Agent Orchestrator (Orchestrator.inc) is a local desktop workspace for running several coding agents across one repository. It gives each task its own agent, workspace, and feedback loop, and tracks every worker, pull request, CI run, and review on a live Kanban board. A local daemon behind the desktop app watches agent activity and source-control state.
Work happens at two levels. A worker is one task, one coding agent, and one isolated workspace: for Git-backed work it gets its own branch and worktree, and standalone agents get a managed directory without a repository. You talk to a worker in structured Chat or in the agent's native terminal UI. The project orchestrator is a persistent planning agent that keeps a project-scoped conversation of goals, decisions, and constraints, turns an outcome into a plan, spawns or redirects workers with the context they need, and coordinates follow-up work.
The orchestrator watches every pull request it opens. Failed checks, review comments, and merge conflicts go back to the worker session that owns the branch until the pull request is approved. Board columns (working, needs you, in review, ready to merge) are derived from session, pull request, CI, and review state rather than set by hand.
About 25 coding agent harnesses are supported, with per-project agent choice, including Claude Code, Codex, Cursor, OpenCode, Aider, Goose, GitHub Copilot, Cline, and Qwen. A mobile companion app pairs with the desktop over LAN or Tailscale to watch sessions and receive notifications, while execution and code stay on the local machine.
Features
- Isolated workers: one task, one agent, and one branch and worktree per worker
- Project orchestrator: a persistent planning agent that splits outcomes into tasks and delegates them to workers
- Live Kanban: cards placed by session, pull request, CI, and review state, with agent, branch, and activity on each card
- CI and review feedback loop: failed checks and review comments routed back to the worker that owns the branch
- Agent reviews: CI, mergeability, reviewer state, and interactive agent reviews shown beside each worker
- Agent-controllable browser: an in-app browser preview per worker, with isolated temporary profiles or reusable named profiles
- Chat or native terminal: structured Chat or the agent's own terminal UI under one supervisor
- Many harnesses: about 25 coding agents supported, chosen per project
- Mobile companion: watch sessions, open a terminal, and get notified from a phone paired over LAN or Tailscale
