fleetd

Software factories, reinvented as fast as the models improve.

You set the direction. Agents build, review and ship it while you sleep.

curl -fsSL https://fleetd.sh | sh

macOS or Linux x64. Open source, MIT. Install · Quick start

Software isn’t an asset. It’s a dozen experiments. Keep one. Drop the rest.

Agents made building cheap, so stop polishing one bet for months. Ship a dozen experiments tonight, see what people actually use, and drop the rest without regret. fleetd is built for that pace: a lead plans the work, agents build it in parallel, and every change is reviewed before it lands.

Ask the lead; it lands the PRs

The lead splits your request between agents, each PR gets its own reviewer, and the merge train lands them together.

Merge train
storefront’s chat in the fleetd app. You asked to let shoppers save their cart and pick it up on the web or the iPhone app; the lead answers it is on it in three parts, each with its own writer and a reviewer for each PR, then that #231 is up with a reviewer on it. The panel lists the agents at work, the three new writers and #231’s reviewer among them.

Your team in one chat

Everyone on the hub writes in the same project chat, under their own name. The lead answers all of you.

People
checkout’s chat in the fleetd app. Bob Okafor asked for the store’s VAT number on receipts, and the lead told him #84 did it. You, Sam, ask whether the app’s checkout can take Apple Pay; Alice Chen, under her name, adds to keep card entry as the fallback; the lead answers both: “Sam, Alice: one order for both.”

Agents message each other, in the open

When one agent learns something another needs, it tells it directly. Every message shows on the agent’s work and in the project’s log, with who read it and when.

Agents’ messages
One agent’s work in the fleetd app: it asks another agent whether #214 keeps an old export, waits, is woken by the answer, edits and tests, then tells three other agents what changed: two read it at their next step, and one that has finished will read it if it runs again.
The project’s log of agents’ messages, newest first: who sent each to whom, when each agent read it, and one that didn’t wake its agent because it had already finished.

Watch an agent’s screen, and take over

Agents test in a real browser or iPhone simulator while you watch. When one needs a code or a login, take over, then hand back.

Screens
storefront in the fleetd app. The lead says the iPhone app’s sign-in texts a code to a test number, so its agent can’t go on alone, and asks you to take over its screen. The panel shows that screen live: an iPhone simulator on the app’s sign-in code page with the code 418093 typed in, its bar saying you’re in control, and Hand back under it.

Each clip and still is the fleetd app on sample data, with the waits cut short.