Release roadmap
| Version | Status and scope |
|---|---|
| v0.23.0 | Previous provider and online-execution baseline. |
| v0.24.0 | Published RetryState, middleware/replay, and lifecycle corrections. See the migration guide. |
| v0.24.1 | Corrective release: usage accounting and verified quota, retry, and DeepSeek compatibility fixes. |
| v0.25.0 | Private attempt-stage and outcome hardening, isolated exception accounting, and SQLite maintenance evidence. See the upgrade guide. |
| v0.26.0 | Shared ownership, artifact and provider fixes, admission/lifecycle hardening, and developer diagnostics. See the migration guide. |
| v0.28.0 | Deprecates positional config arguments and enable_detailed_logging, and moves LLMCallPool to call_pool. See the migration guide. |
| v0.29.0 | Prompt-cache write tracking and pricing, Python 3.15 support, and dependency security updates. |
| v0.30.0 | Per-group prompt-cache statistics, a cache hit-rate floor, cache-key and cache-group model options, CacheGroupTracker, and a prompt-caching guide. |
| v0.31.0 | LLMCallPool fail-fast with pool.resume(), and fail-fast on expired Google credentials. |
| v0.32.0 | Last release before 1.0 and the last to support Python 3.10: public provider timing, token attachment and per-attempt config for custom strategies. |
| v1.0.0rc1 | Release candidate for 1.0: renamed to llm-minder, 0.27/0.28 deprecations removed, Python 3.11+, stable provider-output views, and the frozen stable API. See the migration guide. |
| v0.27.0 | OpenAI Responses API by default, raised SDK floors, 1.0 deprecations, and the API stability draft. See the migration guide. |
Deferred executor, storage, and API simplification work is tracked in #176.