This document outlines the release process for Llama Stack, providing predictability for the community on feature delivery timelines and release expectations.
Llama Stack follows Semantic Versioning with three release streams:
| Release Type | Cadence | Description |
|---|---|---|
| Major (X.0.0) | Every 6-8 months | Breaking changes, major new features, architectural changes |
| Minor (0.Y.0) | Monthly | New features, non-breaking API additions, significant improvements |
| Patch (0.0.Z) | Weekly | Bug fixes, security patches, documentation updates |
Releases follow the X.Y.Z pattern:
- X (Major): Incremented for breaking changes or significant architectural updates
- Y (Minor): Incremented for new features and non-breaking enhancements
- Z (Patch): Incremented for bug fixes and minor improvements
For minor and major releases, release candidates (RC) are published before the final release:
- Format:
vX.Y.ZrcN(e.g.,v0.4.0rc1,v0.4.0rc2) - Python RC packages are published to test.pypi for community testing
- Multiple RCs may be issued until the release is stable
main: Active development branch, always contains the latest coderelease-X.Y.x: Release branches for each minor version (e.g.,release-0.3.x,release-0.4.x)- Patch releases are made from release branches
- Critical fixes are backported from
mainto active release branches using Mergify
- Issues only: Add only issues to milestones, not PRs (avoids duplicate tracking)
- Milestone creation: Create milestones for each planned minor and major release
- Small fixes: Quick-landing PRs for small fixes don't require milestone tracking
A version is released when:
- All issues in the corresponding milestone are completed, OR
- Remaining issues are moved to a future milestone with documented rationale
- Triagers manage milestones and prioritize issues
- Discussions happen in the
#triageDiscord channel - Priority decisions are reviewed in community calls
Each release has a designated Release Owner from the CODEOWNERS group who is responsible for:
- Creating a dedicated Discord thread in
#releasechannel - Coordinating testing activities
- Managing the release timeline
- Publishing release artifacts
- Announcing the release
Testing requirements scale with release type:
- Rely primarily on automated CI tests
- Quick turnaround for critical fixes
- Manual verification only for specific fix validation
- Automated CI tests must pass
- Manual feature testing for new functionality
- Documentation verification
- Community testing window: 1 week
- Release candidates published for community validation
- Comprehensive automated test suite
- Scheduled testing period with predefined test plans
- Cross-provider compatibility testing
- Performance benchmarking
- Community testing window: 2-3 weeks
- Multiple release candidates as needed
For each release, the Release Owner should complete:
- Create release-specific thread in
#releasesDiscord channel - Trigger release workflows
- Generate release notes
- Announce in
#announcementsDiscord channel
Each release includes:
- PyPI package:
llama-stackandllama-stack-client - npm package:
llama-stack-client - Docker images: Distribution images on Docker Hub
- GitHub Release: Tagged release with release notes
- Documentation: Updated docs at https://llamastack.github.io
See CONTRIBUTING.md for general contribution guidelines.
Llama Stack actively maintains the last 2 stable minor releases.
- Bug fixes: Critical bugs are backported to maintained release branches
- Security patches: Security vulnerabilities are patched in maintained releases
- Patch releases (Z-stream): Maintained releases receive regular patch releases
| Release | Status | Notes |
|---|---|---|
| Current minor (0.Y.0) | ✅ Actively maintained | Bug fixes and security patches |
| Previous minor (0.Y-1.0) | ✅ Maintained | Bug fixes and security patches |
| Older releases | ❌ Unmaintained | No backports; upgrade recommended |
If the current release is v0.4.x:
v0.4.x— Actively maintained (current)v0.3.x— Maintained (bug fixes only)v0.2.xand earlier — Unmaintained
Users on unmaintained versions are encouraged to upgrade to continue receiving fixes.