Good for / not for
- Developers building agents that operate real websites where no API exists.
- Tasks a person would otherwise do by hand in a browser: logging in, filling a form, stepping through pages.
- Anyone weighing project health: 112,754 GitHub stars, last pushed on 5 September 2026, MIT-licensed and self-hostable.
- Teams comfortable with metered pricing: $0.02 per browser-hour with a one-minute minimum, and concurrency that scales with lifetime spend.
- Deterministic automation. An agent deciding what to click is not a script, and it will not behave identically on every run.
- Fetching the contents of a page, which is Firecrawl's job and does not need a browser session.
- Anything involving real credentials right now: an open issue reports that sensitive_data is never redacted from the built-in input_text action in saved history.
What it actually does
The name is the confusing part: Browser Use is a specific open-source product, not a general term for using a browser. What it does is put an AI agent behind a real browser. You describe the task in plain English — find this listing, log in, fill this form — and the agent looks at the page, decides what to click or type, does it, then looks again. That is the answer for sites with no API, where the only interface is the one built for people. It is MIT-licensed and self-hostable, with 112,754 GitHub stars and a push on 5 September 2026, and a reported WebVoyager score of 89.1%. The hosted service is priced by browser time: $0.02 per browser-hour with a one-minute minimum, and 10 to 1,000 concurrent sessions scaling with lifetime spend. The free tier is a one-time $15 credit for eligible Google, GitHub and Microsoft signups, restricted to cheaper models, and it does not renew. Two cautions before you build on it. Point releases land several times a week, which is fast progress and a breaking-change risk for integrators. And an issue opened on 5 September 2026 reports sensitive_data going unredacted in saved history.
Sign up with an eligible Google, GitHub or Microsoft account.
A one-time $15 credit, restricted to the cheaper models, with 10 concurrent sessions. It does not renew. At $0.02 per browser-hour the credit goes a long way, but when it is gone the meter starts.
From the vendor's docsPricing and free-plan limits
Free tier: A one-time $15 credit for eligible Google, GitHub and Microsoft signups, with 10 concurrent sessions and cheaper models only. It does not renew.
| Plan | Price | What you get |
|---|---|---|
| Free | Free |
Limits: Signup-gated to eligible Google, GitHub and Microsoft accounts · Cheaper models only · No renewal — the credit arrives once A one-time signup credit rather than a recurring free tier. |
| Pay-as-you-go | Usage-based |
Metered by browser time rather than by month, so there is no monthly figure to quote. |
| Enterprise | Custom pricing |
No figure is published. |
Prices checked on Sep 6, 2026 against the vendor's pricing page. Plans change often — confirm before you pay.
Three honest limitations
- The free tier is a one-time $15 credit, gated to eligible Google, GitHub and Microsoft signups and limited to cheaper models. It does not renew, so it is a trial budget rather than a free plan.
- An issue opened on 5 September 2026 reports that sensitive_data is never redacted from the built-in input_text action in saved history — a real privacy bug if your agent types secrets.
- Multiple point releases a week — v0.13.9 and v0.13.10 both landed on 4 September 2026 — which means frequent breaking-change risk for anything integrated against it.
Alternatives
- Firecrawl — Choose Firecrawl instead if you only need what a page says; reading content does not need a browser driven at all.
- E2B — Choose E2B instead if the code your model writes needs somewhere safe to run, rather than a website to operate.
- Langfuse — Choose Langfuse instead if the agent already works and the open question is what it did on each run.
Getting started in 5 minutes
- Start with the one-time $15 credit if your Google, GitHub or Microsoft account is eligible, and treat it as a trial budget — it does not renew.
- Give it one small task on a site you own or are allowed to automate, and watch what it clicks before trusting it with anything longer.
- Keep real credentials out of it for now: an open issue reports sensitive_data going unredacted in the saved history of the built-in input_text action.
- Pin a version. Point releases land several times a week, and integrators carry the breaking-change risk.
- Ask first whether you need a browser at all. If the answer is only the text of a page, Firecrawl does that for less.
Browser Use questions
- Is "Browser Use" a product or just a phrase?
- A product. Browser Use is an open-source tool under the MIT licence, with a hosted service at browser-use.com and a repository at browser-use/browser-use carrying 112,754 GitHub stars. The name simply collides with the ordinary words "browser use", which is why a plain web search mixes it with unrelated results about using or downloading browsers.
- What does Browser Use cost?
- The free tier is a one-time $15 credit for eligible Google, GitHub and Microsoft signups, capped to cheaper models and 10 concurrent sessions, with no renewal. After that it is pay-as-you-go at $0.02 per browser-hour with a one-minute minimum, and 10 to 1,000 concurrent sessions scaling with lifetime spend. Enterprise is quoted.
- Browser Use or Firecrawl?
- Drive a browser versus read a page. If the task is to click, type and log in — to do what a person would do on a site with no API — that is Browser Use. If all you want is the content of a page as clean text, Firecrawl does it without a browser session. Firecrawl publishes its own "Firecrawl vs Browser Use" page, worth reading with the obvious caveat that one side wrote it.
- Is it safe to give it my logins?
- Not right now, if you can avoid it. An issue opened on 5 September 2026 reports that sensitive_data is never redacted from the built-in input_text action in saved history, which means typed secrets can persist in the run history. Treat it as an open bug and keep real credentials out until it closes.
- How stable is it?
- Fast-moving. Releases v0.13.9 and v0.13.10 both shipped on 4 September 2026, and several point releases a week is the normal cadence, so pin versions and read release notes. On capability, a reported WebVoyager result puts it at 89.1% success — a benchmark figure, not a promise about your sites.
Sources
- Browser Use pricing (one-time $15 credit for eligible Google/GitHub/Microsoft signups, 10 concurrent sessions, cheaper models only, no renewal; pay-as-you-go $0.02/browser-hour with a 1-minute minimum)
- GitHub repository (MIT, 112,754 stars, last push 2026-09-05, v0.13.9 and v0.13.10 on 2026-09-04)
- Open issue 5679, opened 2026-09-05 (sensitive_data is never redacted from the built-in input_text action in saved history)
- Source code: MIT
Research-compiled, not hands-on tested. Compiled from official docs, the vendor's pricing page and community reviews as of Sep 6, 2026. We haven't tested it hands-on yet.