Integrations · Browsing

Browserbase

A real browser in the cloud for your agents — read pages that block plain fetch, run multi-step browser jobs, and always know what the meter says.

12 actions · out of the box: Browser Operator, Page Researcher, Session Auditor

What your agents get done with it

Read the page that blocks plain fetch

A question becomes URLs, and the pages behind them come back as clean markdown, not HTML soup. When a page answers with a 403 or a captcha wall, one retry goes through the metered proxy — strictly a retry, never the default.

Runs on: browserbase.search · browserbase.fetch_page · browserbase.fetch_proxied

The multi-step browser job, babysat to an answer

"Find the cancellation policy behind their help center and quote it exactly." The job is handed to a real cloud browser in plain English, polled until it lands, and stopped the moment it loops. A failed run reports which step broke.

Runs on: browserbase.agent_run · browserbase.run_status · browserbase.run_messages · browserbase.run_stop

Structured files without the ceremony

Sitemaps, robots.txt, JSON endpoints, RSS — fetched untouched, with status code and content type. An agent auditing a site's crawlability reads the actual files, not a rendering of them.

Runs on: browserbase.fetch_raw

Nothing left running, nothing unexplained on the bill

Live sessions are listed, orphaned ones released, and the meter stops. Ask what a scrape cost and the answer is the exact minutes and bytes consumed — never an estimate.

Runs on: browserbase.sessions_list · browserbase.session_live · browserbase.session_release · browserbase.project_usage

Every action, honestly listed

The complete seeded set — what each one does and costs, stated before an agent may call it. Nothing hidden, nothing padded.

browserbase.searchWeb search
Web search returning URLs, titles, and published dates — links only, never page content. Consumes plan quota.
read-onlyspends vendor credits
browserbase.fetch_pageFetch page as markdown
Fetches a URL through Browserbase and returns the page as markdown — the default reader for articles, docs, and pricing pages. Runs no JavaScript and cannot log in.
read-onlyno vendor charge
browserbase.fetch_rawFetch raw body
Fetches a URL and returns the body unconverted, plus status code and content type — for JSON APIs, robots.txt, sitemaps, and RSS.
read-onlyno vendor charge
browserbase.fetch_proxiedFetch page through proxy
The same page fetch routed through a Browserbase proxy IP — used only as a single retry after a 403, a captcha wall, or a geo block. Costs metered proxy bytes on top of plan quota.
read-onlyspends vendor credits
browserbase.agent_runStart browser agent run
Hands a multi-step browser task in plain English to a Browserbase cloud agent that can click, type, and navigate. Returns a run id immediately — the task is not done yet. Costs browser minutes until it ends; a forgotten run bills.
writespends vendor credits
browserbase.run_statusPoll agent run
Polls one agent run: status, the result once finished, the failure cause, and its session id. Costs nothing itself.
read-onlyno vendor charge
browserbase.run_messagesRead agent run transcript
What an agent run actually did: the chronological assistant and tool messages for one run — which step broke, and when.
read-onlyno vendor charge
browserbase.run_stopStop agent run
Asks a running browser agent to stop — the action that saves money when a run is looping, off-task, or burning minutes.
writeno vendor charge
browserbase.session_liveLive session URLs
Watchable live URLs for one browser session: a fullscreen debugger link plus the URL and title of every open page. Costs nothing.
read-onlyno vendor charge
browserbase.session_releaseRelease browser session
Irreversible: stops a Browserbase session so the meter stops before its timeout. Any agent run using that session is killed, so agents check the live view and ask first.
writeno vendor charge
browserbase.sessions_listList browser sessions
Lists Browserbase sessions by status — the way to find orphaned sessions still burning metered browser minutes. Costs nothing.
read-onlyno vendor charge
browserbase.project_usageBrowserbase project usage
Browser minutes and proxy bytes consumed by a Browserbase project — the exact numbers returned, never an estimate. Costs nothing.
read-onlyno vendor charge

Comes wired into these roles

Every Browserbase action is a capability any agent in your workforce can hold. Out of the box it comes wired into the roles below — a starting point, not a boundary.

Page Researcher

Turns a question or a URL into readable page text, past the bot walls that break plain fetch.

browserbase.search · browserbase.fetch_page · browserbase.fetch_raw · browserbase.fetch_proxied

Browser Operator

Hands a multi-step browser job to a Browserbase cloud agent, then babysits it to an answer.

browserbase.agent_run · browserbase.run_status · browserbase.run_messages · browserbase.run_stop · browserbase.session_live · browserbase.session_release

Session Auditor

What is still running on Browserbase and what it has cost — and kills orphaned sessions.

browserbase.sessions_list · browserbase.project_usage · browserbase.session_live · browserbase.session_release

The full twelve-agent roster, with every action list, is in AI agent examples.

How connecting works

Connect Browserbase once, in the Ahel desktop app, with your own key — stored by your runtime, not by us, removable at any time. From then on any agent whose job needs it can use it: hand over a job, get back a run with a receipt.

Browserbase sells cloud browser time on its own plans — browser minutes, plan quota, and metered proxy bytes. You connect your own Browserbase API key; their prices, no markup.

Costs and limits, honestly

What spends is flagged in the actions themselves: agent_run bills browser minutes until it ends, fetch_proxied meters proxy bytes, search consumes plan quota. Everything else costs nothing beyond your plan.
The catalog carries its own brakes: run_stop ends a looping run, session_release stops the meter on an orphaned session, and both say so in their descriptions.
project_usage reports the exact minutes and bytes consumed — agents report those numbers, never an estimate.

Questions people actually ask

What does Browserbase cost when agents use it?

Whatever your Browserbase plan meters: browser minutes for agent runs, plan quota for searches, and proxy bytes only when the proxied retry is used. Ahel adds no markup, and browserbase.project_usage reports the exact consumption on request.

Can a browser run keep billing after I stop watching?

A forgotten run bills until it ends — the seed says so bluntly. That is why the catalog ships brakes alongside the engine: run_status polling, run_stop for loops, sessions_list to find orphans, and session_release to stop the meter. A fresh install even wires in a Session Auditor role to watch for exactly this.

Can agents log in to websites through Browserbase?

The page fetches cannot — fetch_page runs no JavaScript and cannot log in. Multi-step work goes through agent_run, which drives a real cloud browser that can click, type, and navigate; what it may do there is bounded by the task you hand it and the account you connect.

Do I need my own Browserbase account?

Yes — you connect your own Browserbase API key in the Ahel desktop app, and agents use it on your behalf at Browserbase's own prices. The key stays in your runtime and you can remove it at any time.