Audited 2026-10-18 · Sources: GitHub repo + README, project docs site. Tabletop audit — nothing downloaded or executed.
The card
| Repo | stablyai/orca |
| Stars | 97.0k (GitHub API, audit date) |
| License | MIT |
| Born | March 2026 · pushed this week · 8.5k open issues (active, not abandoned) |
| What it is | A desktop app for running a fleet of coding agents side by side — Claude Code, Codex, Grok, Copilot, Amp, Rovo Dev — against real tasks |
| Install surface | brew install --cask stablyai/orca/orca (signed cask) · Android APK · optional cloud/ workspace |
What problem it solves
If you use more than one coding agent, you're already doing this job badly: switching terminals, losing context on which agent holds which task, no shared view of what's running. Orca is the dispatcher's console — one surface where the whole fleet's work is visible. It's the tool we effectively built a private version of for our own crew; seeing the same architecture win 97k stars publicly is a decent validation of the idea.
Our gate findings
| Gate | Finding |
|---|---|
| Upstream identity | Verified — stablyai org, real product site (onorca.dev), brew cask channel, docs |
| License | MIT — clean |
| Install trust | Better than average — a signed brew cask beats a curl-pipe-shell any day. The Android APK is a separate trust channel worth noting. |
| Blast radius | This is the whole finding. A fleet manager's function is spawning agent processes that run commands. Whatever your agents can reach, Orca can drive. On a credential-loaded workstation that's everything; in a scoped workspace or VM it's exactly what it should be. |
| Data boundary | Local desktop app — your task data stays local by default. The optional cloud workspace is where that changes; treat it as opt-in, not default. |
Verdict: CAUTION. Not because it's sketchy — it's a real product from a real company — but because the tool's purpose is the risk. Run it in a container or dedicated workspace. Never pointed at a machine holding raw keys.
Best for
Developers juggling multiple coding agents who want a foreman's console. If you only run one agent, you don't need a fleet manager — the overhead is the product.