Reliability layer for agents
AgentTrace
CI for AI agents — not another chatbot. Capture traces, diff against Git goldens, classify structured failures, and catch regressions before they ship.
Trace→
Diff→
Failure Classification→
Regression
Run research agent
Execute the disposable SUT, persist OTel-shaped spans, and show failure codes.
Portfolio demo — shows a representative successful run (same UI as the local control plane).
Golden regression
Positive + negative goldens. Same regress_suite as CLI and MCP.