Skip to main content
BrowserWire is for AI agent developers, automation engineers, and anyone who needs programmatic access to websites that don’t have an API. Instead of writing brittle scrapers or maintaining CSS selectors, you record a browser session and let BrowserWire’s AI figure out the structure.

​
The problem

Most websites don’t expose APIs. To automate them, you typically write DOM scrapers, maintain XPath selectors, and hope the site doesn’t change its markup. This breaks constantly.

​
How BrowserWire solves it

BrowserWire takes a different approach: record, train, execute.
  1. Record — You open a real browser via the CLI and interact with a website normally. BrowserWire captures every DOM mutation, navigation, and interaction as a structured recording using rrweb.
  2. Train — The recording is uploaded to BrowserWire’s cloud training pipeline. A vision-capable LLM analyzes the session and produces a manifest — a typed description of the site as a state machine with pages, views (data you can read), and actions (things you can do).
  3. Execute — You start a local execution server with bw run. It fetches your manifests and exposes them as REST endpoints. When you call an endpoint, BrowserWire opens a real browser behind the scenes, navigates to the right page, and executes the view or action.

​
Key concepts

ConceptDescription
RecordingA captured browser session (rrweb events) that serves as training data.
TrainingThe cloud pipeline that analyzes a recording and produces a manifest.
ManifestA typed state machine describing a site’s pages, views, and actions.
StateA distinct page or screen in the site (e.g., HomePage, LoginPage).
ViewStructured data that can be read from a state (e.g., product_list, user_profile).
ActionAn interaction that can be performed in a state (e.g., click_login, submit_form).

​
Architecture

┌─────────────┐     ┌──────────────────┐     ┌─────────────────┐
│  bw record   │────▶│  Cloud Training   │────▶│    Manifest      │
│  (Playwright) │     │  Pipeline (LLM)   │     │  (State Machine) │
└─────────────┘     └──────────────────┘     └────────┬────────┘
                                                       │
                                                       ▼
                                              ┌─────────────────┐
                                              │    bw run         │
                                              │  (Local Server)   │
                                              │  localhost:8787   │
                                              └─────────────────┘
  • The CLI runs on your machine. It handles recording and local execution.
  • The training pipeline runs in the cloud. It processes recordings and generates manifests.
  • The dashboard at app.browserwire.io lets you view recordings, monitor training, and browse generated endpoints.
  • Execution is fully local — bw run launches a Playwright browser on your machine and serves a REST API on 127.0.0.1:8787.
Your recordings are uploaded to BrowserWire’s cloud for training. The execution server (bw run) runs entirely on your machine — API calls are served locally through a real browser instance.

​
Next steps

Quick Start

Install BrowserWire and get your first API running in under five minutes.