Orca: an open-source multi-agent orchestration workbench with parallel worktrees and mobile monitoring

Orca is an MIT-licensed open-source desktop tool positioned as a multi-agent orchestration workbench. Core features: dispatch one prompt to up to five agents, each running in an isolated git worktree so you can compare results and merge the winner; a built-in Ghostty-class terminal with unlimited splits; run agents on remote servers over SSH with auto-reconnect and port forwarding; and a companion mobile app for iOS and Android that receives agent completion notifications and sends follow-up instructions. It also offers a design mode—click a UI element in a Chromium window to feed HTML, CSS and a screenshot into the agent prompt—plus an Orca CLI that lets agents create worktrees, snapshots, clicks and form fills via commands. The supported agent list is long, including Claude Code, Codex, Grok, Cursor, GitHub Copilot, DeepSeek Harness, OpenCode, Qwen Code, Kimi, Cline and more; officially, anything that runs in a terminal can connect. It also has built-in account switching and usage tracking, showing Claude and Codex usage and rate-limit resets and hot-switching accounts without re-login. For developers using multiple subscription-based CLI agents, Orca offers a single entry point, eliminating manual switching across terminals and repo copies; teams that need to validate different agents on the same task in parallel can compare outputs directly via worktrees. The tool provides no model capability itself—its value depends on the chosen CLI agents and subscriptions.