Skyvern is an open source browser automation tool that uses LLMs and computer vision to drive websites like a human — a self-hostable RPA alternative that fills forms, clicks through flows, and extracts data without brittle selectors.

What is Skyvern?

Skyvern is an open source browser automation tool that combines large language models with computer vision to operate websites the way a person would. Instead of hard-coded XPath or CSS selectors, it uses a swarm of AI agents to look at a page, understand it, and decide what to click, type, or extract — so automations keep working when a site’s layout changes.

What is Skyvern best for?

Skyvern is best for teams automating repetitive, browser-based work across many sites that have no API — invoice downloads from vendor portals, form filling and submissions, account-payable flows, QA testing, and structured data extraction. Because it reads pages visually, it suits “edge-case-heavy” workflows where traditional selector scripts break on every redesign.

What can Skyvern do?

Drive pages with natural language — act, extract, validate, and prompt methods layered on a Playwright Page object, in Python or TypeScript.
Run multi-step workflows — chain tasks with loops, conditionals, file parsing, and HTTP requests, not just single actions.
Extract structured data — return JSON or CSV, optionally validated against a schema you supply.
Handle authentication — 2FA via TOTP, email, and SMS, plus password-manager integrations (Bitwarden, 1Password, LastPass).
Solve CAPTCHAs and use proxies — native CAPTCHA solving and residential proxy targeting by country/state/zip (cloud-focused features).
Bring your own model — works with OpenAI, Anthropic Claude, AWS Bedrock, Google Gemini, Ollama, and OpenRouter.
Debug visually — livestream the viewport in real time and replay every run with step-by-step screenshots and the agent’s reasoning.
Plug into other tools — integrations for Zapier, Make, n8n, and Model Context Protocol (MCP) support.

Where does Skyvern fall short?

Vision adds latency and cost. Every step runs page understanding through a vision LLM, so Skyvern is slower and more expensive per action than a selector-based script — and you pay for the model tokens on top.
The best anti-bot features are cloud-only. Native CAPTCHA solvers and managed anti-bot detection are reserved for Skyvern Cloud; the self-hosted AGPL build leaves you to handle bot detection yourself.
AGPL-3.0 licensing. The copyleft license can be a blocker for companies that want to embed Skyvern in a proprietary product without releasing their own source or buying a commercial arrangement.

Is Skyvern free?

Partly. The self-hosted code is free and open source under AGPL-3.0 — you only pay for your own infrastructure and LLM API usage. Skyvern Cloud is the paid, managed option: a free tier with 5,000 credits and 1 concurrent run, then Hobby at $29/mo (30,000 credits, 10 concurrent runs), Pro at $149/mo (150,000 credits, advanced CAPTCHA solver, proxies), and custom Enterprise pricing.

What does Skyvern replace?

Skyvern positions itself as an open source alternative to enterprise RPA platforms like UiPath, Automation Anywhere, and Blue Prism. Where those tools rely on recorded selectors and per-robot licensing, Skyvern drives the browser with AI and lets you self-host — trading seat-and-bot fees for infrastructure plus model costs.

FAQ

Is Skyvern open source? Yes. The core is released under the AGPL-3.0 license on GitHub. Some anti-bot and CAPTCHA-solving capabilities are held back for the managed Skyvern Cloud offering.

Can I self-host Skyvern for free? Yes. Run it via Docker Compose or pip and point it at your own LLM. You pay only for your server and the model API calls — there’s no license fee for the self-hosted build.

What do I need to run Skyvern? Python 3.11–3.13, a PostgreSQL database (SQLite works by default for local runs), Docker (optional), and an API key for a supported LLM such as OpenAI, Anthropic, Gemini, or a local Ollama model.

Is Skyvern a good UiPath alternative? For AI-driven, browser-based automation it’s a strong fit — it survives layout changes better than selector scripts and can be self-hosted. It’s less suited to desktop-app RPA, where UiPath’s mature Windows automation still leads.