behave

What problem does it solve?

Analyze and diagnose LLM behavioral patterns in chat sessions by applying structured hypothesis testing to surface root causes and improvement paths.

Core Features & Use Cases

• Loop detection: identify repeated prompts and responses that waste cycles.
• Context degradation: detect forgotten details across turns and context loss.
• Decision inefficiency: surface indecision and delayed commitments.
• Cognitive overload signals: flag confusion, frequent clarifications, and cognitive load indicators.
• Attention drift: spot tangential or off-topic responses.
• Use Case: after a session with unexpected behavior, generate a prioritized list of hypotheses and a testing plan to verify them.

Quick Start

Provide a session transcript and observed symptoms to generate hypotheses and a structured testing plan.