stablyai/orca

Orca is a desktop orchestration shell for CLI-based coding agents — the kind that run in a terminal rather than inside an IDE extension. The core premise is that running one agent at a time is leaving throughput on the table. Orca's parallel worktrees feature lets you fan a single prompt across multiple agents simultaneously, each isolated in its own git worktree, so you can compare outputs and merge whichever result you actually want. That's the architectural bet the whole tool is built around.

The agent compatibility list is deliberately broad. Orca treats any CLI agent as a first-class citizen — Claude Code, Codex, Cursor, GitHub Copilot, Devin, Goose, Cline, Qwen Code, and roughly two dozen others are explicitly listed, with an open-ended "any CLI agent" catch-all. The integration is shallow by design: if it runs in a terminal, Orca wraps it. That's a reasonable trade-off that keeps the tool from being held hostage to any single provider's API changes.

Several features stand out as genuinely practical rather than decorative. Design Mode embeds a real Chromium window and lets you click any UI element to pipe its HTML, CSS, and a cropped screenshot directly into an agent prompt — a real shortcut for frontend iteration that would otherwise require manual copy-paste or screenshot uploads. The diff annotation workflow lets you drop inline comments on any AI-generated diff line and route them back to the agent without leaving the app. SSH worktrees let you run agents on a remote machine with auto-reconnect and port forwarding, which matters when local hardware is the bottleneck.

The mobile companion app — available on iOS and Android — lets you monitor running agents and send follow-up instructions from your phone. That's a small thing that becomes a large thing once you're routinely running agents in parallel on a remote box; you don't need to stay at your desk to babysit them.

The CLI surface (orca worktree create, snapshot, click, fill) means agents can script Orca itself, which closes a loop: agents orchestrating agents through a shared workspace. Whether that's useful or just clever depends entirely on the workflow, but the affordance is there.

What Orca is not is a model router, a prompt manager, or an evaluation harness. It doesn't touch what the agents do internally — it manages where they run and how you review their output. The telemetry is opt-out, the license is MIT, and the team describes shipping daily. The tool is free and cross-platform.