Show HN: Apgar – Catch agent regressions without an LLM judge

Category: observability

Tags: ai-agents, observability, opentelemetry, release-monitoring, regression-detection

Score: 7.3/10 (Innovation: 7, Technical: 7, Documentation: 8, Utility: 7)

Apgar is a self-hosted release health monitor for AI agents that detects behavioral regressions by analyzing OpenTelemetry traces, without using an LLM judge or storing prompt/completion data. It compares releases on metrics like tool errors, retries, loops, cost, and latency, then sends alerts via Slack or webhook. The approach is privacy-preserving and integrates with existing observability pipelines.

Target audience: AI engineers, backend developers, DevOps, MLOps teams

Repository: https://github.com/oimo-ai/apgar · Python · Apache-2.0 · 2 stars

View on Hacker News