AI captain.  Real engineers.  Real results.

Give it a goal, not a prompt.

Fleet runs a crew of AI engineers on your Mac, and it doesn't stop until it's finished, and proven.

50 beta seats, free. Then $149 for the first 100, and $249 after that. One payment, never a subscription.

macOS 14.0+  ·  Apple Silicon  ·  $0 API fees

Last shipped this week19 changes this week, every one verifiedFleet builds Fleet: read the ship's log →
Fleet turning a plain-English goal into a verified draft PR: proof verified

The real bottleneck

Writing code isn't the bottleneck.
Reviewing it is.

A dozen agent tabs, every diff yours to check: your review capacity is the ceiling. Fleet flips it: you review proven work, not raw diffs.

“What is this session even doing?”

One Captain, not a dozen tabs

Brain-dump the goal to one Captain. It plans, routes, and steers the whole crew.

“How much AI code can I sanely review?”

Verified before it reaches you

A different model adversarially checks every job (compiled, tested, screenshotted) before you see it.

“It ran all night and built the wrong thing.”

On-goal, with a human gate

The Captain keeps every worker on the goal; risky calls wait for your OK.

Your review queue
PR #214booking flowproof ✓PR #209rate limitsproof ✓PR #221release notesproof ✓

Meet the fleet

One captain. A whole crew.
You just give the goal.

The Captain is a Claude: your chief of staff, surfacing only the calls that need you.

  • ▸ goal     “build the booking flow and ship it”
    ▸ plan    4 tasks · proof checks designed · 1 needs your OK
    ▸ crew    W1 schema · W2 UI · W3 payments · W4 tests
    ▸ judge   W2 returned · proof attached ✓

    The Captain runs the crew

    It takes your goal, plans the work, designs the checks that will prove it, dispatches workers, and judges what comes back. You direct like a boss, not a programmer.

  • → use the staging API, not prod
    sent to W2 · picked up mid-task

    Live steering

    Every worker is visible and steerable mid-job. Redirect it without restarting.

  • W2 · Billing usage tiersRunning
    cargo test: 214 passed · committing to its branch…

    Workers in parallel

    Several jobs at once, across your whole product, each one a full Claude Code session.

  • main · protected
    └─ fleet/w2-booking-ui
    └─ fleet/w4-payment-tests

    Isolated worktrees

    Every job builds on its own branch, in its own worktree. Main is never touched until you approve.

  • W3 wants to run a DB migrationApproveHold

    Nothing risky without your OK

    Migrations, deploys, deletions: the Captain brings the big calls to you first.

The proof gate

Done means proven.

Nothing counts as done until the worker proves it: tests run, screenshots captured. Fail, and the Captain loops it back. That proof-loop is the difference between started and shipped.

The gate is written out in full: the checks Fleet re-runs before anything is marked done.

See the whole loop
done-check · W4 · API rate limits
$ fleet verify w4
▸ pytest ············ 41 passed
▸ next build ········ exit 0
▸ screenshot ········ captured ✓
▸ curl /api/limits ·· 429 after 100 req ✓
PROOF VERIFIED · draft PR #214 ready for your review
goal“build the booking flow”PlanBuildProvefail: go againcheck passesDone. Proven.

The loop

Engineering runs in loops.

Plan, build, prove. Around again until its check passes. It's how great engineers ship, and every Fleet worker executes this way, automatically, no re-prompting. Give it a goal and a loop is running in seconds. Nothing leaves until it's proven.

See it run, move by move

Your Mac. Your plan.

Full-power Claude Code, on your existing plan.

Every worker is the real Claude Code, with a cockpit on top instead of a terminal.

Runs on your Mac

Your repos stay local. Your keychain stays yours. No cloud IDE, no code sent off to someone else's sandbox. The crew works where you work.

~/dev/jeevz.com · local
keychain ······· untouched
localhost:3000 · live preview ✓

One flat subscription

Fleet runs on your own Claude plan, so a whole crew working all day costs the same flat price. A metered agent is a taxi meter; your plan is a monthly pass.

It rests on one architectural decision, documented in why the whole fleet runs on a flat subscription.

$0
API fees, ever
no per-token meter
your flat Claude plan

How it works

From goal to shipped, in five moves.

new goal
build the booking flow and ship itSend →
plain English · outcome included · no prompt engineering

Lifetime license

One price.No subscription, ever.

Fleet is a license, not a plan. You pay once, the app is yours, and there is no monthly bill for it. Your Claude plan stays your Claude plan.

Three ways in, and all three are the same app on the same lifetime license. What changes is the price, and how many licenses exist at it.

50 beta seats100 founding licenses

Beta

50 seats

$0no payment

Yours to keep, free forever.

Not free for now. Free permanently. When the beta ends the app stays yours at no cost, with nothing to convert and nothing to renew.

It is a gift for being early. In return: run the whole app, tell us what breaks, and shape what ships. That is what the 50 seats are for.

50 42 seats left

Get the free beta

Free, no card, and free to leave. This claims your seat, and the signed build is the next click.

Founding license

100 licenses

Not open yet

$149one time

$249 standard. $100 less for the same license.

The lifetime license at the founding price, for the first 100 people who take it. One payment, and Fleet never bills you again.

It is what comes after the 50 beta seats. Once those are taken this is the way in, and it stays the same one-time payment for the same app.

100 100 licenses left

Listed now so you can see what the beta seat is worth before you take one. Nothing is for sale here yet, and nothing is being held. When it opens, it opens to everyone at once.

Standard license

Not open yet

$249one time

What a license costs once the 100 founding licenses are claimed. Still one payment, still no subscription.

Same app, same license, same $0 in API fees. The only thing the founding price buys is that same license for $100 less.

Listed so the founding price has something real to be measured against. Nothing to do here yet.

What every one of them includes

  • The whole app. Every worker is the real Claude Code, not a stripped-down API wrapper.
  • Runs on your own flat Claude plan, so a crew working all day is $0 in API fees.
  • One payment. No subscription, no per-token meter, no renewal to raise.
  • macOS 14.0+ on Apple Silicon, notarized by Apple, so it opens with no security warning.

three signed, notarized releases in five days, shipped by a founder who is not a career engineer. check the log →

What happens when the 100 are gone?

The founding price closes and the standard license price applies. It is a count of licenses, not a countdown. Nothing here expires by the hour, and the number will not quietly move once it is reached.

Is it really one payment?

Yes. One payment for the license, and Fleet never bills you again. The one recurring cost is your own Claude plan. Fleet drives it rather than reselling it, and you would be paying for it anyway.

What “lifetime” covers

The app is yours with no end date. Updates are included for two years from the day you get it, beta seats included. After that the version you have keeps working. You simply stop receiving new ones.

Questions

Straight answers.

It's early, a private beta, and we'll tell you straight.

Is it safe to run on my repos?

Every worker is visible and steerable from the board, and nothing risky happens without your OK: the Captain brings you the calls that matter (migrations, deploys, deletions) before they run. Workers build in isolated git worktrees, so main is never touched until you approve the merge.

What does it cost?

The app is a one-time lifetime license, not a subscription. There are three ways in, and the prices are listed in the pricing section of this page. On top of that, Fleet drives the official Claude Code on your own flat Claude plan, so a whole crew working all day costs the same monthly price: $0 in API fees, no per-token meter. A metered agent is a taxi meter; your plan is a monthly pass. And it isn't against Claude's terms: it's the official CLI, on your own subscription, used the way it's designed.

Which models does it use?

The real ones. Every worker is a full Claude Code session, the same tool developers run in their terminals, so you get the full Claude models and Claude Code's whole toolkit, not a stripped-down API wrapper.

Do I need to know git worktrees?

No. The Captain creates and cleans up an isolated worktree for every job automatically. You give goals in plain English; the branching, isolation, and merging are its problem.

How is this different from Claude Code alone?

Claude Code can spawn subagents, but a raw subagent is a black box: invisible, unsteerable, and “done” with nothing proving it. Spawning agents is a capability; running a fleet is an operation: visible workers, a human gate, and nothing marked done until it's proven.

Private beta

Don't just start it.Finish it.

You've got a backlog of things you want built and shipped. Fleet is the crew that finishes them: proven, not claimed.

50 beta seats, free. Then $149 for the first 100, and $249 after that. One payment, never a subscription.

macOS 14.0+  ·  Apple Silicon  ·  $0 API fees