Project profile
jevmetrics
An alpha OpenTelemetry Collector processor that asks hosted Jev whether a metric is worth keeping, then applies your thresholds.
Last verified: 2026-09-30
Quick facts
| Project | jevmetrics |
|---|---|
| Creator | ishantanu |
| Type | Developer Tool |
| Language | Go |
| License | Apache-2.0 |
| Uses Jev for | Scores metric metadata; Collector policy keeps or drops |
| Open source | Yes |
| Status | Experimental |
What it is
jevmetrics sits in a metrics pipeline. It sends metric metadata to POST /v1/systemone, caches the answers, and can either annotate the stream or drop metrics under a keep-score policy. The README marks the project alpha.
Why it matters
Retention becomes a typed decision plus a policy file, which is a different product surface from routing a support ticket.
Relationship to Jev
| Hosted Jev | Yes. Default model in the README is jev-latest, at api.typesafe.ai. |
|---|---|
| Questions | Four per metric: relevance, redundancy, and keep as noul; action as a choice of keep, reduce, or drop. |
| What filters | Only the keep noul, after protection rules. The choice is exported for review. |
| Modes | annotate preserves metrics and emits assessments. reduce applies the keep-score policy. |
Architecture
- OTLP metrics
- Metadata summary
- Jev
- Cache
- Annotate or reduce
Key features
- Protected metric names and prefixes skip inference.
- Assessments are cached so later batches can reuse a decision.
- The Docker demo uses mock inference and does not need an API key.
Performance and claims
The README says the processor infers operational value from names, descriptions, units, scope, and attribute keys, before there is a history of dashboard use.
Jev AI Hub read the README and the repository record on 2026-09-30. The repository had 4 stars and an empty GitHub description at capture.
Start in annotate mode, as the README recommends, and decide the keep threshold from the assessments you actually see. The model does not know which dashboards your team opens.
Limitations
- Alpha. The README says filtering savings need evaluation on your own telemetry.
- Inference is asynchronous: the first batch is kept and queued, and the assessment applies to later batches.
- This hub has not run the Collector demo.
Code and repository
https://github.com/ishantanu/jevmetrics
GitHub snapshot captured 2026-09-30: 4 stars, 0 forks, 2 open issues. Last push 2026-09-29. Counts change; this is not a live lookup.
Related Jev concepts
NoulComposite scoringMetric retention
Related guides
Related projects
Sources
- jevmetrics READMEishantanu · accessed 2026-09-30 · github
- ishantanu/jevmetrics repository metadataGitHub · accessed 2026-09-30 · third-party