kunchenguid/firstmate

Talk to one agent. Ship with a crew.

What it solves

Firstmate addresses the "tab-juggler" problem where developers must manually manage multiple parallel coding tasks—such as bug fixes, research, and audits—across different terminal sessions and repositories. Instead of babysitting several individual agents, the user interacts with a single "first mate" agent that manages a crew of autonomous worker agents, handling the dispatching, supervision, and consolidation of results.

How it works

Firstmate acts as an "agent distro"—a collection of instructions, tools, and policies—that turns a general-purpose coding agent (like Claude Code, Grok, or Pi) into a supervisor.

  1. Liaison Model: The user talks to the first mate, which then spawns crewmate agents in isolated session backends (such as tmux, Zellij, or Orca).
  2. Isolation: Each crewmate operates in its own clean git worktree (via treehouse) to prevent parallel tasks from colliding.
  3. Supervision: A bash-based, zero-token watcher monitors the fleet and wakes the first mate only when human intervention or a decision is required.
  4. Delivery: Tasks are categorized as "ship" (resulting in PRs or local merges) or "scout" (resulting in investigation reports).
  5. Extensibility: Users can opt into "secondmates" (persistent agents on local or remote SSH hosts) and a "Relay" system to handle requests from X and Discord.

Who it’s for

Developers who use AI coding agents and want to scale their productivity by running multiple autonomous tasks in parallel without manually managing the context and state of each session.

Highlights

  • Single Point of Contact: Interact with one supervisor agent that manages the entire fleet.
  • Disposable Worktrees: Prevents code collisions by giving every task its own isolated git environment.
  • Multi-Backend Support: Compatible with various terminal multiplexers including tmux, Herdr, Zellij, cmux, and Orca.
  • Harness Agnostic: Works with several primary agent harnesses including Claude Code, Grok, Pi, and Cursor Agent CLI.
  • Restart-Proof State: All state is persisted on disk, allowing sessions to be reconciled and resumed after a crash or restart.

Related

  • Project
    johannesjo/parallel-code

    A GUI for dispatching multiple AI coding agents in parallel, using git worktrees to isolate each task into its own branch and directory for concurrent development.

  • Project
    morapelker/hive

    An open-source AI agent orchestrator that allows developers to run multiple AI coding agents in parallel across isolated git worktrees.

  • Project
    777genius/agent-teams-ai

    A desktop orchestration layer for AI agent teams that allows users to coordinate multiple autonomous agents from different providers to handle coding and project tasks via a Kanban board.

  • Project
    standardagents/dmux

    A tool that manages multiple AI coding agents in parallel by isolating each in its own tmux pane and git worktree.

  • Project
    openteams-lab/openteams

    A local-first AI desktop app that provides a shared workspace for multiple coding agents to collaborate, plan, and execute tasks without interfering with each other.