Orca Desktop IDE: Run Claude Code, Codex, Cursor CLI in Parallel Worktrees

What is Orca

Orca is not a model and does not sell model access. It is a desktop IDE that the creators call an ADE (Agent Development Environment). Its single purpose is to take the AI coding agents developers already use — Claude Code, Codex, OpenCode, Cursor CLI — and run them side-by-side inside one window.

Backed by stablyai, a Y Combinator–funded team.

Codebase is primarily TypeScript; desktop shell uses Electron with installers for macOS, Windows, and Linux.

Licensed under MIT, free for any use.

Extremely rapid release cadence: version v1.4.152 at time of writing, with one to two new versions per day.

Breaking Down Its Capabilities

One Prompt, Five Agents Running Together

This is Orca's core workflow. A single task can be dispatched to multiple agents simultaneously. Each agent operates in its own independent git worktree, so they never interfere with each other. After they finish, you compare the resulting diffs and merge the best one.

A detail the author highlights: these are real git worktrees, not a proprietary format. You can cd into any worktree and run native git commands; Orca does not block you.

Design Mode: Point at Web Elements for AI to Edit

Orca embeds a real Chromium window. When you open a page and click a button or region, Orca captures that element's HTML, CSS, and a cropped screenshot, then injects all three into the agent's prompt. This eliminates the need to verbally describe "the second button in the top-left corner" during frontend work.

Annotate Diffs to Request Revisions

After an agent produces code, you can write a comment on any line in the diff view and send it back to the agent for a fix. Review, modification, and commit happen without leaving the window.

Native GitHub and Linear Integration

Pull requests, issues, and project boards are viewable inside Orca. From an issue you can spawn a new worktree and assign an agent to fix it, removing the context-switch between browser and editor.

SSH Worktrees: Offload Heavy Work to Remote Machines

If the local machine cannot handle many agents, you can connect a remote server via SSH. File editing, git, and terminal all work over the connection, with automatic reconnection on disconnect and built-in port forwarding.

Monitor Agents on Mobile

iOS (App Store) and Android (direct APK) apps let you receive notifications when agents finish and send follow-up instructions while away from your desk.

The project's README shows the author's own usage stats: 5,065 agents run, 26 days 10 hours of cumulative agent time, 359 PRs created — evidence of heavy dogfooding.

Orca CLI: Let Agents Control Orca

A command-line interface enables scripting entire workflows:

orca worktree create
orca snapshot
orca click
orca fill

This means agents can not only work inside Orca but also drive Orca itself. The terminal uses WebGL rendering with unlimited split panes and persistent scrollback (described as Ghostty-class). The file editor is built on the VS Code core, supporting drag-and-drop of files and images into prompts.

Installation and Setup

Prerequisite: Orca does not provide models. It launches Claude Code, Codex, etc. only if they are already installed and authenticated on your machine using your own subscriptions.

Download the installer for your OS from onorca.dev/download or GitHub Releases.

Open Orca, add a local repository.

Create a worktree, choose a base branch, select the agent, and start.

Optional: install the mobile app and pair it with the desktop for remote monitoring.

Use Cases

Let Claude Code and Codex each fix the same bug, then compare diffs and pick the winner — models compete, you judge.

Frontend styling: use Design Mode to point at an element instead of writing lengthy descriptions.

Receive mobile notifications when agents finish; send the next instruction from your phone.

Offload agent execution to a remote server when your local MacBook is overloaded.

My Take

The author's most appreciated point: Orca's positioning is clear — it sells no keys, takes no cut of model API fees, and the tool itself is open-source and free. Monetization strategy is unknown, but there is no forced bundling today.

Compared to other multi-agent managers that have appeared this year, Orca is heavier: it builds its own terminal, editor, browser, and mobile apps. The upside is a cohesive, all-in-one experience; the downside is higher memory usage (Electron plus five or six agents). The 2,000+ open issues alongside 26k stars indicate rapid adoption but also many rough edges.

Open Source Repository

https://github.com/stablyai/orca