Skip to content
Agetor
v1.2.1Release notes

The harness orchestrator.

Run Claude Code, Codex, Cursor, Gemini CLI, fx and Antigravity in parallel from one local kanban board. Every task gets its own git worktree, and questions and approvals land as cards you answer in one place.

  • Free & open source (MIT)
  • Local-first, no account
  • Updates itself
The Agetor board with six columns: Backlog, Ready, Running, Blocked, Review and Done. Cards show a task type, the harness and model, a git branch and Run, Stop, Answer, Open or Done buttons. A header row lists the six harnesses with their status. Highlights: Six harnesses, each with live status; Pipeline run: step 2 of 4, now in Review; A question waits for you: Answer; Every task has its own branch and worktree; Open the worktree, or mark it Done.

Drives the agent CLIs you already use

  • Claude Code

    claude

    stable
  • Codex

    codex

    experimental
  • Cursor

    cursor-agent

    experimental
  • Gemini CLI

    gemini

    experimental
  • fx.sh · Vercel Labs

    fx

    experimental
  • Antigravity

    agy

    experimental

Claude Code is stable. Codex, Cursor, Gemini CLI, fx and Antigravity are experimental.

01 · features

Everything an agent run needs, on one board

Plan, run, review and ship parallel coding agents from a single local app. Every screenshot here is the real app.

The New task form beside the board. It has Task, Bug and Spike types, a title and prompt with @ file autocomplete, a project, branch, the Isolate (worktree) option, a Pipeline, an Agent, the six harnesses, Code and Plan modes, model and effort, and To backlog and Run task buttons.

A kanban board that moves by itself

Cards move through Backlog, Ready, Running, Blocked, Review and Done as runs start, stall, fail or finish, and you can drag a card to override the flow. Create one from the New task form: a Task, Bug or Spike with its project, branch and harness.

  • Isolate (worktree) gives each task its own branch and git worktree, from a base commit pinned when the task is created
  • Type @ to pick files and folders, or / to complete slash commands, skills, MCP servers and plugins
  • Launch from a saved Agent or a Pipeline instead of choosing the harness, model, effort and mode by hand

And the rest of the app

The smaller controls you'll reach for every day, under the names the app gives them.

  • Worktrees

    One list of every task's worktree with its branch, project and state: running, uncommitted or merged. Clean up the ones you're done with.

  • Import Claude Code sessions

    Bring Claude Code sessions you started outside Agetor onto the board. Agetor can look for them each time it launches.

  • Clone repository

    Clone from GitHub, GitLab or Bitbucket Cloud with live progress and cancel, and optionally have an agent write an ELI5.md tour of the code.

  • Git

    Pull requests and issues for your projects in one dialog, with search, labels, assignee filters and New PR.

  • Saved Prompts

    Keep the prompts you reuse in Settings, and insert them from the composer's MCP · Skills · Plugins · Prompts picker.

  • Usage meter

    Plan quota for Claude Code, Codex and Cursor accounts, read with the logins those CLIs already have, plus local token rollups.

  • Notifications

    Native macOS notifications open the task that needs you, and a card shows an unread dot when its agent has said something new.

  • Themes

    Auto follows your system, or pick Dark or Light. Scale the text with ⌘= and ⌘−, and reset it with ⌘0.

  • Model discovery

    Pickers list the models your account actually offers. If a model needs a newer CLI than yours, Agetor refuses the run with an upgrade hint.

How it works

  1. 01

    Write a card

    Click New task, pick a project folder, then choose an Agent, or a harness, model and mode, and write the prompt. Run task starts it now; To backlog queues it.

  2. 02

    Agetor runs it in a worktree

    With Isolate (worktree) on, each task gets its own branch and git worktree. Agetor launches the agent's real CLI there and streams the conversation to the card's run panel.

  3. 03

    Answer, review, ship

    Answer questions as cards, review the diff and send lines back, then Commit & push and open a pull request.

02 · harnesses

Agetor drives the real CLI. It never re-implements the agent.

Every harness runs as its own binary with your existing login. That includes Claude Code, which Agetor runs in interactive mode only.

How each harness runs
Claude Codeclaudestable

Interactive session in a per-task tmux session. Output is read from the transcript Claude Code writes itself.

No claude -p, no Agent SDK.

CodexcodexexperimentalCursorcursor-agentexperimentalGemini CLIgeminiexperimentalAntigravityagyexperimental

One headless turn per message inside tmux, resumed by the harness's own thread, session or conversation id.

fx.shfxexperimental

Agent Client Protocol over stdio, through Vercel AI Gateway. Runs as a child process of Agetor.

How usage is billed depends on each provider's terms. Agetor stays on the same path you'd take by hand.

03 · why agetor

Runs your agents. Doesn't replace them.

Claude Code is still Claude Code, with the same login, skills, MCP servers, slash commands and settings. Agetor adds parallel runs, isolation and one place to review.

Local-first

Everything runs on your Mac. No Agetor account, no cloud relay, no telemetry. State lives under ~/.agetor.

Non-invasive

No hooks, no MCP server, no edits to your CLAUDE.md. Agetor drives the real CLI the way you would.

Restart-safe

tmux-hosted sessions keep running when you quit, and Agetor reconnects to them on the next launch.

Locked to localhost

The local API listens only on 127.0.0.1 and requires a fresh random token on every launch.

Easy to extend

Adding a harness takes three steps across three files. The orchestrator, UI and CLI pick it up automatically.

Open source

MIT licensed and built in the open, mostly by Claude Code running inside Agetor.

04 · install

Install in a minute

Download the app, add the CLI, or build it from source.

Recommended

Desktop app

The signed, notarized .dmg updates itself in place.

  1. Drag Agetor into Applications and open it
  2. Install a harness and log in once, for example by running claude
Download for macOS

Apple Silicon · signed & notarized

Command-line interface

One command installs the agetor binary after verifying its SHA-256 checksum. Works with or without the app.

~ · zsh
curl -fsSL https://github.com/alamops/agetor/releases/latest/download/install.sh | sh
agetor
full-screen live dashboard
Full CLI reference

Build from source

For contributors. Needs Bun 1.3.5 or later, Git, tmux and at least one harness CLI.

~/code · zsh
git clone https://github.com/alamops/agetor.git
cd agetor
bun install
bun run dev:hmr

Dev builds use ~/.agetor-dev, so they never touch your installed app. bun run build needs a Developer ID, so only maintainers use it.

Agetor ships for macOS on Apple Silicon. Linux and Windows aren't built or tested yet. No harness yet? Run npm i -g @anthropic-ai/claude-code.

05 · faq

Frequently asked questions

Local execution, supported harnesses, git worktrees and source access.

What is Agetor?

Agetor is a free, open-source macOS app that turns a kanban board into a control plane for AI coding agents. Each card is a prompt, a repository and a harness. Agetor runs the agent's own CLI in an isolated git worktree and streams the conversation to the board.

Which coding agents does Agetor support?

Six harnesses: Claude Code, which is stable, plus Codex, Cursor, Gemini CLI, fx from Vercel Labs and Antigravity, which are experimental. Agetor launches the real CLI on your PATH with your existing login, and you can add extra accounts for any harness, each with its own isolated home directory.

How does Agetor run Claude Code?

As an interactive session inside tmux, the same way you would run it by hand. Agetor types prompts as keystrokes and reads structured output from the transcript Claude Code writes itself. It doesn't use claude -p or the Claude Agent SDK, installs no hooks or MCP servers, and leaves your CLAUDE.md alone.

Does Agetor send my code through a cloud relay?

No. Everything runs on your Mac, with no Agetor account, no relay and no telemetry. State lives under ~/.agetor, and the local API listens only on 127.0.0.1 behind a token generated at each launch. The agents keep the same privileges and network behavior they have when you launch them yourself.

How do parallel agents avoid conflicts?

Each task runs on its own branch in a dedicated git worktree, with its base commit pinned when the task is created. Several agents can work on the same repository at once without sharing a checkout.

What is the difference between a harness and an Agent?

A harness is the agent CLI that Agetor drives, such as Claude Code or Codex, and you turn each one on or off in Settings. An Agent is a saved setup for a harness: a model, effort, mode, speed, instructions and skills. Launch a task from an Agent in one click, or chain Agents into a Pipeline. In the CLI, an Agent is called a profile.

What are Pipelines?

A pipeline chains saved Agents into a reusable graph, such as Plan, Implement and Review. Steps hand context to each other through structured handoffs, and they can branch, fan out and loop, all from a single board card.

Is there a CLI?

Yes. The agetor CLI drives the same board from your shell, with a full-screen live dashboard and commands to create, run, answer and review tasks. It works with or without the desktop app, and one curl command installs it from GitHub Releases.

Which platforms are supported, and can I build from source?

Agetor ships for macOS on Apple Silicon. Linux and Windows aren't built or tested yet. The code is open source: contributors can clone the repository and run bun install and bun run dev:hmr. Signed release builds need a Developer ID, so most people should download the notarized app.

One terminal at a time is the bottleneck.

Put every agent on the board, give each one a worktree, and review the work in one place.