TraceProbe tutorials

Practical guides for debugging AI agents.

Learn how to read execution traces, find the first meaningful divergence, and turn failed runs into focused engineering changes.

Start with the fundamentals

Short, practical guides for developers working with LLM applications and agent workflows.

Move from a failed answer or slow run to a testable engineering change.

Read spans, timing, tool results, and model decisions while keeping sensitive data local.

Understand what structured traces add to an investigation.

Trace tool selection, arguments, returned data, and the first incorrect decision.

Classify prompt, tool, retrieval, and orchestration failures.

Use a healthy baseline to spot regressions after a change.

Learn how to regression test LLM agents with deterministic fixtures that catch tool, retrieval, and output regressions before release.

Open TraceProbe’s sample runs and explore the debugging workflow locally.