It understands your codebase.

World Model

The World Model is a living dependency graph of your project. It tracks relationships between files, functions, and modules with confidence-scored beliefs. As confidence grows through verified work, the AI gains graduated autonomy, progressing from 'observe only' to 'suggest' to 'act independently'. It also generates improvement goals based on detected issues.

See it in action

How it works

1. 1

   As you work, the World Model builds a dependency graph: which files import what, which functions call which, how modules relate.

2. 2

   Each piece of knowledge has a confidence score (0-1) based on how it was acquired (direct observation vs inference) and how recently verified.

3. 3

   Graduated autonomy: low confidence = observe, medium = suggest, high = act. The AI earns independence through correct actions.

4. 4

   Goal generation: the model identifies improvement opportunities (low coverage, missing error handling, redundant code) ranked by impact and confidence.

Key capabilities

Use cases

• Get proactive suggestions for codebase improvements
• Understand how a change in one file affects others
• Let the AI handle routine tasks it's confident about
• Track technical debt and improvement opportunities

Commands