Public beta for macOS

One chat for all your agents.

Type what you want. Fleetwork picks the project, model and team, checks the work, and asks you only when it matters.

Download for Mac See what it does

Works with the Claude, Codex and Cursor plans you already have.

A harness that manages itself.

Fleetwork handles routing, model choice, failover, memory, checks and approvals for every message.

fix the checkout rounding
ProjectShop TeamBuilder EffortBalanced

Every message finds its project

Fleetwork picks the project, team and effort. If it can't tell, it asks.

QuickHaiku
BalancedSonnet
HighOpus
MaxFable

The right model for the job

Small asks go to light models and hard ones to strong models. You choose which is which.

Claude Code
Codex

Accounts cover for each other

When one plan runs out, the next picks up where it left off.

Tests
Second look
QuickBalanced

Checked before you see it

Every change is tested. If something's missing, it gets fixed or redone one level up.

Memory
ClaudeCodexCursor

One memory, every model

Your instructions and notes go with every task, whichever model runs it.

  • ShopBuilding
  • MobilePlanning
  • APIDebugging
  • BlogWriting

Every project at once

Tasks run side by side, each in its own project and thread.

git push --force origin main

Replaces the remote main branch with your local copy.

DenyAllow

Risky commands, in plain English

Anything dangerous waits for your OK, with one line on what it does.

Summary of recent work

Threads stay small

Long threads fold into a short summary to keep context small.

Wrong project?Which project? Mobile
Mobile
Asks less next time

Learns from your corrections

Move a task or change its effort, and routing gets better.

Working0:42
Stop

Stop means stop

One click ends the task and everything it started.

localhost:5173Open
Open in CursorOpen in VS CodeOpen in Zed

Dev servers and your editor

Open or stop dev servers from the task, and open any project in your editor.

1update the pricing page copy
2then deploy to staging
Ask Fleetwork to do something

Keep typing, it keeps up

New messages wait in a queue while work runs. Edit them before they start.

Big builds run like a team.

Small asks get one agent. Big ones get a plan, milestones and a review.

1

Plan

Splits the work into milestones, each with a clear finish line.

Sync schemaOffline queueConflict UITests
2

Research

Answers open questions before the build starts.

Read docsCheck the repo
3

Build

Ships one milestone at a time. Each one has to pass its tests.

1234
4

Review

Checks every milestone and fixes what it finds.

All milestones met

See it in action.

Every task ends with your move.

See what it did, what changed, and what's left for you to decide.

  • Plain summary
  • Your decisions
  • One-click undo
A finished task with what it did, your move and the changed files

Watch big builds move.

The plan, milestones and review update live. A milestone that fails retries at higher effort.

  • Tests per milestone
  • Automatic retries
  • Switches accounts when one runs out
A big build shown as a plan, milestones and a reviewer

Everything in flight, at a glance.

See what's working, what needs you and what's done.

  • Filter by status
  • Live time and tokens
  • Retry in place
The task list with working, needs you, queued and done tasks

Nothing risky runs without you.

Force pushes, deploys and deletes wait for your OK.

  • Plain-English summary
  • Allow once or deny
  • Stop ends everything
A force push waiting for approval with a one-line summary

Fun facts about how you work.

See when you work, what you ask for, and how much ran on lighter models.

  • Activity heatmap
  • Go-to first word
  • Words in and out
Stats with a heatmap of when you work and fun facts

Built for people who run agents all day.

Fewer tokens

Light models handle small work, so strong ones are saved for hard work.

Separate context

Each project has its own thread. Nothing leaks between them.

Sub-second routing

With a TypeSafe Jev key, routing takes under a second. Without one, a quick model routes.

Local first

Threads, memory and settings stay on your Mac.

Your move

Every finished task ends with what you need to decide next.

Stays awake

Keeps your Mac awake during long work, even with the lid closed.

Questions

Do I need a new subscription?

No. Use the plans you already have: Claude through Claude Code, ChatGPT through Codex, Cursor, or API keys from Anthropic, OpenAI and Gemini.

What does it run on?

Macs with Apple silicon. Agents run through the Claude Code, Codex or Cursor command-line tools.

Where does my data go?

Threads, memory and settings stay on your Mac. Requests go only to the model accounts you connect.

Can agents do something I didn't want?

Risky commands like force pushes, deploys and deletes wait for your OK. Stop ends a task and everything it started.

Is it finished?

Not yet. It's in beta, so expect rough edges and quick fixes.

Give your agents a dispatcher.

Download the beta

For Macs with Apple silicon