Orca: The ADE for Running Claude Code, Codex, and Cursor in Parallel

Orca is an open-source Agent Development Environment (ADE) that runs Claude Code, Codex, Cursor and 20+ other coding agents side-by-side in isolated git worktrees, with a mobile companion app and native SSH support.

  • ⭐ 31658
  • TypeScript
  • MIT
  • Updated 2026-07-29

Compound Engineering: Orchestrate Claude Code, CodexClaude Code Subagent Patterns

Orca desktop app running agents in parallel worktrees, with the mobile companion app in the corner
Orca’s desktop app — official screenshot from github.com/stablyai/orca

What Is Orca? #

Orca bills itself as “the AI Orchestrator for 100x builders.” It’s an open-source Agent Development Environment (ADE) — not a single coding agent, but a cockpit for running several coding agents at once and comparing what they produce.

Instead of running one agent in one terminal tab, Orca lets you:

  • 🤖 Run Claude Code, Codex, Cursor, OpenCode and 20+ other CLI agents side-by-side
  • 🌳 Give each agent its own isolated git worktree, so parallel runs can’t collide
  • 📱 Monitor and steer running agents from a mobile companion app
  • 🖥️ Offload agents to a remote box over SSH, with auto-reconnect
  • 🖱️ Click any element in an embedded browser and send it straight into an agent’s prompt (Design Mode)

🔗 GitHub: https://github.com/stablyai/orca 🌐 Homepage: https://onorca.dev

Built by Stably (YC-backed) and first pushed to GitHub in March 2026, Orca crossed 31,600+ GitHub stars by late July 2026 — up from under 20,000 just two weeks earlier, one of the fastest-growing projects in the “run more than one coding agent at a time” category.


Why “Parallel Worktrees” Matters #

The core idea Orca is built around: a single well-written prompt often gets a better result from one agent than another, and it’s hard to know in advance which agent will nail it. Orca’s answer is to stop guessing:

  1. Fan one prompt across five agents at once
  2. Each agent works in its own isolated git worktree — no shared working directory, no clobbered edits
  3. Review the diffs side-by-side once they finish
  4. Merge the winner, discard the rest

This turns “which agent should I use for this task” from a one-shot bet into a comparison you can actually see.

Parallel worktree orchestration in Orca, showing multiple agents running side-by-side
Parallel worktrees — official screenshot from github.com/stablyai/orca


Key Features #

FeatureDescription
Parallel WorktreesRun the same prompt across multiple agents in isolated git worktrees, then compare and merge
Mobile CompanioniOS/Android app to monitor agents and send follow-ups from your phone
Terminal SplitsGhostty-class terminal rendering (WebGL) with infinite splits and persistent scrollback
Design ModeClick a UI element in an embedded Chromium window to send its HTML/CSS + a cropped screenshot into an agent’s prompt
SSH WorktreesRun agents on a remote server with full file editing, git, and terminal access
GitHub & Linear, NativeBrowse PRs, issues, and project boards in-app; open a worktree straight from a task
Annotate AI DiffsComment on individual diff lines and ship the feedback back to the agent
Orca CLIScript Orca itself — orca worktree create, snapshot, click, fill
Account SwitcherTrack Claude/Codex usage and rate-limit resets, hot-swap accounts without re-logging in

Supported Agents #

Orca doesn’t lock you into one vendor — it works with any CLI agent, including:

Claude Code · OpenAI Codex · Cursor · GitHub Copilot CLI · Grok CLI · OpenCode · Google Antigravity · Devin · Goose · Cline · Continue · Kilocode · Kimi · Kiro · Qwen Code · Mistral Vibe · Rovo Dev · Amp · Auggie · Command Code · Codebuff · Droid · Charm · Hermes Agent · Pi · oh-my-pi · OpenClaude · Autohand Code

You bring your own agent subscriptions — Orca doesn’t sell agent access, it orchestrates the agents you already run.


SSH Worktrees: Running Agents on a Remote Box #

One of Orca’s less obvious but genuinely useful features: agents don’t have to run on the machine in front of you. SSH Worktrees let Orca drive agents on a remote server — full file editing, git operations, and terminal access — with auto-reconnect and port forwarding handled for you. That means you can keep a laptop for reviewing and steering while the actual agent workload (and its CPU/RAM cost) runs on a beefier remote box.

For a fully headless setup with no desktop at all, Orca also ships an orca serve mode.

Remote worktrees running over SSH in Orca
SSH Worktrees — official screenshot from github.com/stablyai/orca


Install #

macOS #

brew install --cask stablyai/orca/orca

Or download the .dmg directly: Apple Silicon · Intel

Windows #

Download the .exe installer from onorca.dev/download

Linux #

# AppImage — grab the latest build
curl -LO https://github.com/stablyai/orca/releases/latest/download/orca-linux.AppImage
chmod +x orca-linux.AppImage
./orca-linux.AppImage

Or on Arch Linux via the AUR:

yay -S stably-orca-bin

Headless Linux Server #

orca serve

See the project’s headless Linux server guide for reverse-proxy and auth setup.

Mobile Companion #


Scripting Orca From the CLI #

Agents can drive Orca itself, not just run inside it. The Orca CLI exposes worktree and UI-automation primitives:

orca worktree create   # spin up a new isolated worktree for an agent
orca snapshot           # capture the current state of a worktree
orca click               # click a UI element (Design Mode automation)
orca fill                # fill a form field programmatically

Full syntax is in the CLI docs — this is enough to script “spin up N worktrees, run an agent in each, snapshot the result” without touching the GUI.


Orca vs. Running Agents Manually #

AspectOrcaTerminal tabs / tmux
Isolation between runsAutomatic git worktreesManual (you set it up)
Compare multiple agents on one taskBuilt-in side-by-sideManual diffing
Remote/SSH agentsNative, with auto-reconnectManual SSH + tmux
Mobile monitoringiOS/Android companion appNone
GitHub/Linear integrationIn-app, no context switchBrowser tab switch
LicenseMIT, open-sourceN/A

Use Cases #

1. A/B Testing Prompts Across Agents #

Send the same feature request to Claude Code and Codex simultaneously, in separate worktrees, and merge whichever implementation is cleaner.

2. Offloading Heavy Agent Runs to a Remote Box #

Use SSH Worktrees to run a long-running refactor or test suite on a rented VPS while your laptop stays responsive.

3. Reviewing AI Diffs From Your Phone #

Kick off an overnight agent run, then use the mobile companion app to check progress and approve or annotate diffs the next morning.

4. UI Work Without Leaving the Agent Loop #

Use Design Mode to click a broken component in a live preview and hand the agent its exact HTML/CSS instead of describing the bug in words.


RepositoryPurpose
Claude CodeOne of the most commonly orchestrated agents inside Orca
OpenCodeOpen-source terminal agent, also orchestrated by Orca
GooseOpen-source agent framework, listed as a supported agent


Conclusion #

Orca is less a coding agent than a control room for the coding agents you already use — Claude Code, Codex, Cursor and 20+ others — with isolated worktrees, a mobile companion, and native SSH support for remote runs. Going from under 20,000 to 31,600+ GitHub stars in about two weeks (as of late July 2026) suggests the “orchestrate many agents, not just one” model is resonating.

Best for: Developers already running multiple coding agents who want isolation between parallel runs, plus the ability to monitor and steer agents from a phone.

GitHub: https://github.com/stablyai/orca


If you want to run Orca’s SSH Worktrees against a dedicated remote box instead of your own laptop:

  • DigitalOcean — $200 free credit for 60 days across 14+ global regions. A common choice for a always-on remote worktree host.
  • HTStack — Hong Kong VPS with low-latency access from mainland China. This is the same IDC that hosts dibi8.com — battle-tested in production.

Affiliate links — they don’t cost you extra and they help keep dibi8.com running.

Last updated: 2026-07-29

References & Sources #

💬 Discussion