This release is the one that actually lands on PyPI after the move to nrl-ai/edgevox. It supersedes v1.2.0, v1.3.0, and v1.3.1 — those were tagged on GitHub but their release workflows failed on Python 3.10 and the publish job was skipped each time, so PyPI is still pinned at 1.1.2 with the old vietanhdev URLs.
What changed since 1.1.2 (effective changelog)
v1.2.0 → v1.3.1 highlights (rolled forward)
- RookApp commentary gate + LLM eval harness + model picker
- Chess-commentary benchmark + canned game-end + board / warmup fixes
- Robot tool-calling benchmark + report audits + slim-directive follow-ups
- Barge-in-safe compaction, absolute doc links, logging cleanup
- Repo moved
vietanhdev/edgevox→nrl-ai/edgevox; PyPI Homepage and Issues URLs retargeted - Docs site restyled to nrl-ai house style — light-only theme, Space Grotesk, hairline brutalism, animated marquee strip, four-corner section grid, numbered principles
edgevox.__version__brought back in sync withpyproject.toml
v1.3.2 — the actual fix
Blackboard.post_requestnow raisesconcurrent.futures.TimeoutError(the futures class) instead of stdlibTimeoutError. On 3.11+ they're aliased; on 3.10 they're distinct classes, which is why the test catch intests/harness/test_new_features.pywas failing only on 3.10 and silently blocking publish for three releases.
What did not change
No surface-level API changes. The agent harness, voice pipeline, STT/TTS backends, and skills are byte-identical to the v1.3.1 GitHub tag (everything in this release except the multiagent.py exception class).