Choose a harness
What each harness is, how Fleet finds or installs it, how to sign in, and what setup it needs. OpenCode 2, Claude Code, GitHub Copilot, Pi and OpenCode 1.
A harness is the coding tool that does the work in a session. Fleet starts it, shows you what it does and keeps its sessions in worktrees. You pick one each time you start a session.
Fleet lists harnesses in Settings → Harnesses. Each one has a card with a status:
- Ready
- Not installed
- Sign-in needed
- Update needed
- Not working
- Disabled
Set up a harness opens a dialog that installs or signs in to the ones that need it, and Check again looks once more. Set Set default on the card you want new sessions to use. If you don’t, Fleet uses the first harness that is ready, and OpenCode 2 is first in the list.
You can install more than one. The What works where page shows what each one can do.
OpenCode 2
Section titled “OpenCode 2”Fleet leads with OpenCode 2, the new OpenCode. It has the most features in Fleet.
- How Fleet finds it: it looks for
opencode2, or anopencodethat reports version 2.x, in a folder of its own, then on your PATH. It needs version 2.0.6 or newer. Fleet says Update needed for an older one. - Install: in Settings → Harnesses, open the OpenCode 2 card and choose where it goes, then Install OpenCode 2 here.
- In its own folder keeps its settings, sessions and sign-ins apart from OpenCode 1. This is the choice to make if you also use OpenCode 1.
- As your main opencode installs to
~/.opencode/bin, and replaces OpenCode 1 there.
- Windows: the installer needs bash. Download the Windows build from the card’s link and save it as
opencode2.exein the bin folder it names. - Sign in: OpenCode 2 keeps its own provider sign-ins. Use the sign-in panel on its card to add an API key or sign in to a provider. OpenCode 1’s sign-ins don’t carry over.
- Profiles: you can keep named config for it in Fleet and pick one per session. Fleet hands it to OpenCode 2 on top of your own
opencode.json.
Claude Code
Section titled “Claude Code”Anthropic’s coding tool. It needs a Claude subscription or an API key.
-
How Fleet finds it: the
claudeprogram, on your PATH or in~/.local/bin. -
Install: from Set up a harness, or in a terminal. On macOS and Linux:
Terminal window curl -fsSL https://claude.ai/install.sh | bashOn Windows, in PowerShell:
Terminal window irm https://claude.ai/install.ps1 | iex -
Sign in: run
claude auth login. Until you do, the card says Sign-in needed. -
Update: Claude Code updates itself. Fleet runs
claude updatefor you from the card.
GitHub Copilot
Section titled “GitHub Copilot”GitHub Copilot is experimental. Fleet runs it through weave-copilot, a small program that drives the Copilot CLI.
- Turn it on: it is off by default. Open Settings → Features and turn on GitHub Copilot.
- Install: Fleet downloads weave-copilot and the Copilot CLI for you, checks the download, and keeps them under
~/.weave/runtimes/copilot/. Choose Install GitHub Copilot in Set up a harness. If you already have weave-copilot orcopiloton your PATH, Fleet uses those. - Sign in: choose Sign in with GitHub on the card. Fleet shows a page and a code. Open the page and enter the code. If you are already signed in to the GitHub CLI, or have
COPILOT_GITHUB_TOKENset, Fleet shows that instead. - HydraFusion: a switch on the card, off by default. It lets Copilot pick a model, or combine work from several, for each task. It is a research preview and what a prompt costs varies a lot, so it only applies to sessions you start after turning it on.
Pi is a small coding tool you run from a terminal. Fleet runs it for you, one pi program per session.
- How Fleet finds it: the
piprogram, on your PATH. Fleet doesn’t install Pi, and the card has no setup button. Install it from pi.dev first, then choose Check again. - Sign in: sign in to your model provider in Pi itself. Fleet doesn’t manage Pi’s sign-ins.
- A first message: a Pi session starts with your first message, so write it before you start.
OpenCode 1
Section titled “OpenCode 1”The original OpenCode. It includes free models, or you can sign in to your own provider in OpenCode.
-
How Fleet finds it: the
opencodeprogram, on your PATH or in~/.opencode/bin. It needs version 1.15.10 or newer. If that program is OpenCode 2, the card says so and tells you how to keep both. -
Install: from Set up a harness, or in a terminal:
Terminal window curl -fsSL https://opencode.ai/install | bashOn Windows:
winget install --id SST.opencode --exact. -
Sign in: none is needed for its free models. For your own provider, sign in in OpenCode.
-
Pooled OpenCode Mode: a switch on the card. New sessions share one OpenCode process that Fleet starts when they need it, so they take a prompt after Fleet restarts without you resuming them.
Using Weave with a harness
Section titled “Using Weave with a harness”Weave adds a team of agents to a harness. Install it the way its own docs say for your harness:
Fleet starts the harness as usual. Use Weave in Fleet explains what to expect.
Where next
Section titled “Where next”- What works where: what each harness can and can’t do in Fleet.
- Your first session: start a session on the harness you set up.
- Troubleshooting: a harness that isn’t found or needs you to sign in.