Blog

Six metrics for judging AI agent traces in production: tool failures, hallucination, logic errors, task completion, safety, and a custom metric you define.

traceroot and Langfuse both trace your agent. traceroot is agent-native: a CLI your coding agent can drive, plus Signals and root cause analysis to improve it.

A LangGraph text-to-SQL agent explained a $200 MRR drop with the wrong reason. We trace the answer back to the reporting view it read in traceroot and show what to record.

An answer that crosses no alert threshold. A traceroot Detector, an LLM-as-a-judge check written in plain English, catches it in production and opens a Signal.

Root cause analysis for a LangGraph SQL agent's wrong answer: the Signal points at a join on subscription_id. We verify it, change both joins and rerun the agent.

An agent evaluation tutorial in traceroot: both runs report the same total, and four code checks on one saved case tell the faulty and corrected runs apart.

A tool call returns 200, the answer is still wrong, and nothing pages anyone. Here is what runs when traceroot judges a trace — and the three decisions behind it.

pi is an open-source agent runtime. Here is how to trace its sessions, turns and tool calls, and how we run our own production agents on pi.

What 1,085 Reddit comments reveal about finding failed agent runs, reconstructing context, and turning production mistakes into regression tests.

The agent in the traceroot app reads your traces and your code at the production commit, and comes back with a change instead of a chart.

Connect TypeSafe AI's Jev to traceroot detectors to screen agent traces and inspect the decisions behind each finding.

The traceroot CLI puts your traces where the work already happens: in the coding agent's terminal, not a browser tab.

A practical introduction to the open-source agent with four tools and one loop.