Blog

Articles on AI agent debugging, testing, and observability.

April 13, 2026

Introducing Rewind: Chrome DevTools for AI Agents

AI agents ship with 30-step chains, but debugging them means re-running everything from scratch. Rewind records every LLM call, lets you fork at any failure, replay with the fix, and prove it works. Open source, MIT licensed.

announcement open-source ai-agents debugging

April 13, 2026

I use Langfuse for tracing. Here's why I added Rewind for debugging.

Langfuse shows you what your agent did. But when step 15 of a 30-step run hallucinates, you can't fix it without re-running everything. Rewind lets you fork at the failure, replay with the fix, and prove it works.

langfuse debugging ai-agents observability