Skip to main content
This guide walks through the full recording workflow — from launching the browser to viewing your generated endpoints on the dashboard.

​
Prerequisites

  • BrowserWire CLI installed
  • Logged in (bw login) — if your token has expired, bw record will prompt you to re-authenticate automatically

​
Step-by-step

1

Start a recording

Run the record command with the URL you want to automate:
bw record https://example.com
A Chrome browser opens and navigates to the URL. You’re not recording yet — take a moment to orient yourself on the site.
2

Press Enter to start recording

When you’re ready, press Enter in your terminal. BrowserWire reloads the page for a clean snapshot and begins capturing all DOM events.
Recording started. Navigate the site, then press Enter to stop.
3

Interact with the site

Navigate around the site and interact with the features you want to automate:
  • Fill out forms — login, search, checkout
  • Click buttons and links — navigation, actions, filters
  • Open dynamic UI — dropdowns, modals, accordions, tabs
  • Visit multiple pages — the more pages you cover, the more states the AI discovers
Take your time. Let pages fully load before interacting.
4

Press Enter to stop recording

When you’ve covered the flows you need, press Enter again. BrowserWire stops capturing events and reports how many were collected:
Captured 847 events.
Uploading recording...
Upload complete.
Starting training...
Opening dashboard: https://app.browserwire.io/dashboard/recordings/abc123
5

Review on the dashboard

The dashboard opens to your recording detail page. Here you can:
  • Watch training progress — the training status badge shows running (with an animated dot), and updates to complete or error when finished
  • Replay your session — an interactive player at the bottom of the page lets you scrub through the recording and verify it captured what you intended
  • Jump to the manifest — once training completes, a link appears next to the status (e.g., → example-com) that takes you directly to the generated API endpoints
6

Browse your endpoints

Go to APIs in the dashboard sidebar (or click the manifest link from the recording page). Click your domain to open the Endpoint Explorer, which lists every discovered view and action with copy-ready curl commands.
7

Run locally

Start the execution server and test your endpoints:
bw run
curl http://127.0.0.1:8787/api/sites
Or paste a curl command directly from the Endpoint Explorer.

​
Tips for better results

Cover critical paths first. Focus on the most important user flows — the ones you’ll actually automate. A focused 2-minute recording of a login flow produces better results than 10 minutes of random browsing. Navigate slowly. Let pages fully load before clicking or typing. The AI needs stable DOM snapshots to accurately identify elements. Interact with dynamic UI. Open dropdowns, expand accordions, toggle state. These interactions help the AI discover actions that aren’t visible in the initial page load. Use a real account. If the site requires authentication, log in during the recording. The AI can only discover pages and actions it can see. Re-record to improve. You can record the same site multiple times. Each recording triggers a new training run that replaces the previous manifest. Add recordings that cover flows you missed the first time.

​
Handling new tabs

If you click a link that would normally open in a new tab, BrowserWire redirects the navigation into the current page so the recording stays continuous. You don’t need to do anything special.

​
Browser disconnection

If you close the browser accidentally while recording, BrowserWire detects the disconnection and saves whatever events were captured up to that point. The upload and training run proceed normally.

​
Training errors

If training fails, the recording detail page shows an error message: “Training was unable to complete. Please try recording again.” Common causes:
  • Very few events — the recording was too short for the AI to identify any meaningful structure. Try recording more pages and interactions.
  • Heavy client-side rendering — some sites use complex rendering that the AI can’t fully parse. Try interacting more slowly and waiting for pages to stabilize.
You can always re-record the same site to trigger a new training run.