Service insights
Continuous LLM-as-judge against real-world traces from agents using your service through the AXRank MCP. Synthetic evals tell you how the surface reads; field telemetry tells you how it's actually being used. Insights blends the two.
Sign-in required
Service Insights dashboard
The dashboard is in build. It surfaces the failure-mode mix on your surface from the last N agent sessions, the diff against the previous synthetic eval, and the single-highest-leverage fix this week.
What you'll see
- —Field failure mix. Counts and ratios for each of the 18 failure-mode categories across real agent traces in the last 7 / 30 / 90 days.
- —Synthetic-vs-field gap. Where the controlled eval and real usage disagree — the leading indicator of which findings actually bite in production.
- —Highest-leverage fix. The one change with the best expected score lift, sourced from the classified failures.
- —Agent Relations feed. A continuous timeline of surface changes, score shifts, and the evidence backing each call.
Heads up
Evals aren't cheap — a full run can cost a few hundred dollars in tokens, and continuous insights means rerunning the loop on a cadence. AXRank is paid for out of pocket; sponsorship keeps it independent and never influences score.
How sponsorship works →