Skip to main content
Bring one real agent behavior that was annoying to debug. The best demo input is not a happy path. Bring the run that made you ask: “why did the agent do that?” Kindred uses Langfuse traces as the source of truth. If you already have a Langfuse trace, we can inspect it and talk through how replay would compare the original run against a replay run.

​
Best examples

Bring an agent flow where:
  • the supervisor routed to the wrong agent
  • the agent selected a different tool than expected
  • tool arguments changed between runs
  • a tool returned a different result
  • the agent skipped a step
  • the agent added an unexpected extra step
  • the final answer changed across runs

​
If you already use Langfuse

Bring one relevant trace. It does not need to be perfect. A suspicious trace, a known-good trace, a screenshot, or a failing example is enough. Best case:
  • one trace that behaved correctly
  • one trace that behaved incorrectly
  • the prompt or user flow that produced the issue

​
If you do not use Langfuse yet

No problem. We can walk through setup together.

​
Helpful for live setup

Bring:
  • access to the agent codebase
  • a prompt or user flow that reproduces the issue
  • Langfuse keys, if you already have a Langfuse project
  • a local run command for the agent
  • a local or deployed agent endpoint, if you want to try replay live
  • access to Cursor, Claude Code, Codex, or another coding agent, if you want help patching instrumentation