48c9-a0a5-782e949cde01"
/>
cc: @leseb @rhuss @franciscojavierarceo
# What does this PR do?
Handles MCP tool calls in a previous response
Closes#3105
## Test Plan
Made call to create response with tool call, then made second call with
the first linked through previous_response_id. Did not get error.
Also added unit test.
Signed-off-by: Gordon Sim <gsim@redhat.com>
# What does this PR do?
This PR adds a step in pre-commit to enforce using `llama_stack` logger.
Currently, various parts of the code base uses different loggers. As a
custom `llama_stack` logger exist and used in the codebase, it is better
to standardize its utilization.
Signed-off-by: Mustafa Elbehery <melbeher@redhat.com>
Co-authored-by: Matthew Farrellee <matt@cs.wisc.edu>
# What does this PR do?
Refactors the OpenAI response conversion utilities by moving helper functions from `openai_responses.py` to `utils.py`. Adds unit tests.
# What does this PR do?
Refactors the OpenAI responses implementation by extracting streaming and tool execution logic into separate modules. This improves code organization by:
1. Creating a new `StreamingResponseOrchestrator` class in `streaming.py` to handle the streaming response generation logic
2. Moving tool execution functionality to a dedicated `ToolExecutor` class in `tool_executor.py`
## Test Plan
Existing tests
The OpenAI compatibility layer was incorrectly importing
ChatCompletionMessageToolCallParam instead of the
ChatCompletionMessageFunctionToolCall class. This caused "Cannot
instantiate typing.Union" errors when processing agent requests with
tool calls.
Closes: #3141
Signed-off-by: Derek Higgins <derekh@redhat.com>
# What does this PR do?
Adds content part streaming events to the OpenAI-compatible Responses API to support more granular streaming of response content. This introduces:
1. New schema types for content parts: `OpenAIResponseContentPart` with variants for text output and refusals
2. New streaming event types:
- `OpenAIResponseObjectStreamResponseContentPartAdded` for when content parts begin
- `OpenAIResponseObjectStreamResponseContentPartDone` for when content parts complete
3. Implementation in the reference provider to emit these events during streaming responses. Also emits MCP arguments just like function call ones.
## Test Plan
Updated existing streaming tests to verify content part events are properly emitted
