Now available for Windows

You're the captain. The AI crew does the work.

ModelCaptain is a desktop workspace where a small team of AI coworkers takes on your work. Say what you want in plain language. The Captain plans it, splits it up, hands it to the right specialists and gets a Reviewer to check the result. Everything lands in a folder on your own machine.

Free to try. Uses the AI subscriptions you already have. macOS and Linux coming soon.

Real footage of ModelCaptain: Codex and Claude coworkers working in parallel on one plan. Long waits sped up 2.5×.

How it works

From “I want…” to “done” in four steps

  1. Say what you want.

    Type your request in plain language, like “build me a landing page,” “make a 30-second promo video” or “research our top five competitors.” No prompt engineering, no flowcharts.

    A real request going in. The Captain’s planning wait is sped up 8×.
  2. The Captain draws up the plan.

    The Captain is a planner AI. It breaks your request into 1–5 tasks and gives each one to a coworker with the right role: Builder, Writer, Designer, Researcher and so on.

    The Captain's plan in ModelCaptain: four tasks, each showing its role, subscription and model, with the Designer on Codex and the Writer on Claude marked Working and the Builder and Reviewer Queued.
    The plan: role, subscription and model for every task.
  3. The crew gets to work.

    Independent tasks run in parallel. When one task depends on another, the Captain passes the finished work along. Each coworker is a full coding agent running on your machine with its own tools, and each runs on the best model for its task.

    ModelCaptain with the Designer task expanded to show its live steps, the Writer task marked Done, and spinners next to working coworkers in the sidebar.
    Codex and Claude working at the same time, with live steps.
  4. The Reviewer checks it. You keep it.

    A Reviewer coworker actually runs and tests what the crew built, then reports concrete issues. The finished work goes to a local workspace folder that you own.

    The wrap-up: what was tested, what still needs you, and a live preview of the result.

Features

Everything a crew needs, nothing you have to manage

The ModelCaptain Tasks screen listing tasks across chats with their status, the coworker role, subscription and model on each, and Hold and Stop controls.
Tasks: every job, who’s on it and what it’s waiting for.

Real specialists, not one generalist

Builder, Writer, Designer, Image Artist, Video Maker, 3D Modeler and more. Each task goes to a coworker with the right role for the job.

The ModelCaptain Models screen showing which model currently gets each kind of work, such as code, web, image and research, and the score formula sliders for quality, efficiency and quota cost.
Models: who gets each kind of work right now, from live scores.

The best model for each task, chosen for you

ModelCaptain picks a model for every task using live benchmarks. You pick the outcome; it handles the model choice.

A coworker's own session in ModelCaptain: the brief handed over from The Captain, then the Codex Designer explaining its logo direction while it keeps working.
Open any coworker to see its brief and talk to it directly.

Full agents with real tools

Every coworker is a complete coding agent that can use a shell, edit files, search the web, generate images and video, drive a browser or the computer, and run tools like ffmpeg and Blender.

A completed plan in ModelCaptain with all four tasks Done, including the Reviewer's mobile test on Claude Sonnet, above an answered permission request.
Every task Done, including the Reviewer’s mobile test.

A Reviewer that actually tests things

The Reviewer doesn't just read the work. It runs and tests what the crew built, then reports specific issues to fix.

The ModelCaptain Connections screen with cards for Claude, Codex, Cursor and Grok showing their subscription, sign-in state, model override and limit cooldowns.
Connections: the subscriptions you already pay for, and their limits.

Bring your own subscriptions

Sign in with the AI services you already use, such as Claude, or ChatGPT for Codex. No new model accounts to juggle.

The end of The Captain's wrap-up in ModelCaptain with an Open preview bar and a chip for each delivered file: HTML, CSS, JavaScript, logo SVGs, brand guide, copy and QA report.
Every delivered file, one click from your workspace folder.

Your files, your folder

All output goes to a local workspace folder on your machine. You own it, you can open it and you can take it anywhere.

A planner that does the planning

The Captain turns one request into a clear set of tasks, picks who does what and decides what can run at the same time.

Parallel work, smart handoffs

Independent tasks run side by side. Dependent tasks wait for their inputs, then pick up exactly where a teammate left off.

The crew

Meet the crew

Every voyage needs a crew. The Captain signs on whoever the job calls for.

  • The Captain

    Plans the work, assigns the crew and keeps everything moving in the right order.

  • Builder

    Writes and runs code for apps, tools, scripts and websites.

  • Writer

    Drafts copy, docs, reports and anything else made of words.

  • Designer

    Shapes layouts, interfaces and how things look and feel.

  • Image Artist

    Generates and edits images, illustrations and graphics.

  • Video Maker

    Creates video and handles editing, cutting and assembly.

  • 3D Modeler

    Builds 3D models and scenes, including in Blender.

  • Game Developer

    Builds playable games from your idea.

  • Sound Designer

    Produces audio, sound effects and music cues.

  • Researcher

    Searches the web, gathers sources and writes up findings.

  • Operator

    Gets tasks done by operating the computer and browser for you.

  • Reviewer

    Runs and tests what everyone else made, then reports concrete issues.

  • …and others

    The Captain can bring on other roles when a job needs them.

What you can make

What will you ask for?

Illustrative examples showing the kinds of requests ModelCaptain handles. Not real customer stories.

You say

“Build a website for my bakery with a menu and opening hours.”

Your crew delivers

A working website in your folder: Designer on the look, Builder on the code, Writer on the words, Reviewer checking it runs.

You say

“Make a small browser game where a cat dodges falling fish.”

Your crew delivers

A playable game, built by the Game Developer and play-tested by the Reviewer.

You say

“Cut these clips into a 30-second trailer with music.”

Your crew delivers

An edited video, assembled by the Video Maker with audio from the Sound Designer.

You say

“Research the top note-taking apps and compare them.”

Your crew delivers

A research report with sources, from the Researcher, written up clearly by the Writer.

You say

“Design a low-poly treasure chest I can use in my game.”

Your crew delivers

A 3D model file from the 3D Modeler.

You say

“Create a set of blog header images in a consistent style.”

Your crew delivers

A folder of matching images from the Image Artist.

You say

“Clean up this spreadsheet and summarise the trends.”

Your crew delivers

Tidied data plus a written summary of what it shows.

You say

“Fill in these forms in the browser using this list.”

Your crew delivers

The task done for you by the Operator, working the computer directly.

Supported agents

Built on the agents you already trust

ModelCaptain doesn't replace your AI tools. It gives them a crew structure. Each coworker is a full coding agent running on your own machine, using your own sign-in.

Supported agents

  • Claude Code
  • OpenAI Codex CLI
  • Cursor Agent
  • Grok Build

Each agent brings its own tools:

  • shell
  • file editing
  • web search
  • image generation
  • video generation
  • browser control
  • computer control
  • ffmpeg
  • Blender
  • and more

Models, picked for you

For each task, ModelCaptain picks the best available model using live benchmarks. Coding goes to a model that's strong at coding, writing goes to a model that's strong at writing, and so on. You don't need to keep up with the leaderboards. Your captain's log has better things in it.

Your subscriptions, your sign-ins

Use the plans you already have, for example a Claude subscription for Claude Code, or ChatGPT for Codex. ModelCaptain works with what you've signed into.

A quick tour: Tasks, Connections (four subscriptions with limit cooldowns) and Models (live benchmark routing). Real time.

Pricing

Pick your ship

Start free. Upgrade when your crew is earning its keep.

Prices in USD; sales tax or VAT is added at checkout where it applies. AI model usage is billed by your own providers under your existing subscriptions, not by ModelCaptain.

FAQ

Questions from the dock

A desktop workspace where a small team of AI coworkers does your work. You describe what you want in plain language. The Captain plans it into tasks, assigns each one to a coworker with the right role, and a Reviewer tests the result.

Everything the crew makes is saved to a local workspace folder on your own machine, and you own it. The coworkers are agents running on your computer. When they call an AI model, that request goes to the provider you signed in with (for example Anthropic or OpenAI) under that provider's own terms. For full details, see our Privacy Policy.

At least one supported agent, installed and signed in: Claude Code, OpenAI Codex CLI, Cursor Agent or Grok Build. You can use plans you already have, such as a Claude subscription, or ChatGPT for Codex. The more agents you connect, the more models ModelCaptain can choose from.

One agent is one pair of hands. ModelCaptain gives you a crew. The Captain splits the job into focused tasks, sends each to a specialist role, runs independent tasks in parallel, passes work along when tasks depend on each other, and picks the best model for each task. Then a separate Reviewer actually runs and tests the result and reports concrete issues.

You get a plan, a division of labour and a second pair of eyes, and the only thing you had to choose was the outcome.

No. ModelCaptain picks a model for each task from live benchmarks. You just say what you want done.

Code, websites, apps, games, 3D models, images, video and video edits, audio, research reports, writing, data work, and tasks done by operating the computer.

Windows is supported today. macOS and Linux are coming soon. You'll also need at least one supported agent installed and signed in. Full requirements are on the Download page. Confirm specs

Refund policy to be confirmed and reviewed by legal

Example wording to confirm: "If ModelCaptain isn't right for you, contact us within [X] days of purchase for a full refund."

ModelCaptain Pro is a subscription for one person. Your license key works on up to 3 of your own computers, and you can move it between computers yourself. Cancel anytime from your customer portal; ModelCaptain keeps working until the end of the period you paid for. Full terms are in the Terms of Service.

ModelCaptain uses your own subscriptions and sign-ins, so model usage is covered by your existing provider plans and their limits. ModelCaptain's own price is shown on the Pricing page. Placeholder

Ready to take the helm?

Download ModelCaptain, connect an agent you already use, and give your crew its first job.

Free to try. Uses the AI subscriptions you already have. macOS and Linux coming soon.