Soft launching TraceFlowLens: record, replay, and compare AI agent executions 🚀
Hi everyone,
I'm soft launching TraceFlowLens.
AI agent debugging is becoming a different problem from traditional software debugging.
The same prompt can produce a different outcome because of model updates, retrieved context, tool responses, prompt changes, or external dependencies.
TraceFlowLens helps engineering teams:
• Record AI agent executions
• Replay executions from recordings
• Compare executions side by side to pinpoint exactly where behavior diverged
• Verify fixes with replayed evidence
It's local-first by default, works with existing AI applications without framework migration, and currently supports the OpenAI SDK, Anthropic SDK, and custom AI workflows.
I'm looking for feedback from engineers building AI agents or production LLM applications.
Website:
I'd love to hear:
- How are you debugging AI agents today?
- What's the biggest challenge you've run into?
