Trace LOCAL-FIRST AGENT RELIABILITY
Did your agent do it—
or just say it did?
Drop a Codex, Gemini CLI, Browser Use, JSONL, or OpenTelemetry trace. Get a claim-to-evidence graph, failure localization, costs, and a reproducible replay—without uploading the run or asking another LLM to judge it.
01INPUTAgent trace
→
02VERIFYObservable facts
→
03OUTPUTShareable report
No opaque judgeEvery diagnosis cites event IDs.
Content-addressedReplay packages verify with SHA-256.
Benchmark-neutralUnlike scores are never merged.