Lizard Agent
Drive a browser with deterministic code and typed Jev decisions.
What Jev decides
Which candidate control or action advances the goal, and whether the task is done.
Lizard builds a closed action list from a page, asks Jev which step to take, and verifies constraints with code rather than an LLM in the loop.
- Jev role
- Select a browser action from a bounded page-derived list.
- Input
- The user goal, current page controls, and available action choices.
- Output
- Typed action choices and supporting probabilities.
- Decision frequency
- For each browser decision step.
- Fallback
- Code enforces measurable constraints and stops on CAPTCHAs.
How it works
Overview
The Python browser agent uses Playwright to inspect page controls and offers Jev typed candidate actions. Code executes the selected action, handles concrete constraints such as prices and dates, and records traces and an on-page HUD. The README reports selected demo runs and stops at CAPTCHAs; it does not claim unrestricted browser coverage.
- Language
- Python
- License
- MIT
- Author
- yousudip
Sources
Related projects
Jev Voice Browser
A browser that starts resolving voice commands while the user is still speaking.
What Jev decides
What the user intends, which page target it applies to, whether the command is complete, and whether it is destructive.
Jev × WebMCP
Turn WebMCP schemas and user wording into inspectable browser tool calls with typed Jev decisions.
What Jev decides
Which tool fits, which allowed value each typed parameter should take, and whether optional values were actually stated.
jev-turbo
Choose browser actions from named page controls and check whether a goal is done.
What Jev decides
Which visible control or navigation action to use next, and whether the goal is met.