What a metric is here
A row in a column store rather than a series in a store of its own.
metrics.hanzo.ai
rate(http_requests_total[5m])- zen
- kai
- enso
a spike is the start of a question
- gateway
- chat.completion
- datastore.query
- model.generate
One store, not four
01 / 06
Metrics go where the logs and traces already are — Hanzo Datastore, a column store. Nothing to run beside it and nothing to keep in sync with it.
PromQL, or the builder
02 / 06
Bring the PromQL you already wrote, or assemble the query in the UI without learning it. Both compile to the same thing.
Labels you would not dare add
03 / 06
Per-model, per-route, per-tenant. A column store counts distinct values for a living, so a label that would multiply series in a time-series database is just another column here.
- In a time-series database
- series = models
- Here
- 1 more column
The number leads somewhere
04 / 06
A spike on a chart is the start of a question, not the end. Click it and you are in the traces for that window, because they were never in a different system.
Alerts are saved queries
05 / 06
You write the query, look at it, then keep it as a rule. What pages you is something you have already seen return the answer you expected.
Scoped to your org
06 / 06
Every query carries the org from your token, so one deployment serves many tenants and no panel can read across them.
rate(http_requests_total[5m])onequerykindnamevaluemodelroutetenant
metrichttp_requests_totalzen/v1/chatorg a
tracechat.completionzen/v1/chatorg a
logrequest servedzen/v1/chatorg a
metrichttp_requests_totalkai/v1/embedorg b
traceembedkai/v1/embedorg b
metrichttp_requests_totalenso/v1/searchorg c
logrequest servedenso/v1/searchorg c
Send a metric, then follow it
Paste and ship
Paste this into any agent. It reads the skill manifest and calls Metrics from there.
Read https://hanzo.ai/skill.md and use Hanzo Metrics in my project. Start with: hanzo metrics queryThe service the session above is changing.
A build is running from the last commit on main.
Static, published by the same pipeline as the rest.
traces for that window · gateway
- gateway
- chat.completion
- datastore.query
- model.generate
api.hanzo.ai/v1/event/insightsOpen in the console ↗
Example project. Every name in it is one you would give your own.