AI Discovery Mode

Autonomous workflow learning powered by Gemini AI

What is Discovery Mode?

Discovery mode uses Google Gemini 2.0 to automatically learn and author workflows from natural language descriptions. Submit a task like "create an account on website.com" and Gemini will pilot a browser, analyze the process, and generate optimized HellScript code.

How It Works

Example Usage

# Run discovery
hellcat run "test login flow on github.com" --follow

🔍 Discovering workflow...
📸 Analyzing page structure...
✍️  Authoring HellScript...
✅ Workflow created: "GitHub Login Test" v1.0.0

Execution result:
{
  "success": true,
  "login_button_found": true,
  "form_fields": ["login", "password"]
}

Fuel consumed: 12,847 tokens ($0.08)

Gemini Models

Cost

Discovery mode typically consumes 10-50 credits (~$0.05-$0.25) per task. Once authored, the workflow can be reused unlimited times at standard execution cost (1-3 credits).

Error Recovery

If a workflow fails, Gemini can analyze the error state (screenshots, DOM, console logs) and suggest fixes or retry with an updated approach.

Local Testing

You can test discovery mode locally with the complete Hellcat stack:

# Start local environment
luther local start

# Run discovery
luther local discover "your task" --url https://example.com --follow