AIAny
AI Agent2026
Icon for item

firstmate

Routes natural-language requests to a single “first mate” agent that spawns and supervises multiple autonomous crewmates, each running in an isolated git worktree and producing finished PRs, approved local merges, or standalone investigation reports. Key features include visible session backends, disposable worktrees, explicit project modes, optional persistent secondmates, and an event-driven zero-token watcher.

Introduction

Why this matters Firstmate treats multi-task agent work not as a single monolithic assistant but as a supervised crew: you speak to one liaison (the “first mate”), which then spawns, isolates, and oversees multiple autonomous crewmates to finish concrete engineering work. This shifts the hard parts of parallel agent workflows—context leakage, colliding repo edits, and orphaned background runs—into engineered boundaries you can inspect and control, rather than invisible model state.

What Sets It Apart
  • Visible, session-backed orchestration: every crewmate runs in its own tmux/herdr/zellij/cmux/Orca terminal you can watch or type into, so automation is observable and recoverable rather than opaque. That makes debugging and trust decisions much simpler than headless multi-agent runners.
  • Disposable worktrees and guarded project boundaries: each task uses a clean git worktree (or Orca-managed worktree) so parallel edits don’t collide, and the first mate enforces read-only defaults with narrow, captain-approved operations to reduce accidental repo changes.
  • Two task shapes and explicit project modes: tasks are either ship (make authorized changes) or scout (investigate and report). Projects declare shipping modes (no-mistakes, direct-PR, local-only) that define how outputs are delivered, clarifying risk and automation scope up front.
  • Event-driven, zero-token supervision and restart-proof state: a lightweight bash watcher wakes the first mate only on actionable events; state lives on disk and in the session backend so interrupted runs reconcile on restart. Optional secondmates let you run persistent isolated homes locally or on SSH-reachable hosts with guarded updates.
Who It's For and Tradeoffs

Great fit if you want to offload routine engineering work (fixes, tests, small features, audits, research tasks) to LLM-backed agents while keeping strong operational control and auditability. Teams that need parallel, observable automation across multiple repos will benefit from the disposable-worktree model and supervised session design. Look elsewhere if you need a pure hosted SaaS orchestration layer (firstmate is a repo-based distro you run where you control tooling), or if you require vendor-hosted model inference—firstmate is a distro and runtime integrator, not a model provider. It also expects you to install and configure a supported harness (Claude Code, Grok, Pi, Codex, OpenCode) and local tools like git and gh; that dependency surface is intentional tradeoff for local visibility and safety.

Where It Fits

Think of firstmate as a local/hosted agent orchestrator and safety layer: it complements LLM/harness providers by providing isolation, supervised lifecycle, and repository-safe delivery modes. Use it when you want LLM-driven automation that produces tangible repo outputs and requires human-in-the-loop merge authority, rather than black-box autonomous pushes.

Quick operational note

The repo is the distro: install means cloning and launching a supported harness. The core value is procedural and architectural—how it isolates work, supervises agents, and guarantees recoverability—so evaluate it by whether visible sessions, disposable worktrees, and explicit shipping modes match your operational risk model.

Information

  • Websitegithub.com
  • Authorskunchenguid
  • Published date2026/06/12

Categories

More Items

Hugging Face
AI Model2026

A 9B agentic multimodal SFT checkpoint distilled from Qwen3.5-9B for coding, general agent tasks, visual coding and cybersecurity. Provided by Xiaomi MiMo as a research seed (77.4B-token SFT mix) to bootstrap agentic RL and tool-use experiments.

Hugging Face
AI Model2026

Preview agentic language model for research and engineering workflows that turns research questions into executable, verifiable workflows via tool use and long-context reasoning; built on a 744B-parameter MoE (GLM-5.2) with MIT-licensed BF16 and FP8 checkpoints.

Hugging Face
AI Model2026

Post‑trained 9B causal language model optimized for agentic workflows, tool use, coding, and long‑context instruction following. Uses a routing‑guided agentic post‑training pipeline that converts harness executions into training signal, improving agentic and coding benchmarks. Text‑only weights (safetensors/BF16), Apache‑2.0.