Introducing Describe
Describe what should happen on a website, in plain words. An agent does it in a real browser you can watch and take over, once or on a schedule.
Most of running a business is doing the same things on the same websites. Reading the inbox and sorting it. Checking a few competitors’ pages. Looking up a new lead and leaving a note on their record. Pulling the week’s invoices and matching them to a sheet. None of it is hard. All of it is time.
Describe is a way to hand that work over by writing it down.
Write what should happen
You type a description in plain words, the way you would brief a new colleague:
Every weekday at 8am, read my unread Gmail from the last 24 hours, group it by what needs a reply and what is waiting on someone else, then post the summary to #ops in Slack.
As you type, Describe reads it and draws it back as a diagram: when it runs, then one node per step, each quoting the words it was read from. We call that drawing a blueprint. If the drawing is wrong, the words are wrong, and you fix the words. Nothing runs until the drawing is right.
A description with a time or an event in it gets two actions: run it once, or turn it on and keep it running on the schedule the words describe.
A real browser does it
The agent works in a real Chromium, with its own profile, on its own display. You can watch it live in the app. When a site asks for a login or a file, you take the controls and hand them back; the file comes from your own machine, never from ours. Logins stay in that browser’s profile, private to you.
This is deliberately not a screenshot of a browser, and not a script that pretends to be one. It is the same kind of browser you use, doing the thing you described, where you can see it.
Any model, per automation
The cheap model sorts your inbox; the deep one reconciles the books. Every model on the picker is on both plans, with its cost marked one to four before you choose. A run is charged by what the model actually used, never a flat fee.
For the people who build on it too
What the app draws is what the API returns: a trigger, the steps, the connections, the model, as JSON. Read a description, check the plan, confirm it as an automation, and webhooks tell you when a run ends and what it cost. The business user and the developer are looking at the same drawing.
Where we are
Accounts, workspaces with members and roles, private browser sessions, automations, the model-backed reader and the credit pool have all shipped; the changelog records each one the day it lands. We are letting people in from the waitlist in order. If the description above sounds like your Tuesday, join it below.