CommandAGI
LearnPricingDocs
Log in
CommandAGI

Vision, intelligence, and control — intelligence you can see, specify, and trust, acting in real environments.

Platform

  • Product
  • Intelligence
  • Perception
  • Foresight
  • Robotics
  • Coordination
  • 3D printing
  • Services
  • Shop
  • Sites
  • Pricing

Explore

  • Worlds
  • Threads
  • Learn CommandAGI
  • Docs
  • Demos
  • Get the app
  • Start a business
  • The Command Economy
  • Ecosystem
  • The Opportunity Fund
  • Earn from your phone
  • Proof of reserves
  • Status

Company

  • Integrity
  • Trust & verifiability
  • About
  • Vision
  • Research
  • Blog
  • Brand
  • Contact

Legal

  • Privacy
  • Terms
  • SMS opt-in
  • Cookies
  • All legal
© 2026 CommandAGI INC. All rights reserved. · GDPR & CCPA aligned · SOC 2 in progress
TermsPrivacyCookiesStatus

Truth-based AI for operating systems.

000 is not a chat model with a screenshot bolted on. It reads a raw screen into things you can act on — windows, buttons, fields, each with a position and a label — acts, and then checks the screen for the change every action was supposed to cause. What a control does is learned from experience, not assumed. It reports what it saw, what it did, and what actually happened, in a form you can audit — and it gets better the longer it works on your screens.

Sees things, not pixels

Typed perception results — every window, button, and field with its position and label — from raw screenshots alone. No DOM, no accessibility tree. The same perception is available as an API for your own agents, including other models.

Verifies every action

Each click is checked against the change it was supposed to cause. An action that did nothing is retried and reported — never papered over.

Refuses to guess

When the screen doesn't support an action, it says so instead of clicking something plausible. Confident misclicks are the failure mode it is built to eliminate.

Learns on the job

What a control does is learned from its own action history. Day one it asks; day thirty it knows your apps.

Live interface

One request. A complete agent run.

Launch CommandAGI 000 behind a thread with a single API call. The thread is watchable while it works and preserves the evidence behind every step.

What it returns

One endpoint — POST /v1/transduce — takes a model id and its typed inputs, and returns that model's typed outputs. GET /v1/models lists every signature.

commandagi/000-perceive
A typed scene graph from raw pixels — every window, control, and field with its box and label.
commandagi/000-ground
A phrase resolved to the one element on screen it refers to, or an honest refusal when it is genuinely ambiguous.
commandagi/000-parse-conversation
A sentence read into the typed triples of a conversation.
commandagi/000-parse-goal
A sentence read into the steps and roles of a goal.

There is no /v1/act. The model emits structure; deciding what to do with it is the policy's job — yours through the API, or ours when you run it inside a thread.

import os, requests

res = requests.post(
    "https://api.commandagi.com/threads",
    headers={"Authorization": f"Bearer {os.environ['COMMANDAGI_API_KEY']}"},
    json={
        "model": "commandagi/CommandAGI-000",
        "environment": "ubuntu",
        "task": "Open a browser and book the cheapest flight to Tokyo next month.",
    },
)
thread = res.json()
print(thread["id"], thread["watchUrl"])

CommandAGI model 000

It doesn't suggest the work. It does it.

Our first agentic model. Truth-based AI that operates, verifies, and learns.

01

Observe

Read the screen into typed, grounded objects instead of guessing from a blur of pixels.

02

Act

Use the real mouse, keyboard, terminal, applications, and attached machines.

03

Verify

Check the change every action was meant to cause, recover, and continue until done.

Next
Put 000 to work

Give it a machine and a goal, then watch every step.

Read the docs