First stable release.
A local-first agentic coding crew: a manager model plans a goal into subtasks, worker models implement each with real file + shell tools behind an approval gate, every subtask is verified by a runnable acceptance gate, and unverified work is surfaced — never counted as a pass. Local via Ollama; optional opt-in escalation of a failed or unverifiable subtask to Claude/Opus, off by default.
Included
- Acceptance-gated crew with completeness + spec-coverage verification.
- 24/7 unattended swarm layer (persistent queue, workers, supervisor, API + GUI).
- Research swarm and a local coding assistant.
pip-installable (pyproject.toml) with a standalone test suite.
See CHANGELOG.md for the full notes.