v0.1.0b1 — 2026-04-24¶
First beta. The 0.1.0a* series is closed. Everything below is the same engine that shipped through 0.1.0a5, just promoted from alpha to beta.
What changed¶
- Version bump:
0.1.0a5→0.1.0b1 - PyPI classifier:
Development Status :: 3 - Alpha→4 - Beta oxi-adapter-referencepinned tooxi-core==0.1.0b1
What "beta" means here¶
- No breaking changes within the
0.1.0b*line. Adapters that work on0.1.0b1will work on0.1.0b2,0.1.0b3, etc. - No promised stability across major versions.
0.2.0a1may break the adapter protocol. Check release notes. - Auto-merge defaults stay off. Operators opt in explicitly per
DispatchPolicy().auto_merge.
Why now¶
The 0.1.0a series proved the engine end-to-end:
- 55 PRs merged in one day, ~40 of them written by the engine itself
- $20.15 dogfood spend hit the daily hard-cap exactly as designed
- First-fork install path verified against PyPI from a fresh venv
- Auto-healing (auto_recover, deep_fix, engine_health, ship_recovery, worktree drift repair) all proven in production
That's enough signal to commit to no breaking changes within the line.
Install¶
--pre is still required because pip's default resolver excludes pre-releases.
What's next¶
0.1.0b2+: polish, docs improvements, T2-9 cleanup, T3-1 product-spec-to-roadmap, T3-2 compute-aware onboarding probe, plus whatever the dogfood loop's auto-observe surfaces.0.2.0(eventually): adapter protocol changes, if any are warranted.
Status of 0.1.0a*¶
All five alpha cuts (a1 through a5) remain on PyPI for archaeology but --pre will route to b1 now. New installs should not pin to alphas.
Audit trail¶
See the merged PRs from 2026-04-24 (#41 through #80) and docs/roadmap.md for the engineering work that went into this beta.