The platform to build and deploy agents that browse and interact with the web like humans, providing cloud browser sessions, browser agents, and identity/observability infrastructure.
Software capabilities Security & access control Allowed domains Restricts a session's top-level page navigation to an allowlist of domains. Bring your own storage Lets enterprise customers store session artifacts such as downloads, uploads, contexts, and extensions in their own AWS S3 buckets instead of Browserbase-managed storage. CAPTCHA solving Automatically solves supported CAPTCHA challenges encountered during a browser session, with support for custom CAPTCHA selectors. IP allowlisting Routes a session's browser traffic through a customer-supplied static proxy or VPN so the traffic can be allowlisted by firewalls or gateways. Proxies Routes browser session traffic through built-in residential proxies, customer-supplied proxies, or domain-based routing rules for consistent geolocation and network identity; when a Browserbase proxy is used, the browser's timezone is now automatically aligned with the proxy's geolocation, and a timezone can also be forced via a CDP `Emulation.setTimezoneOverride` call. Single sign-on (SAML) Authenticates an organization's team members through a SAML 2.0 identity provider with just-in-time user provisioning. Trusted CA certificates Lets a project upload custom PEM-encoded CA certificates into a session's trust store for TLS-intercepting proxies, or set `ignoreCertificateErrors` (now defaulting to `false`) to bypass certificate validation for a session. Verified Uses a purpose-built Chromium browser with real, OS-specific fingerprints recognized by bot-protection partners to reduce CAPTCHA challenges and blocking on protected sites; the older `advancedStealth` session setting is now deprecated in favor of this. Web Bot Auth Lets an agent cryptographically prove it is operated by an authorized user through Cloudflare's Signed Agents program so participating sites can explicitly allow it. Website authentication Handles website logins, 2FA, and passkey challenges in automated sessions through cookie reuse, live-view handoff, and virtual authenticator configuration. Zero data retention Lets an account run sessions without persisting session logs, video recordings, or replay data to Browserbase-managed storage.
Developer tooling & observability Browse CLI Provides a unified command-line interface for interactive browser automation, Browserbase cloud APIs, Functions, and skill or template management, backed by a background daemon that auto-starts on browser commands and can be force-stopped and restarted for recovery. Browse.sh Publishes an open-source catalog of reusable web-task skills that the Browse CLI and coding agents can browse and install. Browser extensions Loads a custom Chrome extension into a browser session to augment page behavior or work around site-specific challenges. Contexts Persists cookies, authentication tokens, and browser storage across sessions so automations can skip repeated logins. Session Live View Streams a running browser session in real time so a developer or end user can watch or take remote control of it. Session recording Automatically records every session as a video across up to ten tabs, retrievable and downloadable per page for later debugging. Session replay Streams a completed session's DOM replay as HLS so it can be embedded in a customer's own application. Skills Packages reusable browser-automation instructions and site-specific playbooks that AI coding agents install and use automatically. Viewports Configures the width and height of a session's browser window for visual testing and layout-dependent automation. Webhooks Delivers signed HTTP POST notifications to customer endpoints for Function invocation and build lifecycle events, with retries, secret rotation and endpoint management.
Core browser platform & API Browser sessions Creates and manages isolated cloud Chromium browser sessions through a REST API and SDKs for Playwright, Puppeteer, and Selenium automation. Browserbase MCP server Exposes a hosted Model Context Protocol server, built on Stagehand, that gives any MCP client natural-language control of a cloud browser, including explicit per-call session targeting and reattachment via an optional sessionId parameter. Functions Deploys browser automation or agent scripts as serverless functions on Browserbase infrastructure, invokable through an API with automatic session lifecycle management. Model Gateway Routes Stagehand LLM requests to OpenAI, Anthropic, or Google through a customer's Browserbase API key with unified billing and no separate provider accounts.
AI agents & automation Agent Optimize tool Analyzes a completed Agent run's event history and proposes system prompt changes or a reusable Stagehand script to make future runs faster and more reliable. Agents Runs an autonomous browser agent from a single API call that browses, searches, extracts data, uses files, and returns a structured result for a natural-language task. Stagehand Provides an SDK that combines Playwright-level browser control with AI primitives (act, extract, observe) for building browser agents.
Scale & reliability Browser regions Runs a browser session in a selected geographic region to reduce round-trip latency for automation workloads. Concurrency management Enforces and reports per-plan limits on concurrent browser sessions and session-creation rate, returning 429 responses and rate-limit headers when a limit is exceeded. Long sessions Extends session duration beyond the default timeout and lets a session survive disconnects for later reconnection through keep-alive.
Account & team administration Plans and billing Defines Free, Developer, Startup, and Scale plans with browser-hour, proxy, and API call allocations, overage rates, and self-serve plan upgrades and invoicing. Team roles Assigns Admin, Contributor, or Viewer roles to organization members to control dashboard permissions and per-project access.
Content capture PDFs Generates, downloads, and displays PDFs from web pages rendered in a browser session. Screenshots Captures viewport or full-page screenshots from a browser session, including fast capture through the Chrome DevTools Protocol.
File transfer Downloads Syncs files downloaded during a browser session to cloud storage and retrieves them through an API. Uploads Uploads local files, including large files through a dedicated API, into a browser session to interact with file input elements.
Usage tracking & analytics Session metadata Attaches custom key-value JSON metadata to a session so sessions can be tagged, filtered, and queried later. Usage tracking Provides an analytics dashboard and a Project Usage API for tracking session volume, browser minutes, and proxy bandwidth.
Web context retrieval Fetch Fetches a URL server-side and returns raw, markdown or schema-guided JSON content, with optional redirects, insecure-SSL bypass, and proxy routing including geolocation-targeted exit locations. Search Returns fast, token-efficient web search results so an agent can discover relevant pages before opening a browser session.
Browserbase's open-source browser-automation framework for agents, built over Playwright with act, extract and observe primitives, an agent that calls tools, shadow DOM and iframe support, and TypeScript and Python SDKs.
Software capabilities Deterministic page control Browser context Manages pages, cookies, extra HTTP headers, init scripts and the active page within one browser context. Clipboard API Reads, writes, copies, cuts, pastes and clears the browser clipboard from automation code. Domain policy Sets and reads a network-level policy on the browser context governing which domains a session may reach. Locator API Targets elements and performs focused DOM interactions including fill, select, file input, visibility and checked state, text and HTML reads, counting and nth selection. Navigation response inspection Exposes the main-document network response returned by a navigation, including status, headers and body. Page API Navigates, inspects and interacts with a page through goto, click, type, hover, scroll, screenshot, snapshot, evaluate, waits, headers, viewport and event handlers.
AI primitives act() Performs an interaction on the current page from a natural-language instruction, resolving the target element at run time instead of from a fixed selector. Batched operations Runs a callback of Stagehand operations against selected pages in one batched round trip, passing data across the worker boundary explicitly. extract() Pulls structured data off a page against a caller-supplied schema, returning typed results in TypeScript, Python and Go. observe() Discovers and plans the executable actions available on a page, returning them for inspection or later replay. WebMCP page tools Lists and invokes the tools a web page registers for itself, so an agent calls a page's own capability instead of driving its UI.
Observability and logging Run metrics and operation history Reports token counts, model calls and per-operation history for a Stagehand instance, with tracing for individual runs. Session observability Tracks a running automation through live session view, a session dashboard, session analytics and filtering, and performance tracking. Structured logging Controls terminal verbosity and forwards structured logs to files or an external platform, splitting LLM, browser and unfiltered records into separate JSONL streams.
SDKs and onboarding AI coding assistant rules Publishes a rules file that gives an AI coding assistant the constraints it needs to write correct Stagehand v4 code. Playwright and v3 migration Provides mapped migration paths from Playwright and from Stagehand v3 to the v4 API. TypeScript, Python and Go SDKs Ships the same act, extract, observe and page surface across TypeScript on Node 22.18+, Python 3.11+ and Go 1.26+.
Session state and deployment Cloud deployment Documents deploying Stagehand automations to the cloud, including a preview of Browser Functions that run act automations on Browserbase with no infrastructure setup. Multi-tab automation Follows a page that opens in a new tab and acts across multiple tabs rather than losing the script to the original tab. User data persistence Persists browser profile data between runs by pointing a local launch at a user data directory that survives shutdown.
Agent framework integrations Agent framework integrations Gives Claude Code, Codex, CrewAI, LangChain Deep Agents, Eve, Mastra, Pi and the Vercel AI SDK persistent Stagehand browser tools backed by one durable session. Third-party MCP server calls Documents orchestrating calls to Model Context Protocol servers from caller code alongside Stagehand primitives; v4 ships no built-in MCP client or autonomous agent.
Browser environments Browserbase cloud sessions Launches or connects to a Browserbase-hosted browser session, with multi-region placement and API endpoint overrides. Local browser sessions Launches a local Chromium or attaches to any already-running browser exposing a DevTools endpoint, with keep-alive, a fixed CDP debugging port and a DOM settle timeout.
Cost and speed controls Action caching Caches act, observe and extract results server-side on Browserbase-managed infrastructure so repeated steps skip the LLM call. Cost and speed tuning Documents the levers for cutting LLM spend and automation latency, including prompt shape, primitive choice and caching thresholds.
Model configuration Custom models and LLM clients Accepts custom model endpoints, custom headers, per-call model overrides and a caller-extended LLM client alongside the first-class supported providers. Model Gateway routing Routes AI primitive calls through Browserbase's Model Gateway when a model is named without an API key, or straight to the provider when a key is supplied.
Web data access Web fetch Fetches a URL over HTTP and returns its content as markdown, schema-constrained JSON or the raw body, without launching a browser. Web search Runs a live web search from the SDK and returns ranked results with URLs, titles and metadata for the agent to act on.