For projects built with AI coding agents
Every agent change gets its own card and a review from a different model before it reaches you. Small changes instead of one giant prompt, and nothing goes out until you give it the okay.
Get more done with the AI you already pay for. Fleet sits between your favorite model and all the others.
Free for 7 days, then $19/month. Your trial starts when you create your business and needs no card. To keep going after it, add a card on the billing page; nothing is charged during the 7 days. Runs on AI accounts you connect, so the AI spend stays yours.
.signup-form { display: flex; display: grid; gap: 12px;}.signup-form button { float: right; min-height: 44px;
Product demo
An example workflow: a worker claims a card, proposes a lesson the librarian adds to the brain, and hands the work to a reviewer on a different model. Once you approve, the work ships. It is scripted for this page, so nothing here is live.
Step 20 of 20
Why it is efficient
Job prompts are still text files, but short and aimed at one outcome. The board, the small tasks and the review do the rest.
Code routes the work
Deterministic code moves the work, not an endless pile of dated notes that chats get confused by. The board is an accurate freeze frame of your business, so any model can walk in and see what is current without reading a pile of files.
Small tasks
Split one job into 3, 5 or 10 small tasks. A fresh model with only the context it needs solves its piece and ends. The board keeps the work moving and catches mistakes along the way.
Plans you already pay for
Run on the Claude plan you already pay for, with no API credits to buy, and add OpenAI (Codex) or Grok with an API key. When one model hits its limit, move the job to another. Choose the model for each job, and coordinate their work, progress and results on one board.
A second opinion
The savings pay for a review layer. A reviewer on a different model reads the work before it reaches you and sends it back for a fix, instead of putting the problem on your list. Every card shows what the agent did, step by step.
Why we built it
A model has no shape until you give it instructions, and like a camera lens it can only bring so much into focus at once. Ask one model to hold a whole project and the edges blur: the context drifts and it starts filling gaps with guesses.
Fleet works the way this actually works in practice. It breaks the job into a chain of small steps and gives each step its own model, a job-specific prompt and only the context that step needs. Each model looks at less, so it checks more.
A different model reviews every step, and nothing ships until you approve.
Fleet's own agents build Fleet this way. A person signs every merge.
How the brain works
When you pay for AI by usage, the provider earns more the more tokens you burn. Fleet earns a flat USD 19 a month and nothing on your AI usage, so our incentive runs the other way: get the job done with less. That is why the brain is minimalist. Short lesson records, small briefs and on-demand lookups keep the model's context clear, so you spend fewer tokens on reading and more on the work.
Most AI setups start every session by loading a wiki: old handoff notes, long rules files, last week's chats. That uses up the model's focus before it does any work, and half of it is out of date.
Fleet keeps what your team learns as lesson records. Each one is four short lines: when it applies, what to do, why, and an id. Never more than 60 words.
An agent starts with a short brief for one task. When its work touches something a lesson names, it looks that lesson up. Nothing else comes along.
Not every lesson makes the cut. Fleet ranks lessons by how closely they match the task and hands the model only the top few. The librarian keeps the set lean: it merges duplicates, retires lessons that are wrong or out of date, and writes down why, so the model's attention goes to the knowledge that matters most. You can see what each lesson is doing for you, and undo any change.
An illustration, not a benchmark.
Each agent goes by a name, works one role and runs on your own AI account, so you can see who is working on what. The names here are examples. Add any department you need.
How it works
Why Fleet is different
Here is what changes when your work runs on a Fleet board instead of in a chat window.
Not right? Send it back with a note, and it goes back to work.
Runs on AI accounts you connect, so the AI spend stays yours.
Your approval
Pricing
The subscription pays for Fleet and nothing else. Your AI usage is billed by the vendor you connect, under their own quotas and limits.
Fleet
$19per month
7-day free trial
Free for 7 days, then $19/month. Your trial starts when you create your business and needs no card. To keep going after it, add a card on the billing page; nothing is charged during the 7 days. When the trial ends, $19 per month starts on that card. Cancel before it ends and you pay nothing. Fleet does not include a server: you run Fleet's runner on your own suitable computer or host and keep it on and online. How much runs in parallel depends on that machine and on your AI providers' limits.
AI provider usage, billed directly by the provider you connect, with its quotas and rate limits. The computer or server you run Fleet's runner on.
Yes. Your agents run on the AI accounts you connect, and each provider bills its own usage. Sign in to Claude Code with a Claude plan that includes it, or add an Anthropic API key. OpenAI (Codex) and Grok connect with an API key from that provider. What a plan allows, and how much, is up to the provider.
No. Every output meant to leave your workspace waits for you to read and approve it. Fleet files approved work as a document and does not send posts or emails itself.
Free for 7 days, then $19/month. The trial starts when you create your business, runs for 7 days and needs no card. To keep going after it, add a card on the billing page; nothing is charged during the 7 days. When the trial ends, $19 per month starts on that card. Cancel before it ends and you pay nothing.
That depends on your setup. Fleet's runner works on your own computer or server, and it can take on several tasks at once as far as that machine and your AI providers' limits allow. Line up as many jobs as you like; the rest wait for a free slot.
Yes, from the billing page, any time. Cancel before the trial ends and you pay nothing.
In your own isolated Fleet workspace. Your board, jobs and approved work stay inside it.