Skip to content

v0.30.0 — Agent Manager: parallel isolated runs

Choose a tag to compare

View all tags
@brcampidelli brcampidelli released this 16 Jul 02:38
v0.30.0
334199b

The Agent Manager — run several coding tasks at once, each isolated, with a live board and an honest conflict report. This is the parallel, agent-first surface, built on Chimera's existing git-worktree isolation.

What's new

  • New "Agents" screen + POST /api/agents. Queue several tasks and run them concurrently, each in its own git worktree (the same chimera solve-batch isolation, reused end to end). A live board shows one card per task with streaming status, then each task's real result — pass/fail, attempts, changed files, per-file diffs — all from the actual run, nothing fabricated.
  • Honest conflict report. A batch bar shows the merged-file count and, prominently, the real cross-task conflicts: files two or more successful tasks both changed are left unmerged for you to resolve (never silently clobbered). Outside a git repo, the UI says tasks ran in-place without isolation — no false claim.
  • Concurrency fix. The run log (runs.jsonl) is now written under a lock, so parallel-batch receipts can't interleave into a corrupt line.

Mid-run cancel/steer is intentionally not in this MVP (future work). The terminal CLI stays sovereign (chimera solve-batch does the same isolation headless). Native installers attach a few minutes after publish; also pip install 'chimera-agent[desktop]'.

Full changelog: https://github.com/brcampidelli/chimera-agent/blob/main/CHANGELOG.md

Assets 6
Loading