Ask.
Click.
Done.

Fast browser automation in the Chrome you're signed into.

Ask in plain words. It takes it from there.

Your Chrome.
Your logins.
Already there.

Computron works in the browser you already use, through an extension. No second profile to sign in to.

Works like you. Faster than you.

It moves the cursor, clicks the links, and types into the fields one key at a time. No typed-in URLs. No hidden shortcuts.

Done once.
Faster next time.

After two matching successes, Computron replays the same clicks with no model calls.

One prompt.

Every site.

Several sites. One message.

Name every site in one message. Computron opens them all at once and reports back when each one is done.

One prompt.
Every site.

Several sites. One message.

Name every site in one message. Computron opens them all at once and reports back when each one is done.

Any site.
No APIs. No scripts.

If a person can do it in a browser tab, Computron can try it there. New service, unfamiliar console? Say what you need, and it finds the way by clicking.

Clicked “Sign in”
Typed “Tokyo”
Copied link address
Task complete
Replayed a trusted workflow

0

model calls when a trusted workflow replays. Computron repeats what already worked, and reassesses when a page changes.

You ask. It does the clicking.

Watch it work.

Scripted interaction demo. Edited playback.

Just say the word.

Describe a browser task in plain English. Computron opens the site, finds its way by clicking, and reports what it found.

Make success stick.

A task that worked twice is replayed from its recorded clicks, with no model calls. Records stay on your machine as plain text.

Recorded steps of a browser task, ready to replay

Change happens. Computron reassesses.

If a page no longer matches what was recorded, the replay stops and a fresh run takes over. It never guesses on a submit or a delete.

A replay falling back to a fresh run when the page changed

Any web app, as a function on your PC.

Computron learns a task once. Then you call it by name, with arguments, from a script, a scheduler or an agent, as many times as you like.

  • Runs on your PCThe browser work happens in your own Chrome, signed in as you.
  • Replay speedA task is learned once. Repeats follow its recorded steps, usually with no model calls.
  • People stay in the loopA phone check or a CAPTCHA is never automated, so it never becomes a function.
API docs

computron

$ computron functions
wikipedia.search   query
x.post             text

$ computron call wikipedia.search --query Osaka --json
{"status":"succeeded","modelCalls":0,"ms":2900, ...}

Questions

Does Computron need Chrome's --remote-debugging-port?

No. It connects through a Chrome extension and native messaging, so it works in the Chrome you are already signed into. Since Chrome 136 the debug port is ignored on the default profile anyway; Computron keeps it only as a last-resort fallback and says so in one line when it has to use it, because that fallback is a logged-out browser.

How the extension bridge works →
Can it use my existing logins, passkeys and 2FA?

Yes. A task runs in a tab of your own profile, so your cookies, passkeys, hardware keys and password manager are all there. When a site asks for a CAPTCHA or a passkey touch, Computron stops, tells you, and carries on once you have done it.

Why headless agents hit a login wall →
Does it run headless?

No, by design. A Chrome profile can be open in one process only, and a headless copy of it has none of your sessions. Computron works in a background tab of your Chrome instead, in its own tab group, and only brings the window forward if you asked to watch.

What does “no model calls” mean?

After two matching successes, a task is replayed from its recorded clicks and keystrokes without asking a model. An irreversible step (submit, delete, pay) replays only when the target matches exactly; on any doubt the replay stops and a fresh, model-driven run takes over. Replays are not always faster, so Computron does not claim a multiplier; it does log the time, model calls and steps of every run.

Where a browser agent's time goes, measured →
How is it different from browser-use, Stagehand or Playwright MCP?

Those start a fresh browser, usually headless, and ask a language model what to do at each step. Computron runs in the Chrome you use, moves the cursor and clicks the way you would, decides each step with one round trip to a decision model (Jev), escalates to a large model only when it is stuck, and replays what has worked before.

What do I need to try it?

Chrome on Windows, macOS or Linux. Install Computron with the one command at the bottom of this page, sign in with Google or GitHub, and you start with 500 free credits.

Getting started →
How much does it cost?

Computron runs on credits, and the AI model costs are included. You start with 500 free credits. After that, there are monthly plans and credit packs; type /credits in Computron to open the pricing page.

Pricing →

Stop clicking through consoles. Get back to building.

curl -fsSL https://thecomputron.com/install.sh | sh

Paste it into a terminal. 500 free credits when you sign in.