Show HN: ML condenses billions of logs into a tiny snapshot your LLM can debug
TL;DR
Rocketgraph is an Apache 2.0 project for self-hosted log clustering and anomaly detection. It is designed to sit beside existing observability stacks such as Loki, Datadog, New Relic, CloudWatch, Sentry, or ClickHouse. The ML engine is described as LLM-free: Drain3 mines structural log templates, Isolation Forest scores templates per service, and Half-Space-Trees score new logs in real time.
Nauti's Take
The strongest part is the separation between detection and explanation. Rocketgraph is not pitching a magic LLM that reads raw logs and guesses; it first builds a reproducible compression layer, then lets Claude triage the result.
The weak part: the numbers come from project demos, not independent benchmarks. For teams drowning in logs, it is still worth a practical trial: run it locally on real exported logs and compare what it surfaces against the existing dashboards.
Briefingshow
Most observability tools assume a human already knows what to search for. With AI-written code and huge volumes of near-duplicate logs, that turns incident response into query guessing. Rocketgraph targets that gap by shrinking noisy logs into structural patterns first, then giving an LLM a small, explainable surface to reason over.
Sources