Hanzo Metrics
High-cardinality time-series database for AI metrics
A time-series store built for the cardinality explosion of modern AI: per-prompt, per-user, per-model labels at millions of series without falling over. PromQL-compatible.
Built for AI Cardinality
Where Prometheus chokes, Hanzo Metrics keeps going.
Millions of Series
Per-prompt, per-user, per-model labels. Active series counts that crush traditional TSDBs run cleanly here.
PromQL Compatible
Drop-in for existing dashboards and alerts. Your Grafana panels, Hanzo Insights queries, and recording rules just work.
Fast Ingest
Single-node ingest north of one million samples per second. Horizontal sharding when one node is not enough.
Tiered Storage
Hot in memory and SSD, warm on object storage. Months of retention without a hardware bill that hurts.
Histograms & Summaries
Native sparse histograms for latency and token-cost distributions. Quantiles without the merge bug.
Long Retention
Years of metrics with downsampling rules. Capacity planning beats throwing data away.