Orca

Key features

• Parallel agents in isolated worktrees Fan a single prompt across five agents, compare outputs, and merge the winner, with no stashing or branch juggling between tasks. Docs frame the same pattern as a recipe: race three agents on one task and keep the winning result.[1][2]

• More than 25 bring-your-own agent CLIs Claude Code, Codex, Gemini, Grok, Cursor CLI, GitHub Copilot, and OpenCode run under the subscriptions and API keys you already hold. Usage tracking surfaces provider credit balances and rate limits, and Codex accounts can be hot-swapped without restarting sessions.[1][2][3]

• Terminal built for agent output The xterm.js-based terminal with Ghostty-style themes runs each agent in its own tab with live working, waiting, and completed states, split side by side, with sessions restored on restart. Panes and layouts persist so a fleet of agents survives a reboot intact.[1][2]

• Per-worktree browser with Design Mode Click any UI element to send its HTML, CSS, and a screenshot straight into an agent prompt for visual fixes. Each worktree gets its own browser tab with saved profiles for repeated flows.[1][2]

• Inline diff annotation Drop markdown comments on any diff line and send them back to the agent, then commit and push without leaving the window. Change attribution records which agent produced what, and hosted reviews, issues, and actions stay linked.[1][2]

• Mobile companion and remote runtimes iOS and Android apps pair with the desktop session, while SSH worktrees and a self-hosted always-on server keep agents running away from your desk. Managed servers can be viewed, updated, rolled back, or forgotten from settings.[1][2][3][4]

• CLI, automations, skills, and MCP The Orca CLI drives orchestration and scheduled automations with per-automation model settings, and a skills registry with MCP support shares context across agents. Agent hooks and memory plus hibernation keep background fleets cheap and resumable.[1][2]

• Local-first and self-hostable under MIT Repositories, worktrees, and sessions run on machines you control, prompts go straight to your providers with no model in the middle, and the app is free and open source. Git history and workspace activity form an audit trail, and SOC 2 readiness work is stated though not certified.[1][2][3]

Who will love it

• Subscribed CLI-agent developers If Claude Code or Codex is already metered on your card, Orca turns those subscriptions into a supervised fleet at no extra license cost.[1][2]

• Teams that review AI output seriously Diff annotation, per-task isolation, and PR-based landing match shops where nothing merges without human eyes.[1]

• Agent-heavy experimenters Racing three agents on one bug and keeping the winner is a first-class workflow, not a workaround.[1][2]

Who should avoid it

• No agent subscription Orca ships no models and no free inference, so there is nothing to orchestrate until you bring your own CLI access.[1][2]

• Non-technical users Worktrees, terminals, and CLI setup assume a developer; there is no no-code path through the product.[1]

• Certified compliance or managed cloud SOC 2 work is stated as readiness rather than certification, hosting is self-managed, and enterprise terms go through direct contact.[1]

Alternatives and compatibility