01 — customers
Three teams, three different bugs
Each of these started with a failure that produced no error. That is the category of problem this product exists for.
Also running Cinder
- Verity Labs
- Northbeam
- Lumen Freight
- Tessera
- Ardent
- Quillwork
99.98% ingestp99142 ms
- Verity LabsLegal technology34 engineers
Found a four-percent silent failure that had been shipping for six weeks
A retrieval step returned empty results without erroring. Verity's contract-review agent kept answering, slightly worse, and nothing alerted.
- of requests silently degraded
- 4.1%of requests silently degraded
- from install to root cause
- 1 afternoonfrom install to root cause
- the bug had been live
- 6 weeksthe bug had been live
Read the story
- Northbeam HealthHealthcare operations52 engineers
Per-customer cost attribution without sending us a single prompt
Northbeam operates under a data agreement that keeps patient-adjacent text inside their VPC. They needed cost and latency attribution anyway, and got it from span structure alone.
- of prompt content leaves their VPC
- 0 payloadsof prompt content leaves their VPC
- inference spend reduced in one quarter
- 31%inference spend reduced in one quarter
- to replace a manual billing spreadsheet
- 2 daysto replace a manual billing spreadsheet
Read the story
- Lumen FreightLogistics19 engineers
Cut agent-incident acknowledgement from 40 minutes to under four
Lumen's on-call rotation had no way to tell an upstream carrier outage from an agent regression. Both looked like "the bot is wrong". One trace view ended the guessing.
- median time to acknowledge
- 40min → 4minmedian time to acknowledge
- dashboards retired
- 3dashboards retired
- new tools added to the rotation
- 0new tools added to the rotation
Read the story
Your incident would fit here.
Start on the free tier and find out what your traces have been hiding.
npx @cinder/cli initOr talk to an engineer — the same people who wrote the ingest path.