A browser your code can call.
A goal goes in; schema-valid data comes out. Drive the sites that never shipped an API, stream every step as events, and when a site demands a human, get a pausable state you can route to a person instead of a stack trace.
- No credit card required
- Free plan to start
curl -X POST $BROWZA_API_URL/v1/runs \
-H "Authorization: Bearer $BROWZA_API_KEY" \
-d '{
"task": "Pull last month's invoices from the billing portal and total them",
"config": {
"profile": "billing-portal",
"outputSchema": {
"type": "object",
"properties": {
"total": { "type": "string" },
"invoices": { "type": "array" }
}
}
}
}'What happens between request and result
Scripted replays — a schema-shaped extraction, and a human check your integration survives. Pick one:
The task
What happens
- Opens the billing portal
- Already signed in — the profile remembered
- Clicks into Invoices
- Filters to last month
- Reads all six invoices
- Double-checks the total
Six invoices, one tidy total
$12,940.00 — handed back as a table you can drop straight into a sheet.
Sped up and simplified for the tour — a real run streams every step live in your dashboard, and you can take over at any moment.
What teams build on it
The browser as a dependable primitive, not a flaky headless script.
Typed data from any site
Supply a JSON Schema and the run finishes with a validated extract — structured results, not scraped guesses.
An API for sites without one
Drive the legacy portal, the vendor dashboard, the government site — from code, as if it had endpoints.
Research & monitoring pipelines
Scheduled collection across many sources, with run events you can stream and results you can diff.
QA the real flows
Exercise sign-up, checkout, and settings flows in a real browser, and catch what broke before your users do.
Ops automations
Wire browser work into your stack: create runs from your app, follow events, hand results to the next system.
Sessions that persist
Profiles keep logins alive between runs, and a paused human-check is an event your integration can route to a person — not a timeout.
Honest failure is a feature
The API is designed around the truth that the web fights back.
Verified, not hoped
Runs check the page before spending work on it and validate the result against your schema — a 200 means the shape you asked for.
Paused, not crashed
A human check pauses the run with the session alive and surfaces as an event. Your code alerts a person; the person clicks; the run resumes.
Idempotent, not duplicated
Send an Idempotency-Key and retries return the original run instead of starting a second one — safe to wire into queues.
Fifteen things teams build with the API
Each one names the endpoint it uses — nothing here is aspirational.
Data & extraction
5Typed data, from any website.
Send a goal and a JSON Schema; get back validated, schema-conformant data from sites that never shipped an API.
The integration that was never built.
Reach the systems that will never expose an API — supplier portals, government sites, internal tools — through one HTTP call.
Data behind a login, reached programmatically.
Pull data from systems that require an account, using a saved session a human established once — no credentials in your code.
Paginated data, collected deterministically.
Walk through paginated results without a model call per page — and get a named reason when collection stops.
Files, fetched and returned.
Have runs download statements, reports, and exports from portals, and pull the files out through the API.
Agent infrastructure
5A browser your agent can call.
Give an LLM application the ability to actually use the web — as one API call that returns a result, not a session you have to drive.
Legacy systems, driven through an API.
Put a modern API in front of the internal tool nobody will rebuild — by describing the task instead of reverse-engineering it.
Teach once, invoke by name.
Record a flow in a real browser, and call it afterwards as a parameterised endpoint — deterministic and cheap.
A person, in the loop when it matters.
Let a run pause for a human — for a login, a challenge, or an approval — and continue from exactly where it stopped.
Every run, inspectable afterwards.
Steps, screenshots, page reads, token accounting, and a live stream — so an agent's behaviour is evidence rather than anecdote.
Pipelines & monitoring
3Recurring collection, without the cron box.
Run collection on a schedule and pull structured results into your warehouse — without hosting browsers or a scheduler.
Change detection, described rather than diffed.
Watch pages for meaningful change and get a structured description of what moved — not a raw HTML diff full of noise.
Critical flows, tested in English.
Check that signup, login, and checkout work from the outside — described in a sentence instead of a file of selectors.
Platform
2Your model, your key, your endpoint.
Use your own provider key — or any OpenAI-compatible endpoint — without your credentials ever reaching the browser.
An agent that proves who it is.
Requests are cryptographically signed with Web Bot Auth, so sites can identify Browza instead of guessing about your traffic.
Ship the integration this sprint
Create a key in the dashboard and send your first request — the docs cover the rest.