Orca

Orca is an Agent Development Environment (ADE) developed by Stably AI (San Francisco), open-sourced under the MIT license.

Core Mechanism#

Orca leverages Git worktrees as the underlying isolation sandbox, enabling developers to parallelly orchestrate 20+ CLI agents—including Claude Code, Codex, Gemini, OpenCode, Cline, Cursor, and GitHub Copilot—each running in an independent worktree without interference, with support for comparing and merging results.

Interaction Capabilities#

• Ghostty-grade Terminal: WebGL-rendered with unlimited splits, scrollback recovery on restart, and full-text search.
• Monaco Editor: VS Code-equivalent kernel with auto-save, quick open, and drag-and-drop of files/images into agent prompts.
• Design Mode Embedded Browser: Each worktree has an independent Chromium instance; in Design Mode, clicking a UI element auto-injects its HTML/CSS/screenshot into the agent's context.
• Inline Diff Annotation: Directly annotate AI-generated diffs and batch-feed them back to agents for revision.
• Multi-format Preview: PDF, images (diff mode), CSV tables, Markdown (with search).

Workflow & Remote Development#

• Deep Git/GitHub Integration: Built-in source control, PR creation, issue tracking, CI status viewing.
• SSH Remote Worktree: Connect to remote machines via SSH to run agents, with reconnection, port forwarding, and password caching.
• Orca CLI: Agents can reverse-control the IDE via CLI—creating worktrees, snapshots, simulating clicks and form fills—enabling self-driven agent workflows.
• Notifications & Checkpoints: Real-time notifications and unread markers on agent completion/blockage, with free-text checkpoints.

Usage Model#

Adopts BYOS (Bring Your Own Subscription)—use your existing Claude Code/Codex subscription without additional login. Supports one-click Codex multi-account hot-switching and visualizes API usage with rate-limit reset timers. All shortcuts (Cmd-J for cross-worktree jump, Cmd-P for quick open, etc.) are customizable.

Architecture Overview#

Built on Electron + Vite, with a TypeScript (98.2%)-dominated frontend combining React and Tailwind CSS. Configuration via orca.yaml, extensible skill system through skills/ directory. Engineering uses pnpm, Oxlint + Prettier + Husky for code quality, with E2E tests. 293 releases shipped with daily update cadence.

Installation#

• Official download: https://www.onorca.dev/
• macOS: brew install --cask stablyai/orca/orca
• Arch Linux (AUR): yay -S stably-orca-bin
• Agent skill injection: npx skills add https://github.com/stablyai/orca --skill orca-cli

To Be Confirmed#

Exact first release date, team size, long-term commercialization strategy, and MCP support details (Server/Client role) are not disclosed on public pages.