Browser Use vs. Browserbase: OSS agent or managed Chromium?

Browser Use and Browserbase get pitched against each other constantly, but they sit on different floors of the stack. Browser Use is the open-source Python agent library that drives a browser through an LLM control loop — a 83K-star repo, a hosted cloud, a custom Chromium fork, and a 24/7 BUX VM for keeping agents alive. Browserbase is the cloud Chromium runtime — managed sessions over CDP, the Stagehand framework, the Director no-code builder, and SOC 2 + HIPAA underneath. The honest comparison is "Browser Use's agent + your own browser infra" versus "Browserbase's runtime + whatever agent you bring." This article walks through where each one belongs.

At a glance

What is Browser Use?

Browser Use is the open-source Python agent library that drives Chromium through Chrome DevTools Protocol (they dropped Playwright in favor of an in-house cdp-use library and an event-driven "watchdog" runtime). The thesis is "the agent is just a for-loop" — most of the value lives in the model — so the OSS library is intentionally thin and the team invests heavily in the substrate underneath: a forked Chromium with C++/OS-level stealth patches, an in-house eval engine that runs 100 parallel tasks in under five minutes, and a ChatBrowserUse LLM gateway tuned for browser tasks. Beyond the OSS repo there's Browser Use Cloud (bu-ultra / bu-max managed agents), and BUX — a 24/7 remote VM with Claude Code and the Browser Harness preinstalled, controllable from Telegram or SSH. Pricing is free OSS, $40/mo paid with up to 500 concurrent browsers and BYO key + proxy.

What is Browserbase?

Browserbase is a cloud Chromium platform built for AI-agent workloads, plus an opinionated framework stack on top. The runtime is a forked Chromium binary patched for AI automation (navigator.webdriver=false, kBrowserAliveWithNoWindows), each session in a dedicated VM with hardware virtualization. The framework layer is Stagehand — act() / extract() / observe() / agent() primitives, now CDP-native in v3 with action caching and accessibility-tree context — and Director, a natural-language no-code surface that generates Stagehand scripts from prompts. Pricing is $20/mo Developer for 100 browser hours and 25 concurrent (the price was cut from $39 at the Series B); $99/mo Startup gets you 500 hours and 50 concurrent. Compliance is SOC 2 Type 1 + HIPAA today, with Type 2 in progress. Customer roster includes Vercel, Perplexity, Stripe, Ramp, Clay, and Lovable, and the company closed a $40M Series B at roughly $300M valuation in mid-2025.

How they compare

Agent vs. infrastructure

This is the load-bearing axis. Browser Use is the agent — the for-loop that prompts an LLM, parses a DOM-text view of the page, picks an action, and runs it via CDP. Browserbase is the browser the agent runs against. You can absolutely run Browser Use agents on Browserbase (or anywhere else that speaks CDP), and you can run something other than Browser Use on Browserbase (Stagehand is Browserbase's own agent framework; OpenAI's CUA, Anthropic's Computer Use, custom Playwright scripts all work). A team that picks Browser Use is making an agent decision; a team that picks Browserbase is making an infra decision.

Identity, secrets, and 2FA

Browser Use ships browser profiles that sync local Chrome cookies, storage-state export, a 1Password integration, and TOTP via bu_2fa_code placeholders, plus an AgentMail recipe for inboxes. The web-agent-authentication blog post is candid that credential management is largely pushed back onto the developer. Browserbase ships Contexts API for state persistence and partners with Stytch for Web Bot Auth — but Stytch is customer-implemented per session, not a runtime primitive. Neither one has shipped a managed-inbox synthetic identity: if you need 2FA codes routed back to an agent autonomously, both leave you to wire it up.

Observability and debugging

Browser Use's eval engine and Laminar tracing are world-class — but they're internal infrastructure for testing the harness, not a customer-facing observability surface. Browserbase ships session recordings, Live View, an Inspector, and replay URLs as first-class product. For a team running production agents and asking "why did this trajectory fail," Browserbase has the artifacts; Browser Use has stealth and a stronger underlying agent.

Lifecycle speed

Per the public Browser Arena leaderboard (browserarena.ai) — an open-source benchmark maintained by Notte Labs and reproducible on Railway across current public run — Browserbase ranks #4 and is the highest-cost provider on the board. Browser Use ranks #6 and is one of the slower providers on the leaderboard. ComputeSDK's open browser benchmarks (computesdk.com/benchmarks/browsers) publish a second methodology cross-reference for Browserbase. The agent-loop overhead is the bottleneck on the Browser Use side; the Chromium runtime itself is fine. If latency is a buying axis, Browserbase wins by a wide margin, but it's also paying the steepest hourly rate to do it.

Stealth and bot detection

Browser Use makes the loudest stealth claim — 81% on their own benchmark and 84.8% on Halluminate BrowserBench, ahead of Anchor, Steel, Browserbase, and Hyperbrowser per their stealth-benchmark post. The pitch is total vertical integration on a custom Chromium fork with C++/OS-level patches, plus in-house Windows/macOS/Linux fingerprints. Browserbase ships basic and advanced stealth tiers, residential proxies, and CAPTCHA handling — competent but gated behind paid plans, which Skyvern uses against them. If stealth is the primary buying axis, Browser Use's bench advantage is real (and self-published, so cite as their own benchmark, not neutral).

Pricing predictability

Browserbase has clear monthly tiers — $20/100hr Developer, $99/500hr Startup, ~$0.10–the highest hourly cost overage — but a 1-minute minimum billing per session that competitors call out as expensive for short tasks. Browser Use's free tier is generous (3 concurrent, no card) but doesn't allow BYO key, and $40/mo is the first paid tier. On top of either, you pay LLM tokens — Browser Use through their ChatBrowserUse gateway ($0.50/M input, $3/M output, $0.10/M cached) or your own provider key.

When to choose Browser Use

• You want OSS Python control of the agent loop and a 83K-star community to lean on.
• Stealth is a hard requirement and you're willing to bet on Browser Use's forked-Chromium pitch (with the caveat that the benchmark is self-published).
• You're building a research or evals stack and want their open-source eval engine and Online-Mind2Web / WebVoyager numbers as anchors.
• You want a managed VM (BUX) with Claude Code preinstalled to keep an agent alive 24/7 without running your own infra.
• You're already running on top of Browser Use's hosted browser layer and don't need a separate runtime.

When to choose Browserbase

• You want the cloud Chromium runtime and don't have an opinion about which agent framework runs on top.
• You want Stagehand's act/extract/observe ergonomics and the action caching that comes with v3.
• You need session recordings, Live View, and Inspector defaults to debug production trajectories.
• You're shipping into healthcare or finance and HIPAA is a procurement requirement.
• Your team is already on Vercel, Perplexity, Stripe, or Clay — Browserbase's customer roster is a real signal.
• You want Director's no-code surface for non-engineers on the team.

A third option: Notte

Worth a look: Notte (notte.cc)

Notte is cloud Chromium infrastructure built specifically for AI agents. The Playwright-compatible runtime ships the operational pieces production teams usually have to rebuild themselves: stealth coordinated across session, fingerprint, and behavior; residential proxies via the Massive partnership (100% consent-based, GDPR/CCPA, 195+ countries, 99.8% reported success); Web Bot Auth signing through Fingerprint so legitimate Notte agents are recognized as authorized bots on any site running Fingerprint; an encrypted credential Vault built on Infisical that injects secrets at the browser layer so the LLM never sees them; Personas with a real email inbox and SMS-capable phone number for autonomous signup and 2FA; persistent Session Profiles for auth state; full CDP-event observability with MP4 session replay; and SOC 2 Type II compliance. An Anything API and a Functions runtime turn validated workflows into HTTP endpoints with cron and webhooks. Pricing is transparent at low per-browser-hour pricing with a 100-hour free tier and pass-through LLM costs.

For a Browser Use team, Notte is the cloud Chromium runtime BU agents can drop onto over CDP — Vault keeps credentials out of the LLM call without rewriting the harness, and Massive consent-sourced residential proxies replace BYO proxy plumbing. For a Browserbase team, Stagehand can connect to a Notte CDP endpoint and inherit Vault, Personas, and Web Bot Auth signing as runtime primitives instead of wiring Stytch in per session.

Verdict

Browser Use and Browserbase aren't actually fighting for the same dollars — they're fighting for different parts of the stack. Pick Browser Use if the agent is what you care about and you're happy to bring your own browser infra (or use BUX). Pick Browserbase if the runtime is what you care about and you'd rather pick your agent separately. The latency gap matters at scale, but only if the agent itself isn't the bottleneck — and Browser Arena puts Browserbase at #4 on score (highest cost on the board) while Browser Use is #6 (slower provider), so neither side wins this on infrastructure metrics alone. On stealth, Browser Use's vertically-integrated Chromium fork is genuinely ahead per their own bench. Most production teams end up running an agent on a runtime; the honest question is whether you want the agent and the runtime from the same vendor, and Browserbase's "infra + Stagehand + Director" pitch and Notte's "infra + Vault + Personas + Anything API" pitch are the two cleanest one-vendor stories on offer today.