Skip to content
Terminal-native agent orchestration

Run many coding agents without losing control of your repo.

Agent Factory gives every task its own git worktree and branch, keeps every agent visible from one terminal, and lets you jump in only when a session needs judgment. The TUI, JSON CLI, and local HTTP API all share one daemon-owned state model.

Claude Code · Codex · Aider · Gemini · Amp · opencode · Devin Needs only tmux + git Linux · macOS · WSL

Home

One task, one agent, one reviewable branch

Agent Factory is built for the moment after "ask an agent" becomes "supervise several agents." It gives each normal session an isolated git worktree, keeps the agent running under daemon supervision, and brings every session back to a normal git review path.

  • Create isolated work


    Start a session for a task. Agent Factory creates a branch and git worktree, then launches your chosen agent inside it.

    Sessions and worktrees

  • Watch every agent


    The TUI shows each session, status, and Agent tab from one terminal. Scan progress without attaching to every running process.

    The TUI

  • Jump in when needed


    Interact in-pane, attach full-screen, or open helper tabs for shells, dev servers, and test watchers in the same worktree.

    How it works

  • Review and merge normally


    The result is a real branch. Use your existing git diff, pull request, CI, and merge flow instead of trusting a hidden agent workspace.

    Why Agent Factory

  • Automate recurring work


    Cron tasks and watch scripts can create sessions or deliver prompts into existing ones, hosted by the same daemon that keeps sessions alive.

    Tasks and automation

  • Script the control plane


    Every session and task operation is available through JSON CLI commands and a local owner-only HTTP API over a Unix socket.

    CLI guide

How Agent Factory Fits

Agent Factory is not a chat UI and it is not just tmux with a README. The TUI, CLI, and HTTP API are thin clients over a daemon that owns all session state, task schedules, worktree operations, and usage-limit recovery.

flowchart TB
    tui["af TUI<br/>sidebar · Agent tab · attach · tabs"]
    cli["CLI + HTTP API<br/>JSON commands · local Unix socket"]
    daemon["daemon<br/>single writer of all state<br/>sessions · tasks · usage limits"]
    sessions["sessions<br/>one git worktree + branch per agent<br/>local or remote hooks"]

    tui -->|"read projection + RPC"| daemon
    cli -->|"RPC"| daemon
    daemon -->|"owns"| sessions

That design is what lets you close the TUI, reboot, script a task, or call the API without splitting the world into different sources of truth.

Start Here

  • Install and try it


    Install af, create a first worktree-backed session, attach, detach, and archive it.

    Getting started

  • Understand the workflow


    Follow the path from prompt to worktree, supervision, interaction, review, and cleanup.

    How it works

  • Compare alternatives


    See where Agent Factory sits relative to Herdr, GUI worktree apps, kanban dashboards, Claude Agent View, and plain tmux.

    Comparison