Orca: parallel coding agents, each in its own Git worktree

My notes

Summary

Orca is a desktop app for running several AI coding agents in parallel, each isolated in its own Git worktree so their changes don’t collide. A single prompt can be fanned out to multiple agents, with the diffs compared and annotated before merging whichever result looks best, plus a mobile companion for monitoring and optional SSH worktrees for remote execution.

Key Insight

• Each agent gets its own Git worktree, so running five agents on the same repo at once doesn’t produce merge conflicts or overwritten work. Isolation is structural, not just a convention.
• “Fan one prompt across five agents” is effectively a built-in A/B test for agent output: same task, multiple models or attempts, pick the best diff instead of committing to one agent’s first pass.
• Diff annotation adds a review layer before merge, closer to a lightweight code-review step than a blind accept.
• SSH worktrees separate orchestration (local desktop UI) from execution (remote machine), which matters when agent runs are compute-heavy or need a beefier box than a laptop.
• The mobile companion is for monitoring and follow-ups on long-running jobs, not full editing. It is a notification and check-in surface rather than a second IDE.