Skyvern

Visit Website

Introduction

Skyvern automates browser workflows with AI. The open-source repository Skyvern-AI/skyvern is licensed AGPL-3.0, and the company runs a hosted cloud. The official site positions it for workflows on portals with no API: logging in, downloading invoices, copying data and filling forms.

Skyvern is not a test framework, but QA teams use browser agents like it to drive flows that traditional selectors handle badly. The site lists CAPTCHA solving, 2FA support, structured data extraction, Python and TypeScript SDKs, webhooks, MCP readiness and Docker self-hosting.

Features

1. Vision plus LLM agent: Completes tasks on unfamiliar pages without hand-written selectors.
2. Workflows: Multi-step workflows with data extraction to CSV or JSON in a schema you choose.
3. CAPTCHA and 2FA: Native CAPTCHA solving; 2FA/TOTP support on Pro.
4. SDKs: Python and TypeScript SDKs.
5. MCP ready: The site states Model Context Protocol support.
6. Self-hosting: The open-source build runs in Docker.
7. Observability: Runs are recorded for review; webhooks and event streaming are available.
8. Compliance: The site lists SOC 2 Type II and HIPAA compliance for the cloud.

Pros

1. Open source: AGPL source means you can inspect and self-host the agent.
2. Handles messy sites: Vision-based actions cope with layouts that defeat brittle locators.
3. Published prices: Hobby and Pro prices and credit allowances are on the pricing page.
4. Large community: The repository has over 23,000 GitHub stars as of review.
5. Enterprise controls: Vault integrations and human-in-the-loop on Enterprise.

Cons

1. Automation first, testing second: There is no built-in assertion and reporting layer like a test runner; you add that yourself.
2. AGPL obligations: Running a modified version as a network service carries AGPL source-sharing duties. Check with your legal team.
3. Non-deterministic: LLM-driven actions can vary between runs, which matters for regression testing.
4. Credit usage: Cloud usage is credit based; heavy suites can be costly.
5. Model costs when self-hosting: Self-hosted runs need your own model access.

Usage and Links (Example)

Cloud quick start, from the site’s code sample (Python):

from skyvern import Skyvern
skyvern = Skyvern(api_key="YOUR_API_KEY")
# create a browser session and run a task in natural language

For self-hosting, follow the Docker instructions in the GitHub repository. For QA use, wrap Skyvern runs in your test runner and assert on the structured output it returns.

Resources:

• Skyvern website
• Pricing
• GitHub repository

Pricing

The open-source edition is free to self-host under AGPL-3.0. From Skyvern’s pricing page, every self-serve cloud plan starts with a 14-day trial (30,000 credits):

1. Hobby: $29 a month. 30,000 credits (about 1,200 actions), 10 concurrent runs, basic CAPTCHA solver, webhooks.
2. Pro: $149 a month. 150,000 credits (about 6,200 actions), 25 concurrent runs, advanced CAPTCHA, 2FA/TOTP, residential proxy.
3. Enterprise: Custom. 100 concurrent runs, HIPAA, SOC 2 report, vault integrations, human-in-the-loop.

Recommended for?

1. QA engineers automating flows on third-party portals where selectors are unstable.
2. Teams that want an open-source agent they can self-host.
3. Operations-style checks such as verifying a partner portal still works end to end.
4. Not a fit as your only regression framework; pair it with Playwright or Selenium for deterministic checks.

Conclusion

Skyvern is a capable open-source browser agent with a hosted cloud and clear pricing. It shines where pages are unpredictable and no API exists.

For testing, treat it as a driver and add your own assertions and reporting. Review the AGPL terms before modifying and hosting it.