Browserbase vs. Skyvern: cloud Chromium runtime or RPA-replacement workflow platform?

Browserbase and Skyvern get compared when an engineering team and an ops team are staring at the same portal problem from opposite sides. Browserbase says: give developers a reliable cloud Chromium runtime, Stagehand primitives, Director, replay, and tracing so they can build an agentic product. Skyvern says: give operations a workflow platform that already knows how to look at a screen, log in, validate actions, extract structured data, and run across portals. If the buyer wants a browser primitive, Browserbase belongs in the shortlist. If the buyer wants the workflow itself to be the product, Skyvern is the closer match.

At a glance

What is Browserbase?

Browserbase runs a custom-patched fork of Chromium for AI-automation use cases (navigator.webdriver=false, "Chrome" UA, kBrowserAliveWithNoWindows for long-lived sessions) inside per-session VMs with hardware virtualization. Each session is reachable over Playwright, Puppeteer, Selenium, or CDP. On top of the runtime, Browserbase ships three productized layers: Stagehand (an open-source AI automation framework with act/extract/observe/agent primitives, ~500K weekly downloads at v3 launch, CDP-native and Playwright-free in v3), Director (a no-code natural-language surface that generates Stagehand scripts from prompts), and Functions (TypeScript code deployed alongside the browser session for low-latency execution).

Compliance is SOC 2 Type 1 + HIPAA, with Type 2 in progress. Pricing is $20/mo for the Developer plan (100 browser hours, 25 concurrent), $99/mo for Startup, with a 60-minute free tier. The 1-minute session minimum is a real cost driver for short-task workflows. Backing is a $40M Series B at ~$300M valuation; customers include Vercel, Perplexity, Stripe, Ramp, Clay, Lovable, Commure, 11x; claimed scale is 100M+ sessions lifetime, 50M in 2025.

What is Skyvern?

Skyvern is an open-source AI browser-automation platform that uses vision-LLMs plus a planner-actor-validator agent loop to drive websites the way a human would, instead of relying on XPath/CSS selectors. The pitch is explicit: replace brittle Selenium/Playwright scripts and human SOPs across portal-driven verticals — healthcare/EHR integration, insurance quoting, government forms (SS4 filings, IRS), procurement / AP teams downloading invoices from supplier portals, payroll, mortgage, job applications.

Skyvern 2.0 hit 85.85% on WebVoyager (vs ~45% for Skyvern 1.0); the full eval is published at eval.skyvern.com. The team also created Web Bench (5,750 tasks / 452 sites) with Halluminate, focused on WRITE tasks — forms, logins, downloads — where most agents fail. The Validator inspects the screen after each action to detect "fake successes." Multi-step workflows are defined in YAML; structured JSON/CSV output schemas validate extraction. Native 2FA/TOTP and CAPTCHA solving are built in. Credentialing is native plus integrations with Bitwarden, 1Password, and Azure Key Vault — credentials never sent to the LLM. A "Route Memorization" / compile-to-code engine lets the agent solve a workflow once, then compile to a fast/cheap Playwright script that self-heals via the LLM when it breaks.

The platform is Apache-licensed open source; you can self-host it (often the choice for health-tech buyers) or run on Skyvern Cloud at ~$0.10 per step/page. Compliance is SOC 2 Type 2 + HIPAA. Funding is a $2.7M seed (Dec 2025), 3,000+ GitHub stars at HN launch and growing. Integrations include Zapier, Make.com, n8n, plus a REST API and an MCP server.

How they compare

Category: infrastructure vs. workflow platform

This is the entire game. Browserbase gives you a Playwright-compatible Chromium session; you write the agent loop. Skyvern gives you a workflow platform with a vision agent loop already running — you define the goal, not the clicks. Stagehand sits between the two in spirit (high-level act("click sign in") calls) but it's still a framework you call from your own code; it does not replace the workflow-platform shape Skyvern targets. The Skyvern blog's framing — "Browserbase manages headless browsers while you write the scripts" — is fair, even if the "60% Browserbase failure rate" cited in the same article is unsourced and shouldn't be reproduced.

Director is not the same bet as Skyvern

The easy mistake is to treat Browserbase Director and Skyvern's hosted workflow product as the same "no-code" category. Director is useful because it can turn natural language into Stagehand code that an engineering team can review, deploy, and maintain. Skyvern is useful because the workflow stays inside the platform: credentials, validation, run logs, retries, integrations, and non-engineer ownership all live there. Director reduces the amount of code a developer writes. Skyvern reduces the need for a developer to own the workflow at all.

Category split (why a head-to-head speed comparison doesn't apply)

Skyvern is not on the public Browser Arena leaderboard, and that omission is itself part of the framing. Browser Arena measures browser runtimes; Skyvern is a workflow platform that runs on top of someone else's browser, with vision-LLM steps and a validator inspecting every action — there's no session endpoint to point a benchmark harness at. Browserbase is on Browser Arena: it ranks #4, is fast, and is also the highest-cost provider on the leaderboard. ComputeSDK's open browser benchmarks (computesdk.com/benchmarks/browsers) publish a second methodology cross-reference that also covers Browserbase. If you need a managed Chromium runtime as a primitive, Browserbase is in the buying conversation; if you need a complete workflow execution layer, the runtime-speed conversation simply isn't the metric — that's the architecture working as intended for Skyvern's category.

Authentication, credentials, identity

Skyvern ships credentials, 2FA/TOTP, CAPTCHA solving, and integrations with Bitwarden, 1Password, and Azure KV natively. The credential never reaches the LLM. For per-customer-account identity in a single workflow (a procurement bot logging into vendor X with vendor X's creds), this is a clean primitive.

Browserbase leaves identity to your code. The Stytch partnership covers Web Bot Auth at the request-signing layer, but credential handling is roll-your-own — env vars, your own secret store, your own data-dir / Context API juggling. For an engineering team that already has secret management and identity in their stack, this is the right primitive. For an ops team without that infrastructure, Skyvern's batteries-included posture is the easier sell.

Pricing model and predictability

Browserbase is metered: $20/mo gets you 100 browser hours, $99/mo gets you 500 hours, with overages at ~$0.10-0.12/browser-hour and ~$10-12/GB residential proxy. The 1-minute session minimum is a real cost on short tasks. Skyvern Cloud charges ~$0.10 per step/page on Basic, Pro, and Enterprise tiers (no public dollars on tier pricing — flag). The OSS self-host option is genuinely free if you run it.

The dichotomy is "browser-hour metering" versus "value-per-step pricing." A noisy variable workload at scale costs less on Browserbase per-hour; a high-value low-volume portal flow (filing a tax form, downloading an invoice) costs less per outcome on Skyvern.

Observability

Browserbase: Live View, Session Replay, Session Inspector, OpenTelemetry tracing, token-level Stagehand reports. Best-in-class for the infrastructure layer. Skyvern: live viewport streaming, action viewer, run logs, full eval results published at eval.skyvern.com. Best-in-class for the workflow layer — you watch your agent reason and act, not just your browser session.

Compliance

Both ship credible posture. Browserbase: SOC 2 Type 1 + HIPAA, Type 2 in progress. Skyvern: SOC 2 Type 2 + HIPAA. Skyvern's HIPAA + open-source-and-self-hostable combination is a real wedge for health-tech buyers who need credential handling on-prem. Browserbase's HIPAA is real and matched.

When to choose Browserbase

• You're an engineering team shipping an agentic product and you write code.
• You want a Chromium-fork-with-AI-automation-patches as your primitive and Stagehand or custom CDP code on top.
• Director's no-code natural-language surface is useful, but it's still output that your team deploys and maintains.
• Debugging matters and you want the most polished defaults in the category.
• Incumbent gravity (Stagehand ~500K weekly downloads, $40M Series B, Vercel/Perplexity/Stripe customer logos) closes your procurement.
• Sessions are 1-100 concurrent and 6h or shorter.

When to choose Skyvern

• You're an ops team replacing RPA — UiPath, Automation Anywhere, Power Automate — and you don't want to write Playwright code.
• Portal-driven workflows: procurement / AP downloading invoices from dozens of supplier portals, healthcare/EHR, government/regulatory forms, insurance quoting, payroll, mortgage, job applications.
• "One workflow, many sites" cross-portal capability matters — the agent adapts to layout differences across vendors.
• Native 2FA/TOTP/CAPTCHA + credential vault + Bitwarden/1Password/Azure KV integrations are out-of-the-box requirements.
• You want open-source-and-self-hostable for HIPAA/security reasons; Apache license is acceptable to your legal team.
• Pricing per outcome (~$0.10/step) maps better to your unit economics than per-browser-hour.

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.

Notte sits closest to Browserbase in category — both are managed cloud Chromium runtimes — but the Anything API and Functions narrow the gap to Skyvern's workflow productization without giving up the SDK shape engineering teams want. Vault and Personas (real inbox + SMS for autonomous 2FA) are integrated runtime primitives where Skyvern's Bitwarden/1Password/Azure KV integrations live in the workflow layer.

Verdict

The Browserbase versus Skyvern question is not "which is better?" — it's "which job am I hiring this tool to do?" Browserbase is the right pick if you have engineering capacity, want a managed Chromium runtime with Stagehand, and intend to ship an agentic product. Skyvern is the right pick if you're replacing portal-based RPA and want a workflow platform that does the whole job — vision-LLM execution, validator-checked actions, native 2FA/TOTP, structured-schema extraction, n8n/Zapier integrations, optional self-host. Both are credible on compliance (SOC 2 + HIPAA). Don't pick on lifecycle speed — Browserbase is on the public Browser Arena leaderboard (at #4, the highest-cost provider on the board) because it's a runtime; Skyvern isn't on the board because it's a complete workflow execution layer. Pick on which side of the build-versus-buy line your team sits.