Skip to main content
BrowserWire is a contract layer between AI agents and websites. Instead of writing brittle DOM scrapers or maintaining fragile selectors, you browse a site once and BrowserWire builds a versioned manifest of everything it can do — entities, actions, views, and workflows — served as a REST API. 
You browse a site → BrowserWire watches → You get a typed API

​
Who it’s for

BrowserWire is built for AI agent developers and automation engineers who need to give agents reliable, typed access to web interfaces without maintaining low-level browser automation scripts. If you’ve ever written a Playwright script to log into a dashboard, fill out a form, or extract structured data — and then had it break the next time the site updated — BrowserWire is designed to solve that.

​
Key capabilities

Vision LLM discovery — BrowserWire uses a vision-capable LLM to perceive what’s on the page. It identifies interactive elements, understands their purpose, and synthesizes reliable locators without relying on fragile CSS or XPath selectors. Typed APIs — Discovered pages produce callable operations like submit_login(email, password) with input schemas. Your agent gets a structured interface, not a blob of HTML. OpenAPI output — Every discovered site gets an OpenAPI 3.0 spec at /api/sites/:slug/openapi.json. Point any tool-using LLM at this spec and it can discover and call actions without any manual integration. Chrome extension — A Chrome extension runs alongside your browser during discovery. It captures page state and communicates with the BrowserWire server over a local WebSocket connection. The extension only activates when you explicitly start an exploration session. Desktop app — An Electron-based desktop app bundles the server and extension management into a single native window, as an alternative to running the CLI manually.

​
How the discovery pipeline works

When you start an exploration session, BrowserWire’s pipeline runs the following stages:
  1. Perception — The Chrome extension captures the current page state and sends it to the local server. The server passes a screenshot and page structure to a vision LLM, which identifies what’s visible and what the user can interact with.
  2. Locator synthesis — For each discovered element, the pipeline generates reliable locators (semantic selectors, ARIA attributes, text content) so actions can be replayed consistently even if the page layout shifts.
  3. Manifest compilation — Discovered interactions are compiled into a typed manifest describing entities (data objects on the page), actions (things you can do), and workflows (multi-step sequences). The manifest is versioned and stored locally.
  4. API generation — The manifest is automatically translated into a REST API with an OpenAPI spec. You can query the spec, browse interactive Swagger docs, and execute actions over HTTP — all without writing any code.
BrowserWire runs entirely on your machine. The extension only sends page data to the local CLI server — nothing leaves your computer except the LLM API calls you configure.

​
Next steps

Quick Start

Install BrowserWire and discover your first site in under five minutes