Orca
GitHub
A new tool called Orca lets devs run multiple AI coding agents like Claude Code and Codex at once, each in its own workspace. It's basically mission control for juggling AI coders, and you can steer them from your phone.
Based on reporting by GitHub — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
There's a specific kind of chaos that comes with running more than one AI coding agent at a time — different terminals, different git branches, no single view of what's actually happening. Orca is built to fix exactly that problem, and it does it by treating each agent like a tenant in its own apartment: isolated git worktrees, separate terminals, all managed from one desktop app.
The headline trick is fanning a single prompt out across five agents simultaneously, each working in its own worktree, so you can compare results side by side and just merge whichever one actually solved the problem. That's a fairly blunt way to deal with the reality that Claude Code, Codex, OpenCode, and the rest all have different strengths — instead of picking one and hoping, you run them in parallel and let the results argue it out.
Orca also leans hard into removing friction around review. There's a design mode that lets you click on a UI element in a live Chromium window and pipe its HTML, CSS, and a screenshot directly into an agent's prompt, which is a genuinely clever shortcut for the usual copy-paste dance between browser devtools and a chat window. Diff annotation works the same way — drop a comment on a line of an AI-generated diff and it goes straight back to the agent, no context switch to GitHub required. GitHub and Linear are baked in natively too, so PRs and issues live inside the same window as the code.
A mobile companion app rounds things out, letting you check on and nudge agents from your phone — useful if you've got five worktrees running against a beefy remote SSH box and don't want to be glued to a laptop waiting for a notification. Orca supports basically any CLI-based agent, from Codex and Cursor to smaller names like Goose and Kilocode, and the whole thing is MIT-licensed and open source, with builds available for macOS, Windows, and Linux plus a headless server mode.
What's notable is the pace: the team says they ship daily, which tracks with how quickly this space is moving — a tool built around orchestrating agents has to keep up with agents themselves getting better every few weeks.
My take — AI-written commentary, not fact-checked reporting
This is the natural next layer once you accept that no single model or agent wins every coding task — you just run several and pick the winner, which is a very brute-force, very effective way to deal with the fact that model quality is still inconsistent. I like that it's open source and agent-agnostic rather than locking you into one vendor's CLI tool, that's the right instinct in a market where everyone from Anthropic to xAI is shipping a coding agent every quarter. The real test will be whether running five agents in parallel on the same prompt burns through API credits fast enough to make this a rich person's toy.
Read more about this at: GitHub