A fleet of coding agents, each in its own boat.

Run Claude Code, Codex and OpenCode side by side. Every thread gets a whole VM on boat.dev, with its own files, processes and ports, so agents can't trip over each other or over your laptop.

Get it on GitHub

Free and open source, MIT licensed. For the Mac and Linux, with an iPhone app coming.

Every agent gets its own machine

A thread is a whole VM. Its agent can install what it needs, start a dev server and run the test suite 30 times in a row without touching another thread or your Mac.

A quiet VM stops on its own after 5, 10 or 30 minutes, and wakes when you send. For a long job, turn on Keep running.

Flotilla's sidebar on the Mac. A hover card on the Slugify thread shows its agent, Claude Code with Opus 5.5 on high, and that its VM has 23 minutes left.

From a prompt to a merged pull request

Add a GitHub repo and Flotilla sets it up once, then keeps it current as main moves. Each new thread starts in seconds, in a fresh VM with your repo already set up.

The pull request sits on the thread with its checks. If a check fails, a merge conflicts or a reviewer asks for changes, one button writes the prompt: Fix, Resolve or Address review. When it's green, hold to merge.

A thread on the Mac where Codex ran the test suite 30 times and found no flaky tests. Its pull request, #19, sits above the message box with 2 of 2 checks passing and ready to merge.

Agents hand work to other agents

An agent can pass part of its task to another agent, in the same VM or in a new one. Ask Opus 5.5 to get second opinions from Codex on GPT-6.1 Sol and from Claude Fable 5.1. Both reviews come back to the thread that asked, and Opus reads them, fixes what they found and tells you who caught what.

In this thread, both caught Straße coming out as strae, and each found something the other missed.

The same thread on the Mac after both reviews came back. Opus 5.5 sums up who caught what: both caught letters like ß being dropped and a negative max_length; only Codex caught Unicode dashes merging words; only Fable caught non-Latin titles coming out empty, apostrophes and slugify(None). The 11 tests pass.

Close the laptop. The threads keep going.

Flotilla is a server as well as an app. The server can live inside the Mac app, or on a Linux box that's always on and that your devices reach over Tailscale. Then your Mac, your iPhone and a Linux desktop all show the same threads, live.

Pair a device with a QR code or a link.
Each device gets its own token, and you can revoke one at any time.
Your boat.dev key and GitHub sign-in stay on the server.

On the Linux box

$ flotilla setup --tailscale-serve
$ flotilla service install

setup asks for your boat.dev key, signs in to GitHub and prints a pairing link and QR code. service install keeps it running.

Your threads, on your iPhone

A native iPhone app is coming, with the same threads streaming live, hand-off cards and all. Start a thread with a photo attached, hold to merge with your thumb, and when an agent finishes, a notification takes you straight to its thread.

Every threadGrouped by what needs you
Hand-offsCards for every agent that answered
Hold to mergeIt fills under your thumb

It isn't on the App Store or TestFlight yet.

Free and open source

Flotilla is MIT licensed, and the app costs nothing. Here's who bills for what:

Flotilla: Nothing
boat.dev: VM time, on a paid plan
Your model providers: Tokens, as they do now

Idle VMs stop on their own, so a quiet thread costs nothing.

Build it from source

There's no download yet. You'll need:

A Mac with Xcode, or a Linux desktop running Debian or Ubuntu, and Rust
A boat.dev account on a paid plan
For projects, the GitHub CLI, signed in
To keep threads going with the laptop shut, a Linux box that stays on

On your Mac

git clone https://github.com/joseph-lozano/flotilla.git
cd flotilla
script/bundle-macos
open target/release/Flotilla.app

The README covers Linux, your boat.dev key and running a server.

Thanks to Theo and T3 Code

Flotilla is inspired by T3 Code, a GUI for coding agents from Theo. Much of Flotilla was built by pointing my coding agents at the T3 Code repo. The new-thread page, the small touches in a thread, and the way the server pairs devices and talks to them all started there. Thank you.

Joseph Lozano