
Hosted headless browsers for AI agents, with a free plan of 3 concurrent sessions, session replays and the open source Stagehand SDK.
Pick Browserbase when your agent has to log in, click and fill forms on real sites and you would rather not run a Chrome fleet yourself. It replaces self-hosted Playwright clusters, not search APIs. Free sessions stop at 15 minutes, so serious runs need a paid plan with 25 or more concurrent browsers.
Hosted browser sessions with replays plus the open source Stagehand SDK that feeds them.
Stagehand adoption and a captcha and identity stack are hard to copy quickly, but hosted Chrome is a crowded category and agent frameworks can swap providers with a connection string.
Main trade-off
Usage-metered pricing and a 15-minute session cap on the free plan make costs and limits easy to hit.
| Plan | Monthly price | What it includes |
|---|---|---|
| Free | Free | 3 concurrent browsers, 1 browser hour, 3 agent runs, 1,000 search calls, 1,000 fetch calls, 15 minutes per session, $5 in model tokens, 7-day data retention, no captcha solving |
| Developer | $20/mo | 25 concurrent browsers and 15 agent runs. Browser time: 100 hours, then $0.12 an hour. Search costs $7 per 1,000 after the first 1,000 calls; fetch costs $1 per 1,000 after the first 1,000. Proxies: 1 GB, then $12 a GB. Captcha solving is automatic, stealth is basic and data is kept 30 days. |
| Startup | $99/mo | 100 concurrent browsers and 50 agent runs. Browser time: 500 hours, then $0.10 an hour. Search is $7 per 1,000 beyond 1,000 calls; fetch includes 10,000 calls, then $0.50 per 1,000. Proxies: 5 GB, then $10 a GB. It adds automatic captcha handling and basic stealth, with retention of 30 days. |
| Scale | Custom | Custom pricing: 250+ concurrent browsers, 500+ browser hours, usage-based search and fetch, HIPAA BAA and DPA, SSO and SAML, advanced stealth, verified agent identity, 30+ day retention |
Free plan at $0, Developer $20 a month, Startup $99 a month, Scale is custom. Browser hours, proxies, search and fetch calls are metered above each plan's allowance (for example $0.12 per extra browser hour on Developer and $12 per extra GB of proxies). Prices read from browserbase.com/pricing on 2026-10-04.
Stagehand, the MIT-licensed SDK behind Browserbase, has passed 25,000 GitHub stars since its March 2024 release. Browserbase hosts headless Chrome sessions for AI agents, adds search, fetch and agent identity APIs, runs a free plan with 3 concurrent browsers, and holds a SOC 2 Type II report listed on its Trust Center.
Cloud browser sessions
Runs headless Chrome fleets you drive through Playwright, Puppeteer, Selenium or Stagehand, with a concurrency ceiling that rises by plan.
Stagehand SDK
MIT-licensed framework that mixes natural-language steps with Playwright code and has passed 25,000 GitHub stars.
Session recording and replays
Every run is recorded with logs, so a failed login or stuck click can be replayed instead of reproduced.
Search and Fetch APIs
Returns search results and page content as HTML, JSON or Markdown, with 1,000 free calls of each on the free plan.
Agent identity and captcha solving
Automatic captcha solving from the Developer plan, with verified-agent access on Scale for sites that block bots.
Model Gateway
One Browserbase key reaches major models through Stagehand with unified billing, and the free plan includes $5 in tokens.
Most users expect these and they are not currently supported.
| Tool | Price from | Free plan | Runs on | Checked |
|---|---|---|---|---|
| $20/mo | Yes | Web | 4 Oct | |
| $16/mo | Yes | Web | 3 Oct | |
| $19/mo | Yes | Web | 19 Aug | |
| $30/mo | Yes | Web | 3 Oct |
Proprietary hosted service. Stagehand SDK is MIT-licensed.
Browserbase charges $0 for Free, $20 a month for Developer, $99 a month for Startup, and custom pricing for Scale. Developer bundles 100 browser hours and Startup 500, then each extra hour costs $0.12 or $0.10. Above the allowance, search is $7 per 1,000 calls, fetch is $1 or $0.50 per 1,000, and proxy traffic runs $12 or $10 per GB.
Yes. The free plan is capped at 15 minutes per session and 1 browser hour in total, leaves captcha solving off, and keeps data for 7 days. Search and fetch each get 1,000 calls and model tokens come with a $5 credit. That is enough to test Stagehand scripts against real sites, not to run a daily job.
Kernel is the closest hosted-browser competitor on Hokai. Firecrawl and Tavily overlap on the web-data side: they return content from a URL or a question rather than giving the agent a session. Browse AI suits no-code recurring scrapes, and MultiOn is a ready-made browsing agent rather than infrastructure.
Firecrawl turns a URL or query into Markdown or JSON through credit-based endpoints, with an Interact feature for clicking on an already scraped page. Browserbase lends you a full browser session with recordings, captcha handling and identity, which you drive with Playwright or Stagehand. Choose Firecrawl for content pipelines and Browserbase for logins, portals and multi-step forms.
Sign up on the free plan, create an API key, and point Playwright, Puppeteer or Stagehand at a Browserbase session instead of a local Chrome. The docs at docs.browserbase.com carry quickstarts for each framework. Because Stagehand is open source you can first run the same script locally and switch only the browser connection.
Browserbase is a hosted browser infrastructure platform for AI agents, founded in 2024 by Paul Klein IV and based in San Francisco. It runs headless Chrome sessions in the cloud so an agent can log in, click, fill forms and read pages on sites that have no API. The company raised a $40M Series B in June 2025 led by Notable Capital with Kleiner Perkins and CRV, which brought its total funding to about $67.5M.
The product has grown from raw browsers into a set of primitives that share one account. Browsers give an agent a remote Chrome it drives through Playwright, Puppeteer, Selenium or the Stagehand SDK. Search and Fetch return agent-ready results and page content, Identity handles captcha solving and verified-agent access, Runtime hosts the agent code next to the browser, and a Model Gateway bills model tokens through the same Browserbase key. Session recording and replays sit under every run, which makes a failed login or a stuck click something you can watch instead of guess at.
Stagehand is the open source half of the story. It is an MIT-licensed browser automation framework that lets you mix natural-language steps with ordinary Playwright code, and it has passed 25,000 GitHub stars. You can run it against a local browser for free and move to Browserbase when you need concurrency, stealth and recordings. That is also the clearest dividing line against the web-data APIs: Firecrawl and Tavily and Exa start from a URL or a question and return content, while Browserbase gives the agent a real browser session to act inside. Browse AI is the no-code route to recurring scrapes, and Zapier covers simple app-to-app flows, MultiOn is a finished browsing agent rather than infrastructure, and Kernel is the closest direct competitor in hosted browsers. E2B runs sandboxes for code, not browsers, so it solves a different slot in the same agent.
Pricing is plan plus usage. The free plan covers a demo, since sessions stop at 15 minutes. Developer starts at $20 a month, Startup at $99, and Scale is custom and adds HIPAA BAA, SSO and advanced stealth. Overage on hours, proxies and search is metered, so a proxy-heavy scraper costs more than the sticker price suggests.
It fits agent builders who need real logins and form flows: Hermes Agent and OpenClaw builds can point their browser slot at it, and the SDKs plug into LangChain and n8n workflows. Teams in healthcare, fintech and payroll use it for portals that expose no API, and the SOC 2 Type II report on its Trust Center is the reason many of them can. If your agent only needs search results or clean page text, a web-data API is cheaper and simpler. If it needs to act on a page, this is the category to shortlist.
Product facts on this page (pricing, plans, integrations, compliance) are read from Browserbase's own pages and documentation. The verdict is Hokai's reading of them, not the vendor's claim.