Skyvern – AI Browser Automation with Vision LLMs

TL;DR

TL;DR: Skyvern is an open-source AI agent that drives browsers using vision LLMs instead of brittle XPath selectors — it can navigate any website it has never seen before.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: skyvern.com — verified HTTP 200
• Source repository: github.com/skyvern-ai/skyvern — README read in full
• License: AGPL-3.0 (verified via GitHub API)
• HN launch thread: news.ycombinator.com/item?id=41936745
• Source last checked: 2026-07-24 (commit 4a7f2c1, pushed 2026-07-24T06:59:49Z)

What Is Skyvern?

Skyvern automates browser-based workflows using LLMs and computer vision. Unlike traditional browser automation that relies on DOM parsing and XPath selectors, Skyvern uses Vision LLMs to understand a page’s visual layout and plan interactions adaptively.

The project describes itself as:

Skyvern automates browser-based workflows using LLMs and computer vision. It provides a Playwright-compatible SDK that adds AI functionality on top of Playwright, as well as a no-code workflow builder to help both technical and non-technical users automate manual workflows on any website.

Skyvern was inspired by the task-driven autonomous agent design from BabyAGI and AutoGPT, with the key addition of browser interaction through Playwright.

The core advantages over XPath-based automation:

1. Zero per-site configuration — Skyvern can operate on websites it has never encountered before
2. Resistant to layout changes — no hardcoded selectors to break when a site redesigns
3. Generalizable workflows — a single workflow can be applied across many similar websites

Setup Options

Skyvern offers two setup paths: a managed cloud service or a self-hosted option.

Skyvern Cloud

Skyvern Cloud is a hosted version at app.skyvern.com. It handles infrastructure, parallel execution, anti-bot detection, proxy networks, and CAPTCHA solving. Sign up at the link above to get started without managing servers.

Local Setup: pip install

Requires Python 3.11, 3.12, or 3.13.

pip install "skyvern[all]"

Then initialize and run:

skyvern quickstart
skyvern run server

The quickstart command defaults to SQLite at ~/.skyvern/data.db. For a local Postgres container instead:

skyvern quickstart --postgres

To use an existing Postgres database, pass --database-string with your connection string.

Local Setup: Docker Compose

For a fully containerized setup with Postgres, API, and UI:

git clone https://github.com/skyvern-ai/skyvern.git && cd skyvern
cp .env.example .env
# edit .env to add your LLM API key
docker compose up -d

Open http://localhost:8080 when ready.

LLM Provider Configuration

Skyvern needs an LLM API key. After installation, configure your provider in .env:

# .env
OPENAI_API_KEY=sk-...
# or ANTHROPIC_API_KEY=sk-ant-...

The quickstart wizard prompts for this if running interactively.

How the Agent Works

Skyvern uses a swarm of agents to comprehend a website and plan/execute actions:

1. Perception agent — reads the page using computer vision
2. Planning agent — determines the next action based on the goal
3. Action agent — executes the action through Playwright

This is distinct from Playwright scripts with fixed XPath selectors. Skyvern’s Vision LLM approach means it can reason about a page’s visual structure — buttons, forms, modals — even when the underlying HTML is unfamiliar.

Skyvern 2.0 scored 85.8 percent on the WebVoyager benchmark, a test of autonomous web navigation performance.

Architecture

The main components:

| Component | Role | |---|---| | skyvern CLI | Installation, quickstart, server management | | Playwright SDK | Browser control layer | | Vision LLM | Page understanding and visual reasoning | | Task queue | Parallel workflow execution | | SQLite / Postgres | Workflow state and data persistence |

Practical Evaluation Checklist

• GitHub stars: 22.5k (as of July 2026)
• License: AGPL-3.0
• Language: Python
• Python versions: 3.11, 3.12, 3.13
• Browser backend: Playwright
• Cloud offering: Yes (Skyvern Cloud at app.skyvern.com)
• Self-hosted: Yes, via pip or Docker Compose
• No-code builder: Yes, visual workflow builder included

Known Issues

A known bug in skyvern==1.0.31 causes a SQLite migration error:

sqlite3.OperationalError: table organizations already exists

Fix by removing the stale database and upgrading:

rm ~/.skyvern/data.db
pip install --upgrade skyvern
skyvern quickstart

Version 1.0.32 and later contain the fix.

FAQ

Q: How is Skyvern different from a Playwright script? A: Playwright scripts use hardcoded XPath or CSS selectors that break whenever a website updates its layout. Skyvern replaces those selectors with Vision LLMs that can understand a page visually and adapt, so the same workflow survives site redesigns.

Q: Does Skyvern handle CAPTCHAs? A: Skyvern Cloud includes built-in CAPTCHA solving. Self-hosted setups do not include this — you would need to integrate your own CAPTCHA solver if the target sites are protected.

Q: What LLM models does Skyvern support? A: Skyvern supports any LLM with a vision capability. Configuration is done via API key in the .env file. Check the official docs for the current list of tested providers.

Q: Can Skyvern run workflows in parallel? A: Yes. Skyvern Cloud runs multiple instances in parallel by default. Self-hosted Docker Compose also supports parallel task execution through the task queue.

Conclusion

Skyvern targets a real problem: XPath-based browser automation is fragile. When a website changes its class names or DOM structure, every automation script breaks. Skyvern’s vision-first approach means workflows survive site changes and can even generalize across similar websites with no per-site code.

The self-hosted setup via pip or Docker Compose is accessible for anyone with Python 3.11 plus and an LLM API key. The cloud option removes infrastructure overhead entirely.

If you are building automations against websites you do not control — price monitors, data scrapers, form submissions, recurring reporting tasks — Skyvern is worth evaluating. The 22.5k star count and active development signal it is beyond the prototype stage.

Try the cloud version at app.skyvern.com or run it locally:

pip install "skyvern[all]"
skyvern quickstart