Cloud browsers for AI agents

Your agent needs a browser. BCTRL gives it one in the cloud - drive it yourself with Playwright, or hand it to a hosted agent, and capture every run as a replayable record. Spaces scope what each agent can touch. Subaccounts bill it back to your customers. You run none of the infrastructure.

Quickstart

Launch a cloud browser and connect to it in 60 seconds.

API Reference

Every endpoint, schema, and example for the HTTP API.

CLI

Drive BCTRL from your terminal or CI.

How BCTRL works

Four nouns, and one decision: do you drive, or does an agent?

Create a space

The boundary. It decides which storage, secrets, and AI credentials the runtimes inside it can reach.

Launch a runtime

A cloud browser starts inside the space, with the stealth, proxy, and fingerprint you configure.

Drive it - two ways

Connect over CDP and drive it yourself with Playwright or Puppeteer, or hand it to a hosted agent.

Inspect the run

Every run is recorded: events, live view, recording, and files. Nothing to wire up.

Two ways to drive

One runtime, two control modes. Hold the controller with a CDP connection from your own process, or let a hosted invocation run. One at a time.

1 import { Bctrl } from '@bctrl/sdk';
2 import { chromium } from 'playwright';
3 
4 const bctrl = new Bctrl({ apiKey: process.env.BCTRL_API_KEY! });
5 
6 const runtime = await bctrl.runtimes.create({ type: 'browser', name: 'first-run' });
7 const { connectUrl } = await bctrl.runtimes.start(runtime.id);
8 
9 const browser = await chromium.connectOverCDP(connectUrl);
10 const page = browser.contexts()[0]?.pages()[0] ?? (await browser.newPage());
11 await page.goto('https://example.com');
12 
13 await bctrl.runtimes.stop(runtime.id);

The platform

Spaces

Scope what a runtime can touch - storage, secrets, AI credentials - per space.

Runtimes

Cloud browsers you configure for stealth, proxy, and fingerprint.

Invocations

Hand a runtime to a hosted agent - act, observe, extract, browser-use.

Runs

Replay any session: events, activity, live view, recording, files.

Vault

Store secrets and TOTP, and inject them into runtimes by reference.

AI Providers

Bring your own model keys. Agents use them without exposing the secret.

Proxies

Attach your own proxy or pull from managed pools by country.

Extensions

Load Chrome extensions into a runtime.

Tools & Toolsets

Give agents callable tools - versioned, promotable, bundled.

Start building

Connect with CDP

Drive a runtime with Playwright, Puppeteer, or any CDP client.

Hosted agents

Run Stagehand and browser-use agents with one call.

Integrations

Wire BCTRL into LangChain, CrewAI, MCP, and computer-use frameworks.

Use cases

Browser agents

Give AI agents a real browser, hosted or self-driven.

Web data extraction

Pull structured data from any site, with proxies and CAPTCHA handling.

Automated testing

Run end-to-end flows in isolated cloud browsers, with full recordings.

Form & login automation

Automate logins and multi-step forms with vaulted credentials and TOTP.

Built for platforms

Most cloud browsers stop at the session. BCTRL is built to sit underneath your product: spaces scope each agent’s blast radius, a full tool-calling system lives in the platform, and subaccounts give every customer an isolated, separately-billed environment under one organization.

Spaces & mounts

Decide exactly what each runtime can reach - storage, vault, AI.

Tools & toolsets

A tool-calling system for agents, versioned and bundled.

Subaccounts

Per-customer isolation and usage metering, out of the box.

Keep exploring

SDK Guide

Spaces, runtimes, invocations, and runs - in depth.

API Reference

The per-endpoint HTTP reference.

Changelog

What’s new across the platform, SDK, and CLI.