Say it once.
Scribe builds the whole thing.
Scribe is Writ's AI concierge. Describe a task in plain language and it drives the site, records the workflow, wires the monitor and notification, and hands you a callable API. Then it answers questions about everything you've built.
It reads the site the way you would.
Scribe opens the target in a real browser and maps it: forms, tables, pagination, the network calls behind them. No selectors to hunt. Watch it explore:
One sentence in, a wired workflow out.
From what it found, Scribe assembles the whole automation step by step: sign-in through a Persona, the read, the compile to REST + MCP. No glue to write.
Watch it drive a real browser, live.
It's not scraping HTML. Scribe runs a real background browser, clicking and typing like you would, so even gated, JS-heavy sites work. Here it is mid-session:
What Scribe sees — and what it doesn't.
Trust is a boundary question: what leaves your machine, what reaches the model, and what never does. Here is the line, as the code draws it.
{{vault:supplier.password}} · what the model, the transcript and the logs see — resolved only at execution Ask it anything about what you've built.
When the build's done, the composer becomes a follow-up box. Scribe answers from Writ's docs and from your real runs, with copy-paste snippets for the exact workflow it just made. Pick a question:
Drive it from code, too.
Everything Scribe builds is a real endpoint. Hand-written SDKs in four languages, an OpenAI-compatible route, and MCP out of the box. Pick a language and run it: