Vibe Browse

Built with Hyperbrowser

Control a visible Chrome browser through natural language. Built on HyperAgent and Claude Agents for conversational web automation.

🆕 New: Export Sessions as Training Data

Every conversation = training data for your web agents.

Vibe Browse now exports your browser automation sessions to OpenAI/Anthropic fine-tuning formats. Perfect for training custom web agents on real interactions.

npm start "scrape Hacker News" --export
# → Generates training_data/openai_*.jsonl, anthropic_*.json, trajectory_*.json

What It Does

• Conversational browser automation through natural language chat
• Navigate, click, type, extract data, and observe pages with simple commands
• Interactive CLI that maintains context across multiple requests
• Runs a visible Chrome instance you can watch
• NEW: Export sessions as LLM training data for fine-tuning web agents

Get an API Key

Get your Anthropic API key at https://console.anthropic.com

Setup

npm install

Create a .env file:

ANTHROPIC_API_KEY=your-key-here
ANTHROPIC_MODEL=claude-3-7-sonnet-latest

Usage

Start an interactive session:

npm start

Or with an initial prompt:

npm start "Go to Hacker News and get the top post title"

Example Commands

Go to github.com/hyperbrowser/hyperagent and summarize the README
Navigate to example.com and extract all contact info
Search for TypeScript on Google and click the first result

After each response, continue the conversation or type exit to quit.

Training Data Export

Export your browser automation sessions as LLM fine-tuning data:

# Auto-export when session ends
npm start "scrape Hacker News" --export

# Or export during/after any session
> export

Outputs to ./training_data/ in OpenAI, Anthropic, and raw JSON formats.

Perfect for:

• Fine-tuning your own web agents
• Building domain-specific automation datasets
• Benchmarking agent performance
• Training models on real browser interactions

See TRAINING_DATA.md for full documentation.

Growth Use Case

Perfect for automating growth tasks like scraping competitor data, monitoring pricing, extracting user feedback from review sites, or generating lead lists from directories.

Resources

• Hyperbrowser Docs
• HyperAgent SDK

Follow @hyperbrowser for updates.