v0.4.5

What's Changed

• chore: bump llama-stack-client to 0.4.4 in UI lockfile by @cdoern in #4791
• fix: MCP CPU spike by using context manager for session cleanup by @derekhiggins in #4851
• fix(vector_io): eliminate duplicate call for vector store registration (backport #4925) by @mergify[bot] in #4944
• chore: bump version to 0.4.5 for release by @cdoern in #4954

Full Changelog: v0.4.4...v0.4.5

v0.5.0

What's Changed

• docs: Added a new oci-llamastack notebook for how to build agents with OCI and llama stack by @omaryashraf5 in #4418
• docs: Add guide to migrating from Agents to Responses by @jwm4 in #4375
• feat: convert models API to use a FastAPI router by @nathan-weinberg in #4407
• chore: update mcp dependency constraint to >=1.23.0 by @derekhiggins in #4457
• feat(ci): added codeql scanning workflow by @gmatuz in #4462
• feat: migrate Conversations API to FastAPI router by @leseb in #4342
• fix(faiss): add backward compatibility for EmbeddedChunk deserialization by @leseb in #4463
• fix: Removed duplicate parameters from integration test by @gyliu513 in #4461
• fix: removed scan on push by @gmatuz in #4466
• chore: Document release process by @raghotham in #4470
• feat: build ARM64-based UBI starter image by @rhdedgar in #4474
• chore: add "Discussion" issue template by @nathan-weinberg in #4469
• chore: Updated test integration guide by @gyliu513 in #4460
• fix: update CONTRIBUTING.md to reflect pre-commit version used in CI by @eoinfennessy in #4468
• fix: skip resources with empty IDs from conditional env vars in config processing by @Elbehery in #4455
• fix: Fix Vector Store Integration Tests by @franciscojavierarceo in #4472
• chore: Delete CHANGELOG.md by @terrytangyuan in #4480
• ci: run ARM64 builds on nightly schedule only by @rhdedgar in #4479
• chore(github-deps): bump actions/checkout from 4.3.1 to 6.0.1 by @dependabot[bot] in #4491
• chore(github-deps): bump astral-sh/setup-uv from 7.1.6 to 7.2.0 by @dependabot[bot] in #4490
• chore(github-deps): bump docker/setup-qemu-action from 3.2.0 to 3.7.0 by @dependabot[bot] in #4489
• chore(github-deps): bump github/codeql-action from 3.31.9 to 4.31.9 by @dependabot[bot] in #4488
• chore(github-deps): bump stainless-api/upload-openapi-spec-action from 1.9.0 to 1.10.0 by @dependabot[bot] in #4487
• chore: Add backwards compatibility for Milvus Chunks by @franciscojavierarceo in #4484
• fix: aiohttp HTTP Parser auto_decompress feature susceptible to zip bomb by @leseb in #4494
• chore: Add backwards compatibility for qdrant chunks by @Ygnas in #4495
• chore: Updated CONTRIBUTING guidance for integration test by @gyliu513 in #4459
• fix: fonttools security advisory by @leseb in #4503
• chore: Add backwards compatibility for pgvector chunks by @Ygnas in #4506
• refactor!: change image_name to distro_name in StackConfig by @cdoern in #4396
• fix: Add backwards compatibility for sqlite-vec, chroma, and weaviate chunks by @ChristianZaccaria in #4502
• fix: urllib3 vulnerable to decompression-bomb safeguard bypass by @leseb in #4512
• fix: disable together banner by @cdoern in #4517
• chore: switch to monthly minor release by @leseb in #4518
• chore: change discussion template label by @nathan-weinberg in #4525
• chore: add maintenance policy to release doc by @leseb in #4514
• docs: fixed outdated links for api overview, routed to the updated links by @lalexandrh in #4524
• chore: upgrade virtualenv by @raghotham in #4585
• chore: resync client dep with main by @leseb in #4591
• fix: llama-stack-api packaging by @cdoern in #4593
• docs: add guidance for contributing new providers by @leseb in #4478
• feat: migrate post_training API to FastAPI router by @eoinfennessy in #4496
• fix(memory/rag): remove file:// uri prefix by @r-bit-rry in #4286
• fix: benchmark registration via registered_resources config by @leseb in #4600
• feat(api): migrate Eval API to FastAPI router (#4345) by @r-bit-rry in #4425
• feat: convert shields API to use a FastAPI router by @nathan-weinberg in #4412
• feat: convert datasetio API to use a FastAPI router by @nathan-weinberg in #4400
• feat: Elasticsearch integration for VectorIO by @ezimuel in #4007
• fix: enable vector store registration from config with OpenAI metadata by @are-ces in #4616
• docs: Update RAG Agent Documentation using vector_stores by @robinnarsinghranabhat in #4485
• chore(github-deps): bump github/codeql-action from 4.31.9 to 4.31.10 by @dependabot[bot] in #4640
• chore(github-deps): bump actions/cache from 5.0.1 to 5.0.2 by @dependabot[bot] in #4639
• chore(github-deps): bump docker/setup-buildx-action from 3.11.1 to 3.12.0 by @dependabot[bot] in #4638
• chore(github-deps): bump actions/setup-node from 6.1.0 to 6.2.0 by @dependabot[bot] in #4637
• fix: update responses limitations doc to track latest state by @iamemilio in #4392
• feat: Convert scoring API to use a FastAPI router by @gyliu513 in #4521
• fix: Removed unused para for test score by @gyliu513 in #4645
• fix: fix list-deps quoting in deps-only output by @gyliu513 in #4653
• feat(api): Implement connector support via static configuration by @jaideepr97 in #4263
• fix: default ollama URL in Quickstart was incorrect in 2 places by @damian0815 in #4646
• fix: unregister function first before register by @gyliu513 in #4473
• feat!: migrate safety API to FastAPI router by @r-bit-rry in #4643
• feat: convert prompts API to use a FastAPI router by @nathan-weinberg in #4649
• feat: add scheduled CI workflow for release branches by @cdoern in #4510
• fix: Fix redundant MCP tools/list calls by @jwm4 in #4634
• docs: Move demo script to step 3 for quick start by @gyliu513 in #4661
• feat: convert scoring_functions API to use FastAPI router. by @EleanorWho in #4599
• feat(ci): add Bedrock integration tests with record/replay by @skamenan7 in #4292
• feat: Core Changes for default embedding dims by @rriley99-oci in #4671
• fix: use SecretStr for AWS credentials by @eoinfennessy in #4681
• feat: Implemented reasoning.effort parameter in LLS Responses by @Nehanth in #4633
• fix: file_search_call results missing document attributes/metadata by @are-ces in #4680
• fix!: usage input_token_details and output_token_details are not optional by @mattf in #4690
• fix: completed_at is required output by @mattf in #4692
• fix: store is required output by @mattf in #4693
• docs: update contrib guidelines on PR reviews by @leseb in #4676
• docs: require test plan with script and output for API PRs by @leseb in #4659
• feat: Add OpenAI API conformance coverage analyzer by @leseb in #4668
• feat!: use global vertext API endpoint by @ktdreyer in #4674
• fix: Concurrent calls into SentenceTransformer() cause failures of cl...

v0.4.4

What's Changed

• fix: Enable session polling during streaming responses (backport #4738) by @mergify[bot] in #4756
• feat: add scheduled CI workflow for release branches (backport #4510) by @mergify[bot] in #4769
• fix: make release-branch-scheduled-ci compatible with older branches (backport #4753) by @mergify[bot] in #4767
• fix: pass branch explicitly to install-llama-stack-client action (backport #4759) by @mergify[bot] in #4763
• fix: llama-stack-api packaging by @cdoern in #4777
• feat(ci): unify PyPI/npm release workflow with dry-run support (backport #4774) by @mergify[bot] in #4785
• fix: install setuptools-scm in CI (backport #4782) by @mergify[bot] in #4786
• build: bump llama-stack-client to 0.4.4 for release by @cdoern in #4787
• fix: override version from release tag for all packages (backport #4788) by @mergify[bot] in #4789

Full Changelog: v0.4.3...v0.4.4

v0.4.3

What's Changed

• fix: enable vector store registration from config with OpenAI metadata (backport #4616) by @mergify[bot] in #4631
• fix: Fix redundant MCP tools/list calls (backport #4634) by @mergify[bot] in #4663
• fix: file_search_call results missing document attributes/metadata (backport #4680) by @mergify[bot] in #4686
• fix: Concurrent calls into SentenceTransformer() cause failures of client.vector_stores.file_batches.create() (backport #4636) by @mergify[bot] in #4698
• feat: Add shutdown functionality to LlamaStackAsLibraryClient and AsyncLlamaStackAsLibraryClient (backport #4642) by @mergify[bot] in #4733
• feat(PGVector): implement automatic creation of vector extension during initialization of PGVectorVectorIOAdapter (backport #4660) by @mergify[bot] in #4740

Full Changelog: v0.4.2...v0.4.3

v0.4.2

What's Changed

• fix: disable together banner (backport #4517) by @mergify[bot] in #4519
• fix: llama-stack-api packaging (backport #4593) by @mergify[bot] in #4596
• fix(memory/rag): remove file:// uri prefix (backport #4286) by @mergify[bot] in #4603
• fix: benchmark registration via registered_resources config (backport #4600) by @mergify[bot] in #4604

Full Changelog: v0.4.1...v0.4.2

v0.4.1

What's Changed

• fix(faiss): add backward compatibility for EmbeddedChunk deserialization (backport #4463) by @mergify[bot] in #4464
• fix: skip resources with empty IDs from conditional env vars in config processing (backport #4455) by @mergify[bot] in #4475
• chore: Add backwards compatibility for Milvus Chunks (backport #4484) by @mergify[bot] in #4493
• fix: Fix Vector Store Integration Tests (backport #4472) by @mergify[bot] in #4481
• chore: Add backwards compatibility for qdrant chunks (backport #4495) by @mergify[bot] in #4499
• fix: aiohttp HTTP Parser auto_decompress feature susceptible to zip bomb (backport #4494) by @mergify[bot] in #4498
• fix: fonttools security advisory (backport #4503) by @mergify[bot] in #4504
• chore: Add backwards compatibility for pgvector chunks (backport #4506) by @mergify[bot] in #4507
• fix: Add backwards compatibility for sqlite-vec, chroma, and weaviate chunks (backport #4502) by @mergify[bot] in #4513
• fix: urllib3 vulnerable to decompression-bomb safeguard bypass (backport #4512) by @mergify[bot] in #4515

Full Changelog: v0.4.0...v0.4.1

v0.4.0

llama-stack v0.4.0

🎯 Notable Features in llama-stack v0.4.0

⚠️ Breaking Changes + Noteworthy Functional Changes to be Aware of

Configuration Files

• StackRunConfig → StackConfig: Configuration class renamed
• run.yaml → config.yaml: Runtime configuration file renamed
• build.yaml removed: Build configuration consolidated into config.yaml
• New required config sections: vector_stores, safety, and structured storage.stores configuration

Vector Store API

• vector_db_id → vector_store_id: Renamed across all endpoints and data structures (#3923)
• ChunkMetadata now required: No longer optional in Chunk class (#4413)
• Embeddings refactored: Removed from base Chunk class, added new EmbeddedChunk class (#4413)
• Removed chunk_id property: Dropped from Chunk class (#3954)
• Search response structure changed: Updated return format (#4080)

API Removals

• Agents API removed: Sessions and turns endpoints deleted—use Responses + Conversations instead (#4055)
• SDG API Stubs removed: Synthetic data generation endpoints deleted (#4035)
• /v1/openai/v1/* routes removed: Use /v1/* routes instead (#4054)
• Deprecated v1 routes with v1alpha equivalents were removed (#4054)
• Register/unregister resources deprecated: Resource registration APIs deprecated (#4099)

Provider Configuration

• Bedrock env variable: Token variable renamed to match AWS/boto3 conventions—update your environment (#4152)
• Inference base_url standardized: Unified configuration format across providers (#4177)

API Behavior Changes

• Parallel tool calls: New parallel_tool_calls parameter affects execution flow (#4124)
• Logprobs parameter: Use include parameter instead of direct field (#4261)
• Telemetry architecture: Complete redesign around OpenTelemetry auto-instrumentation (#4127)

🏗️ Architecture Improvements

• API/Provider Separation: Split API and provider specs into separate llama-stack-api package for better modularity (#3895)
• FastAPI Migration: Converted multiple APIs to FastAPI router system (Files, Providers, Inspect, Datasets, Benchmarks) for improved performance and maintainability
• Inspect API Update: /v1/inspect now only lists v1 APIs by default (#3948)

🔍 Vector Store Enhancements

• Query Rewrite Support: Added query rewrite capabilities in vector_store.search (#4171)
• Qdrant Improvements: Hybrid and keyword search support (#4006)
• ChromaDB Enhancements: Keyword search and delete_chunk implementation (#3057)
• Persistence: Vector stores now persist across server restarts (#3977)
• Metadata & Embeddings: Return embeddings and metadata from vector store methods (#4046)

🤖 Model & Inference

• Model Discovery: List available models via provider_data header (#3968, #3928)
• New Providers:
  • OpenAI-compatible Bedrock provider (#3748)
  • OCI GenAI service integration (#3876)
  • OCI embeddings support (#4300)
• Standardized Configuration: Unified base_url for inference providers (#4177)

📋 Responses API Improvements

• Parallel tool calls support (#4124)
• tool_choice parameter (#4106)
• max_tool_calls parameter (#4062)
• Logprobs via include parameter (#4261)

🆕 New APIs

• Read-only Connectors API (#4258)
• File Processor API skeleton (#4113)
• Admin API
  • Stack Administration: New /admin API (v1alpha) for administrative operations (#4401)
  • Endpoints: Provider management, health checks, version info, and route listing
  • Deprecates: Standalone /providers and /inspect APIs (still functional for backward compatibility)

🗑️ Deprecations & Removals

• Deprecated register/unregister resource APIs (#4099)
• Removed Agents (sessions/turns) API (#4055)
• Removed SDG Stub API (#4035)

🔒 Security Fixes

• Fixed RBAC bypass vulnerabilities in model access (#4270)
• Prevented ABAC bypass in vector store operations (#4394)
• JWT token redaction in logs (#4325)

🛠️ Infrastructure & DX

• Multi-architecture builds with ARM compatibility (#4290)
• OpenTelemetry auto-instrumentation support (#4281)
• SQLite WAL mode to prevent database locking (#4048)
• File deletion permission enforcement (#4275)

What's Changed

• feat: Adding Demo script by @franciscojavierarceo in #3870
• chore: use --no-cache in Containerfile by @ehhuang in #3884
• feat: Add rerank models and rerank API change by @jiayin-nvidia in #3831
• fix(conversations)!: update Conversations API definitions (was: bump openai from 1.107.0 to 2.5.0) by @dependabot[bot] in #3847
• fix(logging): ensure logs go to stderr, loggers obey levels by @ashwinb in #3885
• fix(responses): sync conversation before yielding terminal events in streaming by @ashwinb in #3888
• chore: better error messages for moderations API by @ehhuang in #3887
• test: suppress expected error logs in SSE test by @ashwinb in #3886
• fix(ci): enable responses tests in CI; suppress expected MCP auth error logs by @ashwinb in #3889
• chore(misc): update datasets, benchmarks to use alpha, beta prefixes by @ashwinb in #3891
• chore(telemetry): add an arguement to select conatiner runtime explicitly by @ehhuang in #3896
• chore: support default model in moderations API by @ehhuang in #3890
• chore(telemetry): code cleanup by @ehhuang in #3897
• feat: Add static file import system for docs by @franciscojavierarceo in #3882
• chore: update docs for telemetry api removal by @ehhuang in #3900
• chore: update run configs by @ehhuang in #3902
• chore(ui-deps): bump @testing-library/jest-dom from 6.8.0 to 6.9.1 in /llama_stack/ui by @dependabot[bot] in #3914
• chore(ui-deps): bump @types/react-dom from 19.2.1 to 19.2.2 in /llama_stack/ui by @dependabot[bot] in #3915
• chore(ui-deps): bump @tailwindcss/postcss from 4.1.14 to 4.1.16 in /llama_stack/ui by @dependabot[bot] in #3913
• chore(ui-deps): bump @types/node from 24.8.1 to 24.9.1 in /llama_stack/ui by @dependabot[bot] in #3912
• chore(github-deps): bump astral-sh/setup-uv from 7.1.0 to 7.1.1 by @dependabot[bot] in #3906
• chore(github-deps): bump actions/upload-artifact from 4.6.2 to 5.0.0 by @dependabot[bot] in #3905
• chore: remove duplicate provider definition by @leseb in #3917
• fix(tests): limit vector store providers for record mode in CI tests by @derekhiggins in #3898
• fix: Avoid BadRequestError due to invalid max_tokens by @luis5tb in #3667
• fix!: Enhance response API support to not fail with tool calling by @luis5tb in #3385
• feat(prompts): attach prompts to storage stores in run configs by @r3v5 in #3893
• chore(package): migrate to src/ layout by @ashwinb in #3920
• feat: add provider data keys for Cerebras, Databricks, NVIDIA, and RunPod by @mattf in #3734
• fix: add missing shutdown methods to PromptServiceImpl and ConversationServiceImpl by @nathan-weinberg in #3925
• chore!: BREAKING CHANGE: vector_db_id -> vector_store_id by @ehhuang in #3923
• chore(telemetry): add grafana dashboards by @ehhuang in #3921
• chore(telemetry): more cleanup: remove apis.telemetry by @ehhuang in #3919
• fix(context): prevent provider data leak between streaming requests by @ashwinb in #3924
• chore: bump openai package version by @leseb in #3918
• chore: remove leftover llama_stack directory by @leseb in #3940
• chore(mypy): add mypy and type stub packages to dev deps by @ashwinb in #3930
• feat(responses)!: introduce OpenAI compatible prompts to Responses API by @r3v5 in #3942
• fix(mypy): resolve OpenTelemetry typing issues in telemetry.py by @ashwinb in #3931
• fix(mypy): resolve OpenTelemetry typing issues in telemetry.py by @ashwinb in #3943
• fix(mypy): resolve type issues in MongoDB, batches, and auth providers by @ashwinb in #3933
• fix(mypy): resolve model implementation typing issues by @ashwinb in #3934
• fix(mypy): resolve provider utility and testing type issues by @ashwinb in #3935
• fix(mypy): re...

v0.3.5

What's Changed

• chore(docs): Remove Llama 4 support details from README (backport #4178) by @mergify[bot] in #4323
• fix(inference): respect table_name config in InferenceStore (backport #4371) by @mergify[bot] in #4372
• fix: InferenceStore workers being cancelled on event loop change by @leseb in #4373

Full Changelog: v0.3.4...v0.3.5

v0.3.4

What's Changed

• chore: bump starlette version (backport #4158) by @mergify[bot] in #4248
• fix: uninitialised enable_write_queue by @leseb in #4264
• fix: Add policies to adapters (backport #4277) by @mergify[bot] in #4279
• fix: Avoid model_limits KeyError (backport #4060) by @mergify[bot] in #4283
• chore: bump mcp package version (backport #4287) by @mergify[bot] in #4288
• fix: RBAC bypass vulnerabilities in model access (backport #4270) by @mergify[bot] in #4285

Full Changelog: v0.3.3...v0.3.4

v0.3.3

What's Changed

• fix: allowed_models config did not filter models (backport #4030) by @mergify[bot] in #4223
• fix: Vector store persistence across server restarts (backport #3977) by @mergify[bot] in #4225
• fix: enable SQLite WAL mode to prevent database locking errors (backport #4048) by @mergify[bot] in #4226
• fix(docs): fix glob vulnerability (backport #4193) by @mergify[bot] in #4227
• fix: enforce allowed_models during inference requests (backport #4197) by @mergify[bot] in #4228
• fix: update hard-coded google model names (backport #4212) by @mergify[bot] in #4229

Full Changelog: v0.3.2...v0.3.3