Skip to content
PostHogPublic

About

Fleet engine: one command sets up T3 Code, agents and opt-in workflow packs

Resources

Code of conduct

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

Fleet

Your Mac runs T3 Code. It links to your devboxes, where agent threads work in their own worktrees. An optional hub shares MCP servers with them. Every machine reads the same private profile repo on GitHub.

Coding agents get much better with good rules, skills, and a machine that stays on. Setting that up by hand takes a day per machine, and the machines drift apart.

Fleet sets up T3 Code, Claude Code, Codex, and a set of agent workflows on your Mac and your Linux devboxes. You pick the workflows as packs. Your answers live in your own private profile repo, so every machine gets the same setup.

Getting started

  1. Download T3 Code and open it. Its Welcome page sets up one agent, Claude Code or Codex, and signs you in.

  2. Sign in with T3 Connect. In T3 Code, open Settings → Sharing & access, sign in, and turn on T3 Connect. It links your Mac to your devboxes and your phone.

  3. Paste this prompt into a new thread in T3 Code on your Mac, or open it in T3 Code:

    Set up Fleet on this Mac with me. Read https://raw.githubusercontent.com/PostHog/fleet-engine/main/SETUP.md and follow it step by step.
    

From here the agent leads. It asks what you know and what you want, explains each part, and runs the steps itself. You answer its questions and do your own sign-ins.

See what it does

Open the Fleet site to choose where Fleet runs: on your Mac only, or on your Mac plus a hub and a node. Switch the Pull requests, Orchestration, and Communication packs on or off. The site builds the prompt above with your picks in it, so the agent skips those questions.

The Packs page shows every pack, and each pack has its own page. Select packs there, and copy one prompt that adds them to a Fleet you already run.

bun scripts/build-page.ts generates the site from the pack data into site/. To see it, open site/index.html.

Status

Version 0.1.0 is the first public release. It is early. On every pull request, CI runs the install as a new Linux user, then fleet setup and fleet doctor, with a sandbox GitHub account. Other paths are built but not yet proved end to end, such as a run on a clean Mac user and the real sign-ins. The changelog lists them under "Known gaps".

Set it up by hand

Run this as your login user, not with sudo, on macOS or on Linux (x64 or arm64):

curl -fsSL https://raw.githubusercontent.com/PostHog/fleet-engine/main/get.sh | sh

Read get.sh first if you want to know what runs. It:

  • installs what Fleet needs if it is missing: the GitHub CLI, Bun, and T3 Code. On a Mac it first installs the Command Line Tools if needed, and Homebrew only when the T3 Code app is missing. On Linux, install curl, git, tar, sha256sum, python3, and cron first. Linux also needs systemd's user manager, so T3 Code can run as a service. To keep it running after you log out, an administrator runs sudo loginctl enable-linger <your user> once.
  • fetches the engine into ~/.local/share/fleet/engine and puts fleet in ~/.local/bin.
  • adds one line to your shell's login file, so ~/.local/bin and ~/.bun/bin are on your PATH.
  • starts fleet setup.

fleet setup asks plain-language questions. It asks you to sign in to GitHub, and to Claude, ChatGPT, or both, depending on the plan you pick. It then creates the private repo fleet-profile in your GitHub account, pushes your answers there, and applies your packs to this machine. Running get.sh again picks up where it stopped.

Getting started walks through it step by step, from a new Mac to a linked Coder devbox, with what each sign-in asks.

Develop

bun install
bun run test
bun run typecheck

See GLOSSARY.md for the words Fleet uses.

License

MIT. See LICENSE.

About

Fleet engine: one command sets up T3 Code, agents and opt-in workflow packs

Resources

Code of conduct

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages