See how it thinks.

Reasoning Display

Models that support reasoning (gemma4, deepseek-r1, qwen3) stream their thinking tokens in real-time. You see exactly how the AI classifies your request, how many tokens it uses, how long it takes, and the full reasoning chain. No black box. Complete transparency into every response.

See it in action

How it works

1. 1

   Each request is classified (SIMPLE, MODERATE, COMPLEX, ANALYTICAL) to set appropriate response parameters.

2. 2

   Token counts are displayed: estimated input tokens and actual output tokens, with per-model calibration.

3. 3

   Reasoning models emit <think> tokens that are streamed to you in real-time as the model works through its logic.

4. 4

   Valence tracking shows the AI's confidence level, which changes based on task success/failure history.

Key capabilities

Use cases

• Understand why the AI made a particular decision
• Debug unexpected responses by reviewing reasoning
• Compare reasoning quality across different models
• Monitor response efficiency (tokens per quality)

Commands