Skyvern
by Skyvern AI
Open-source browser automation that uses LLMs and computer vision
Last reviewed 2026-06-19
Skyvern automates browser-based workflows using LLMs plus computer vision. Instead of brittle CSS or XPath selectors that break when a site changes, it takes a screenshot and uses a vision model to find and act on the right element, which lets it operate on sites it has never seen before and survive layout changes. It positions itself as a replacement for fragile scrapers and traditional RPA, and exposes natural-language primitives on a Playwright-compatible SDK. Skyvern ships both as a self-hostable open-source project (AGPL-3.0, Docker) and as Skyvern Cloud, a managed platform with a no-code visual workflow builder, a Copilot chat, record-and-replay, anti-bot and CAPTCHA handling, geo-targeted proxies, and 2FA support. It is Y Combinator-backed and is marketed for regulated, form-heavy verticals like insurance, recruiting, procurement, and healthcare.
What it can do
Act on unfamiliar sites via computer vision
SupervisedTakes a screenshot and uses a vision model to locate and click the right element, so it works on sites with no pre-mapped selectors and survives layout changes.
sourceOrchestrate multi-step workflows
AutonomousRuns configured workflows with loops, conditionals, 2FA, and data extraction end to end once defined.
sourceFill forms and download documents across portals
SupervisedCompletes form-heavy tasks such as insurance quotes, invoice retrieval, and government forms.
sourceBuild automations from plain language or an uploaded SOP
CopilotA Copilot chat and no-code builder let a human describe or upload a process that the agent then executes.
source
Strengths
- +Vision-based approach is resilient to site changes and works on never-before-seen pages, unlike selector-based RPA
- +Serves both developers (SDK) and non-technical ops teams (no-code builder, Copilot, SOP upload)
- +Enterprise posture with self-host option and a permissive cloud free tier
Limitations
- −AGPL-3.0 license is copyleft and can block some commercial closed-source self-hosting (vs MIT peers)
- −Paid cloud pricing is opaque, with tier prices gated behind signup or a demo
- −Vision-per-step is token-heavy, so high-volume autonomous workflows can be costly and not fully hands-off
Overview
Skyvern automates browser workflows with LLMs and computer vision. Rather than fixed selectors, it screenshots the page and uses a vision model to find and click elements, which lets it work on unfamiliar sites and survive redesigns. It is pitched as a more durable alternative to scrapers and legacy RPA.
What it does
It navigates sites, fills forms, downloads documents, and orchestrates multi-step workflows with loops, conditionals, and 2FA. A configured workflow can run end to end (autonomous for that defined task), while building and supervising the automation is human-driven. A Copilot chat and no-code builder let non-developers describe or upload a process.
Integrations & setup
Self-host via Docker or use Skyvern Cloud. It is model-agnostic (OpenAI, Anthropic, Google, AWS Bedrock, Ollama and OpenAI-compatible endpoints), integrates with Zapier, Make, and n8n, supports password managers and MCP, and exposes a REST API.
Pricing
The open-source project is free under AGPL-3.0. Skyvern Cloud has a free tier with monthly credits; paid tier prices are gated behind signup, and enterprise is custom.
Best for / not for
Best for teams automating form-heavy, regulated workflows (insurance, recruiting, healthcare, procurement) that need resilience to site changes. Less ideal where the AGPL license conflicts with closed-source distribution or where per-step vision cost is prohibitive.
Alternatives
Browser Use is the closest open-source peer (MIT, framework-first); MultiOn and OpenAI Operator are hosted browser agents; UiPath and Automation Anywhere are incumbent RPA platforms adding AI.
What people are saying
We aggregate real LinkedIn discussion into sentiment for the agents people search most. Skyvern isn't tracked yet, want it added? Request tracking.
FAQ
How is Skyvern different from traditional RPA?+
Traditional RPA relies on fixed selectors that break when a site changes. Skyvern uses a vision model to find elements visually, so it can operate on sites it has never seen and tolerate layout changes.
Is Skyvern open source?+
Yes, the core project is open source under AGPL-3.0 and can be self-hosted via Docker. Skyvern Cloud is a managed paid version with a no-code builder and a free tier.
Sources
- Skyvern (GitHub) · accessed 2026-06-19
- Skyvern (official site) · accessed 2026-06-19
- Skyvern pricing · accessed 2026-06-19
- Launch HN: Skyvern (YC S23) · accessed 2026-06-19
Last reviewed 2026-06-19