Llama Stack 1.4.0
Meta · SDK / library · v1.4.0 · Released
View official release notes ↗ Copy link
What's Changed
- chore: bump fallback_version to 1.3.1.dev0 by @cdoern in #6384
- fix(starter): default trust_remote_code to false for sentence-transformers by @yashanil98 in #6318
- feat: Add Meta AI to ogx go auto-detection by @mattf in #6383
- feat: add local_api_key auth provider with attributes and startup validation by @mattf in #6400
- feat: default ogx go to localhost instead of all interfaces by @mattf in #6402
- fix(file-processor): Preserve content in async Docling uploads by @pgustafs in #6394
- docs(api): describe supported search rankers by @KXHXK in #6395
- fix(file_processors): Preserve Docling chunk structure metadata by @pgustafs in #6398
- fix(providers): stop counting cached tokens twice in the Anthropic-compat usage by @arthi-arumugam-git in #6415
- feat: Allow disabling chat completions persistence by @meffmadd in #6412
- fix(vector-io): persist vector store metadata to kvstore in Milvus, Chroma, and Weaviate by @goingforstudying-ctrl in #6371
- fix(ci): use generated ADC for Vertex recordings by @skamenan7 in #6425
- feat: implement reactive truncation for truncation=auto in Responses API by @mattf in #6410
- fix(vertexai): map "default" service_tier to None instead of "standard" by @Lifto in #6424
- feat: add ssl_mode and ca_cert_path to PostgresSqlStoreConfig by @omkarjoshi0304 in #6196
- feat: redact sensitive data from structured logs by @mattf in #6433
- feat(file-processors): Add an option to serialize Docling tables as Markdown by @pgustafs in #6408
- refactor: Harden upstream_header provider with trusted-proxy verification by @rhdedgar in #6436
- refactor: consolidate SSE streaming helpers in ogx_api.utils by @mattf in #6438
- fix(prompts): raise typed not-found errors to preserve HTTP keep-alive by @asimurka in #6440
- test(batches): skip text completions batch test for Anthropic by @derekhiggins in #6023
- fix(core): filter persisted registry objects by active providers by @Artemon-line in #6429
- fix(ci): return full test matrix for unrecognized changed file paths by @leseb in #5944
- fix(tests): use collection_path in pytest_ignore_collect for pytest 9+ by @rome-xi in #6444
- fix(vertexai): persist Gemini thought_signature across tool-call round trips by @asimurka in #6443
- fix(revert): remove remote::passthrough inference provider by @franciscojavierarceo in #6453
- fix(tests): preflight-check provider dependencies in the integration test runner by @mattf in #6471
- fix(deps): update packages for CVE remediations by @derekhiggins in #6488
- fix: repair mangled route_policy warning and existance docstring typo by @simpleqt in #6498
- fix: Add Qdrant to unit test dependencies by @varshaprasad96 in #6476
- docs: Repair broken GitHub links by @varshaprasad96 in #6477
- fix: qualify tenant column in PostgreSQL upserts by @eoinfennessy in #6509
- ci: Make tenancy configurable in the ci-tests distribution by @eoinfennessy in #6512
- feat: migrate MilvusClient to AsyncMilvusClient in vector IO provider by @mattf in #6493
- fix(docs): fix broken internal links and fail the build on new ones by @elinacse in #6478
- fix(ui): resolve critical and high severity vulnerabilities in production dependencies by @elinacse in #6479
- fix(ui): enable TypeScript build validation and fix 198 type errors by @elinacse in #6480
- fix(paper): qualify unreferenced production deployment claim by @elinacse in #6486
- docs(batches): clarify security model and add authorization tests by @elinacse in #6485
- fix: Stop integration tests when dependency installation fails. by @franciscojavierarceo in #6525
- docs: fix Docker Hub links in Kubernetes deployment docs by @tarilabs in #6513
- fix(paper): pin bibliography citations to reproducible source snapshots by @elinacse in #6487
- fix: Prepare reproducible releases after publishing exact client versions. by @franciscojavierarceo in #6526
- fix: close upstream streams when streaming wrappers are abandoned by @goingforstudying-ctrl in #6506
- fix(docs): correct list-deps CLI commands after the OGX rename by @mattf in #6459
- fix(responses): report failed instead of completed when truncation=auto exhausts retries by @elinacse in #6504
- docs: fix stale meta-llama/ogx references across docs by @simpleqt in #6494
- docs: Shorten the JOSS manuscript and correct its references. by @franciscojavierarceo in #6532
New Contributors
- @KXHXK made their first contribution in #6395
- @arthi-arumugam-git made their first contribution in #6415
- @meffmadd made their first contribution in #6412
- @Lifto made their first contribution in #6424
- @omkarjoshi0304 made their first contribution in #6196
- @rome-xi made their first contribution in #6444
- @simpleqt made their first contribution in #6498
- @tarilabs made their first contribution in #6513
Full Changelog: v1.3.1...v1.4.0
Source: Meta's official Llama Stack release notes at github.com. Text reproduced from the source for reference; all rights belong to Meta. Patchwire is not affiliated with Meta.
More Llama Stack releases
- v1.5.0 · Oct 9, 2026
- v1.5.1-dev: fix(conversations): align item id semantics with OpenAI (#6716) · Oct 8, 2026
- v1.0.4 · Oct 8, 2026
- v1.4.1-dev · Sep 23, 2026
- v1.0.3 · Sep 10, 2026
- v1.3.1 · Sep 1, 2026
- v1.2.5 · Sep 1, 2026
- v1.2.6-dev · Aug 31, 2026