The problem

Most LLM-powered applications with real use cases are more than just an LLM call:

A retrieval step, some external API calls, maybe a re-ranking step and a chain of prompts.

If you're building a chat app, you might have multiple steps for each message. This can make it really hard to have visibility into the entire system.

That's why we're really excited to introduce:

What we built

**A simple system for logging complex traces (**See an interactive demo here)

We've extended our logging framework, to support prompt chains and complex traces.

This means you can now log:

• Retrievals
• Prompt chains
• Conversations / chats
• External API calls
• Custom metrics (user clicked, user rejected, thumbs down, dwell time)
• • anything else you can think of.