Browserbase Changelog

Browserbase Agents is a managed solution that enable you to deploy agents on the web with a simple API call. No code to maintain or to deploy, you describe the task goal in natural language and trigger a run using a simple prompt.

Browserbase Agents is the perfect fit to reliably run at scale if you want to:

• Automate a site too dynamic or varied to script reliably
• Scale automation past a handful of sites without one script per target
• Ship browser agents without standing up infrastructure (compute, orchestration, evals)
• Cut maintenance on fragile, high-upkeep scripts

Learn more

Stagehand 3.6.0 is live.

With WebMCP support, a page can expose typed, first-party tools directly to your agent — so Stagehand calls those tools instead of inferring actions from the DOM. Discover what a page offers with page.listWebMCPTools() and run one with page.invokeWebMCPTool().

const page = stagehand.context.pages()[0];
await page.goto("https://example.com");

// Discover the tools the page exposes
const tools = await page.listWebMCPTools();

// Invoke one directly
const result = await page.invokeWebMCPTool(
  "searchFlights",
  { origin: "SFO", destination: "JFK" },
  { frameId: tools[0].frameId },
);

Read the WebMCP docs →

Plus:

• Azure OpenAI auth via Microsoft Entra ID.
• claude-fable-5 support, with adaptive thinking and the new xhigh effort.
• Smarter act caching — URLs that differ only in query-param order now hit the cache.

Fixes:

• Shadow-root XPath resolution so deterministic actions can target elements inside web components.
• Security and dependency updates.

Stagehand 3.5.0 is live.

Pass screenshot: true to extract() to send the viewport screenshot alongside the accessibility tree, so the model can read charts, layouts, and visual-only content the a11y tree misses.

const data = await page.extract({
  instruction: "extract the pricing tiers and prices",
  schema,
  screenshot: true,
});

Plus:

• Native clipboard API for copy/paste flows.
• Claude Opus 4.8 is now supported for Computer Use (CUA).
• Google Vertex AI auth parameters across the SDK and API schemas.
• Use ignoreDefaultArgs to drop chrome-launcher's built-in flags (e.g. --disable-extensions) when running locally.
• Snapshots now include [selected] / [checked] state for more reliable form handling.
• Constructor and per-request model config now forwards to API-backed sessions.

Fixes:

• Anthropic CUA triple_click action mapping.
• structuredOutputMode for newer Anthropic models.
• Local viewport settings are now respected when attaching over CDP.

Use the updated Fetch API to fetch and return web content as markdown or structured JSON from any URL. Faster and cheaper than spinning up a browser, in formats that are easier for humans and models to read.

Data cap raised from 1MB to 5MB.

Extract ~$4 / 1k pages, ~$7 / 1k with proxies.

Read the docs →

Session replays can now be embedded directly in your product and streamed to your end users within seconds of a session ending.

Includes:
- API for embedding session replays in your own product
- Playback in any HLS-compatible player
- Storage, fMP4 encoding, and CDN delivery handled by Browserbase

Your backend fetches the session's .m3u8 playlist from our API, your frontend hands it to a player, and end users stream segments directly from our CDN

Free on every plan, up to 120 sessions per minute.

Learn more

Stagehand 3.4.0 is live.

Use ignoreSelectors to keep noisy parts of a page out of extract() and observe(): ads, nav, modals, related posts, and anything else your agent should ignore.

const article = await stagehand.extract(
  "extract the article title and body",
  z.object({
    title: z.string(),
    body: z.string(),
  }),
  {
    ignoreSelectors: [".ad", ".newsletter-modal"],
  },
);

Plus:

• Agent variables are now supported in the Stagehand API schema without the experimental requirement.
• Agent mode now defaults to hybrid for compatible models and DOM mode otherwise.
• New CUA model support: openai/gpt-5.4-mini, openai/gpt-5.5, and anthropic/claude-haiku-4-5.
• Better OOPIF frame handling and stronger observe element ID prompting.

Read the Stagehand docs for extract() and observe().

New and improved Downloads API

Every file your agent downloads is now individually addressable.

What changed?

The Downloads API now treats every file as a first-class object with its own ID and metadata. You can list downloads for a session and filter by filename, MIME type, size, or timestamp; fetch an individual file or just its metadata; and delete a single file on demand.

The session-level zip endpoint continues to work, no migration required. Your existing downloads are already addressable.

Read the docs →

Stagehand 3.3.0 is live.

Verified mode for improved agent identity. Sessions get a verifiable identity that helps on sites that gate bot traffic. Enable in your session params:

const stagehand = new Stagehand({
  env: "BROWSERBASE",
  browserbaseSessionCreateParams: {
    browserSettings: {
      verified: true,
    },
  },
});

Plus:

• Adaptive thinking with Anthropic models. Claude CUA now scales the extended-thinking budget to match task complexity — faster on simple steps, deeper when it counts.
• Agent usage in stagehand.metrics. agent.execute() calls now contribute to stagehand.metrics on API-backed sessions, so token counts and model calls land where you expect them.
• Strict structured outputs. Supported model paths now use strict JSON schema enforcement, eliminating malformed responses on extract() and other structured calls.
• Clearer file upload elements. File inputs are now explicit in the page snapshot, so agents can identify and drive upload flows reliably.

Managing model providers adds friction to every Stagehand setup: separate API keys, accounts, rate limits, and tier gating. Model Gateway removes that layer.

Pass your Browserbase API key and pick a model. We handle routing, retries, and billing at market price with no markup.

const stagehand = new Stagehand({
   apiKey: process.env.BROWSERBASE_API_KEY,
   model: "openai/gpt-5", // or “anthropic/claude-sonnet-4-6” / “google/gemini-3-flash-preview”
});

Switching models is a one-line change. No new accounts, no reworked setup. Rate limit handling, retries, and action caching all run through Browserbase automatically.

Supported: OpenAI, Anthropic, Gemini. Need something else? Reach out.

Browserbase and Prime Intellect have partnered to make browser agent training more accessible. You get scalable browser environments, live website access, and the full Browserbase platform as your training substrate with Prime Intellect’s full training platform.

BrowserEnv, the reinforcement learning environment built on this foundation, is available today at browserenv.com.

Most web agents need to be able to search the web before performing actions on a website. Browserbase Search, powered by Exa, now handles searching with a single API call.

Send a plain-text query and get back the URLs your agent needs to act on next.

import { Browserbase } from "@browserbasehq/sdk";

const bb = new Browserbase({ apiKey: process.env.BROWSERBASE_API_KEY! });

const searchResponse = await bb.search.({
  query: "browserbase documentation",
  numResults: 10,
});

Results return a url, title, and description, up to 25 results.

Each plan includes 1,000 free searches.

Learn more

We love improving our Free plan.

Free users can now use up to 3 browsers concurrently (previously this was limited to 1).

Happy building!