Show HN: ML condenses billions of logs into a tiny snapshot your LLM can debug
TL;DR
Rocketgraph is a self-hosted open-source observability tool: it clusters raw logs into structural templates and flags anomalous patterns without adding a separate ingest pipeline. It uses Drain3, Isolation Forest and Half-Space-Trees. LLMs are optional for triage explanations, while the detection path is deterministic and reproducible.
Nauti's Take
The strong part is the compression step: templates and anomaly scores first, an LLM as a debugging helper after that. That is the sane way to bring AI into observability, because raw logs are expensive, noisy and often sensitive.
Be careful with the HN framing: billions sounds impressive, while the public repo proves a smaller benchmark. Teams should judge it on false positives, missed incidents and whether it surfaces new patterns before the dashboard does.
Briefingshow
As AI writes more code, teams will see more small and unfamiliar production failures. Traditional dashboards often show only the metrics someone already expected and wired into an alert. Rocketgraph shifts attention to new or rare log patterns.
That can give SREs and LLMs a smaller, more explainable case file instead of a pile of raw logs.
Sources