Orca: The IDE Built to Run a Fleet of Coding Agents in Parallel

The tooling for agentic coding has quietly split into two problems. The first — which model or agent is best — gets all the headlines. The second — how do you run several of them at once without chaos — is where the real day-to-day friction lives. Orca, an open-source project from Stably AI (Y Combinator-backed), tackles the second one head-on, and it has become a fixture in our own workflow.

An ADE, Not Another IDE

Stably calls Orca an Agent Development Environment (ADE). The distinction is deliberate. Cursor, VS Code and the rest are built around one human typing in one workspace. Orca is built around a fleet of agents working simultaneously — humans and agents both treated as first-class citizens.

The mechanism that makes this work is simple and, in hindsight, obvious: one isolated git worktree per agent. Point three agents at the same repository and each gets its own worktree and branch. They never collide. You can fan a single prompt across Claude Code, Codex and OpenCode at once and compare the three results side by side — a small tournament for every non-trivial task.

What It Actually Gives You

Beyond worktree isolation, the features are the ones you reach for once you're running more than one agent:

• 30-plus CLI agents supported — Claude Code, Codex, OpenCode, Gemini, Grok, Copilot, Cline, Goose and more. Any agent that runs in a terminal fits.
• Ghostty-class terminals — WebGL-rendered, infinite splits, full scrollback search. You watch several agents work at once without losing the plot.
• Design Mode — a real Chromium window per worktree, so you inspect the UI an agent just built without switching apps.
• SSH worktrees — run agents on a remote or on-prem machine with full file and git access.
• Diff annotation — drop markdown comments straight onto a code change and send them back to the agent as a feedback loop.
• Native GitHub / Linear integration — PRs, issues and boards without the context switch.
• Mobile companion — iOS and Android apps to monitor a running fleet from away from the desk.

The pace is part of the pitch: the team's tagline is "we ship daily," and the changelog backs it up. At the time of writing the repository sits above 28,000 GitHub stars — remarkable for a project that launched in early 2026.

Why It Fits a Sovereign Setup

Two properties make Orca a comfortable fit for the way we build. First, it is MIT-licensed and open — you can read exactly what it does with your repositories and your agent credentials, which is not a given for tools in this category. Second, SSH worktrees mean the agent can run where the code and the compute already are. Pair that with an on-prem model endpoint and the control plane stays on your own infrastructure while inference points at a self-hosted model — the same hybrid principle we apply across our agent stack.

Our Take

Orca does not replace Claude Code or Codex — it conducts them. If you have already accepted the multi-track approach to coding agents, Orca is the missing orchestration layer: it turns "several agents I switch between" into "a fleet I run in parallel," with the worktree isolation that keeps the parallelism from turning into a merge nightmare. For a small team that wants the leverage of many agents without the chaos, it is one of the most useful open-source tools to land this year. As always: bring your own accounts, keep sensitive inference on-prem, and let the tool accelerate the work, while the professional judgment stays with you.