Orca has passed 95,200 GitHub stars and was on the daily Trending page on October 17, 2026, with 14,808 commits on main and two releases in two days (v1.4.227 on October 16 and v1.4.229 on the 17th). Its pitch fits in one line: run Claude Code, Codex, OpenCode or any terminal agent side by side, each in its own git worktree, and track them all from one window.
What it is and how it works
Orca is a desktop app for macOS, Windows and Linux that calls itself an AI agent orchestrator. The problem it tackles is familiar to anyone who has run several agents at once: if they share a directory they trample each other's changes, and if each lives in its own terminal you lose track of which one has finished or is waiting for input.
The fix is to isolate each agent in a git worktree, a working copy with its own branch on top of the same repository. You can send the same prompt to five agents, compare the results and keep the best one. Around that idea, Orca adds:
- WebGL-rendered terminals with splits and scrollback that survives restarts.
- A code editor, diff review with line comments that go back to the agent, and built-in GitHub PRs, issues and Linear boards.
- SSH worktrees to run agents on a remote machine.
- A CLI (
orca worktree create,snapshot,click,fill) so agents can drive Orca themselves. - A mobile app that notifies you when an agent finishes and lets you reply.
The package.json makes the stack clear: Electron with React and TypeScript, xterm.js for terminals, Monaco as the editor and node-pty for processes. It's MIT-licensed, copyright Lovecast Inc.
Installing it
There are binaries for every OS on the releases page, plus package managers:
# macOS (Homebrew)
brew install --cask stablyai/orca/orca
# Arch Linux (AUR)
yay -S stably-orca-binThen you open a repository, create a worktree per task and pick which agent runs in each. Orca doesn't bring its own models: it drives the terminal agents you already have installed and signed in.
What I like and what I don't
I like that isolation relies on a standard git primitive rather than a proprietary format: the worktrees are still there if you stop using Orca, and each branch gets reviewed and merged like any other. I also like that it works with any terminal agent instead of locking you into one.
What holds me back is the pace. The README itself says they ship daily and the feature list is always behind the changelog; with 4,300 open issues and 4,200 open PRs, the surface changes faster than anyone can audit it. And it's a tool that concentrates a lot of power: it hot-swaps Claude and Codex accounts without signing in again, offers computer use, and pairs your phone with your desktop through a relay whose code lives in the repository's cloud/ folder. Before running it on a laptop with production credentials, check what that pairing exposes, and turn off the anonymous telemetry if your policy requires it; the docs explain how.
Orca vs Claude Squad
Claude Squad is the closest alternative: a terminal app that also gives each agent its own worktree, but manages sessions with tmux.
| Orca | Claude Squad | |
|---|---|---|
| Approach | Desktop IDE with editor, diffs and browser | TUI on top of tmux |
| Agents | Any CLI | Claude Code by default, Codex, OpenCode, Amp, Gemini, Aider and custom profiles |
| Install | Desktop app, Homebrew cask, AUR | brew install claude-squad or install script |
| License | MIT | AGPL-3.0 |
| Community | 95,200 stars | 8,600 stars |
When to pick which: Orca if you want to review diffs, open PRs and follow your agents from your phone without leaving one app; Claude Squad if you live in the terminal, work over SSH or prefer a small tool you can read end to end. If you only run two agents now and then, git worktree add and two terminal tabs are still enough.
Sources
- stablyai/orca, GitHub
- Orca releases, GitHub (2026-10-17)
- Orca package.json, GitHub
- GitHub Trending (daily), GitHub (2026-10-17)
- smtg-ai/claude-squad, GitHub