Overview
Agent Orchestrator (AO) is a local desktop workspace for running several coding agents on one repository at the same time. Its unit of execution is a worker: one task, one coding agent, one isolated workspace. For Git-backed work each worker gets its own branch and worktree, and the task, conversation, terminal, changed files, browser preview, pull request, CI status and review state all stay attached to that session from start to finish.
Above the workers sits a project orchestrator — a persistent planning agent scoped to the repository rather than to a task. It is meant for the level above implementation: exploring an idea, weighing approaches, identifying high-impact work, then breaking a plan into focused tasks, spawning or redirecting workers and passing each one the context it needs. Its conversation keeps goals, decisions and constraints, and it reads live AO state (active workers, ownership, pull requests, CI, reviews) alongside repository context. The division is explicit: the orchestrator owns planning and delegation, workers own implementation, tests, commits and pull requests.
The Kanban is what makes the fleet legible. Every worker lands on the same live board — Working, Needs you, In review, Ready to merge — and AO derives each card's column from session, pull-request, CI and review facts rather than from manual status updates. Cards carry the task, agent, branch, activity and status together, and open into the conversation or native terminal UI, the diff, the PR summary, reviews and an isolated browser preview per worker.
AO is agent-agnostic: the README advertises any harness, naming Claude Code, Codex and 25+ others, and it works with your existing source-control workflow. It ships as an Apache-2.0 desktop app for macOS (Apple silicon and Intel), Windows and Linux (AppImage, deb, rpm), with a local daemon that the desktop app starts for you — no CLI required — and automatic update checks.
What it does
- One worker per task, each with its own branch and Git worktree (or an AO-managed branchless directory for scratch work)
- A project-scoped orchestrator agent that plans, splits work, spawns workers and passes them context
- A live Kanban whose columns are derived from session, pull-request, CI and review state
- Pull requests and agent reviews beside the worker, with requested changes routed back to the owning agent
- An agent-controllable browser preview with per-worker isolated profiles, so parallel UI tasks do not share state
- Harness-agnostic: Claude Code, Codex and 25+ other agent CLIs, with desktop, web, mobile and cloud agents
Getting started
AO is a desktop application backed by a local daemon. Install the build for your platform, point it at a repository, and create your first worker — the app runs the daemon itself, so there is no CLI setup step.
Download the desktop app
Builds are published on the GitHub releases page: .dmg for macOS (arm64 and x64), .exe for Windows, and AppImage, .deb or .rpm for Linux. AO checks for updates automatically.
https://github.com/Untrivial-ai/agent-orchestrator/releases/latestAdd your repository
Open AO and point it at the repository you want it to manage. The installation guide covers setting up the agent CLIs you plan to use and troubleshooting.
Create a worker for a clear task
Use New task when the work is already defined: describe the outcome, choose the coding agent and model, attach relevant files, then work with the agent in structured chat or its native terminal UI.
Or start with the orchestrator
For a larger, vaguer outcome, talk to the project orchestrator first; it turns the plan into focused tasks and delegates them to workers with the context they need.
Supervise from the board
Follow every worker on the Kanban — Working, Needs you, In review, Ready to merge — and open a card to inspect the conversation, terminal, diff, PR, CI and reviews, or to send review feedback back to the same agent.
Commands and code are distilled from the project's own documentation — always check the official repo for the latest.
When to use it
- Reach for it when more than one coding agent is working in the same repository and branches, context and reviews start colliding
- Use the orchestrator to turn an ambiguous outcome into sequenced tasks before any agent starts editing
- Watch CI failures and review comments land back on the agent that owns the branch, instead of re-explaining them yourself
- Run parallel UI tasks with isolated browser profiles so previews do not share session state
How Agent Orchestrator compares
Agent Orchestrator alongside other open-source coding-agent harnesses & runtimes tools AI/TLDR tracks, ranked by GitHub stars.
| Tool | Stars | What it does |
|---|---|---|
| CC Switch | ★ 133k | A cross-platform desktop app that manages provider configuration for eight coding-agent CLIs from one place, with a local failover proxy, MCP and skill management, and spend tracking. |
| Oh My OpenAgent | ★ 69.1k | Opinionated agent bundle for OpenCode and Codex CLI that ships preconfigured sub-agents, lifecycle hooks and built-in MCP servers behind a single ultrawork command. |
| oh-my-claudecode | ★ 39.2k | A plugin and CLI that adds multi-agent orchestration to Claude Code: staged autopilot workflows, parallel tmux workers, and advisors that route a second opinion to Codex, Gemini, Grok or Cursor. |
| herdr | ★ 39k | A single-binary terminal multiplexer built for coding agents: persistent sessions across local and SSH machines, per-pane working/blocked/idle status, and a socket API agents drive themselves. |
| AionUi | ★ 32.9k | An open-source desktop workspace that pairs a built-in agent engine with dozens of external CLI agents, adding scheduled runs, MCP tools and remote access from a WebUI or chat apps. |
| cmux | ★ 27.2k | A Ghostty-based macOS terminal built around coding agents: vertical tabs that show git branch, PR status and ports, attention notifications when an agent needs you, and a scriptable built-in browser. |
| cc-connect | ★ 15.5k | A Go daemon that bridges local coding agents — Claude Code, Codex, Cursor Agent, Gemini CLI, OpenCode, Kimi CLI and more — to Feishu, DingTalk, Slack, Telegram, Discord, LINE and WeCom, so you drive them from chat. |
| Agent Orchestrator | ★ 12.1k | Plan, run and supervise a team of coding agents from one desktop board |