Answer Reasoning (Traces)
Ever wonder how your AI agent arrives at its answers? The Answer Reasoning feature lets you peek behind the curtain to see the agent's thought process in real-time. Reasoning begins streaming back long before the answer itself is returned from the AI.
What you get
When enabled, users can see:
- How the agent thinks - Step-by-step reasoning as it works through problems
- When things go wrong - Clear explanations if the agent can't answer something
- Tool usage - Which external tools the agent used and why
- Quality checks - How confident the agent is in its response
Who can see traces?
You control who gets access to reasoning traces:
- Basic traces - Show general thought process (good for most users)
- Detailed traces - Include technical parameters (typically for admins only)
Traces help you understand why an agent gave a particular answer, making it easier to improve prompts and troubleshoot issues.
You can enable traces for specific user roles (like content admins) while keeping them hidden from regular users.
The Answer Reasoning (Traces) feature must be enabled in your site-wide pika-config.ts file before chat apps can use it. Without this enablement, traces will not be available regardless of individual chat app settings.