Browser

Browser gives agents a real browser for sites with no API: one-shot reads, interactive sessions, and logins kept between sessions. Every session is recorded step by step.

ContractCopy Anchor Link

• A one-shot read returns a page as Markdown, a screenshot or a PDF, or crawls a site.
• A session acts on plain-language steps with act, returns typed data with extract, and describes the page with observe.
• A persistent session keeps its login under a name, so the next session starts signed in.
• Secrets are typed into pages on the server. The model never sees them.
• Sessions on government, military and sign-in provider sites run undisguised, or hand off to a person.
• handoff passes a live session to a person where the law requires one, then resumes.
• Every session is a recording kept forever, with each step's Cost and model calls.
• Usage meters by session second and by one-shot read.
• An Agent uses Browser to sign up for outside services, to reach sites with no Integration, and to test its own Site.

VerbsCopy Anchor Link

None.

InterfacesCopy Anchor Link

• API
• SDK
• CLI
• MCP

ExampleCopy Anchor Link

// pantrystock.co restocks corner shops from a wholesaler whose portal has no API.
import { $ } from 'startup-builder'

const s = await $.browser.session({ persist: 'wholesaler-login' })
await s.act('open Reorders and search for "oat milk 1L"')
const item = await s.extract({ sku: 'string', price: 'money', inStock: 'boolean' })
if (item.inStock) await s.act('add 24 to the cart and place the order')
await s.close()