let Jev score OpenTelemetry logs before a bigger LLM sees them
Jev Logs is an MIT-licensed, open-source TypeScript triage layer that uses TypeSafe's Jev via the Vercel AI SDK to score OpenTelemetry logs before expensive LLM analysis, assigning a 0 - 100 diagnostic value, priority (critical/high/normal/low), actionable probability, and a route (analyze/retain). It skips deeper analysis only when priority is low, value ≤25, and actionable probability <0.1, and integrates by wrapping an existing OTLP exporter (Node.js 22+), with annotation mode preserving all records and rule-based filters (e.g., `^GET /health` → retain) to avoid Jev calls. Run `npx jevlogs` for an offline demo or `--live` with a local OTLP receiver on `127.0.0.1:4318` to cut LLM spend by sending only high-signal logs to reasoning models.
- why now
- New open-source log triage.
- topic
- AI Tools & Agent Workflows
- source
- DEV Community