Skip to main content
aiAxonIQ exposes a Prometheus-compatible query API, so Grafana and other Prometheus clients can read your metrics directly. There is no exporter, bridge or adapter to run. This is the answer to “we already have Grafana dashboards and we are not rebuilding them.”
Before you start. You need metrics in aiAxonIQ and a bearer token for the API. See Authentication.

​
Supported endpoints

POST is accepted on /api/v1/query and /api/v1/query_range as well, which is what Grafana uses for a long expression. That is the subset Grafana needs to drive a dashboard and populate its query builder’s dropdowns.
These are the only routes that do not need a tenantId. Everywhere else in the API a tenant-scoped request without one is a 401 — but Prometheus has no concept of a tenant and Grafana cannot be made to send one, so these routes take the organization from your session alone. See API reference.

​
Configure the datasource

1

Add a Prometheus datasource in Grafana

Set the URL to your aiAxonIQ API origin. It is the same host you sign in to.
2

Add the credential as a custom HTTP header

Under Custom HTTP Headers, add:
Authentication is the standard bearer token used by the rest of the API — not a license key. License keys are for sending data; sessions are for reading it. See Authentication.
3

Save and test

Then build a panel with a metric name and confirm data appears.

​
The limits — read this before relying on it

This is a simplified PromQL subset, not PromQL. The supported shapes are:
Aggregations avg, sum, max, min and count are supported.Rate functions are accepted and approximated by an average. rate(m[5m]), increase(m[5m]) and irate(m[5m]) all parse, run, and return the mean sample value over each step rather than a per-second rate — a plausible number that is not a rate.This is the single most important thing on this page. A copied Grafana dashboard full of rate() expressions will render and will not error. The response carries a warnings entry saying the function was approximated, and Grafana surfaces it beside the panel — but the line is drawn either way, so verify every migrated panel against the same data in aiAxonIQ before trusting it.
Everything else the subset cannot express is refused with a 400 and a message naming what was unsupported, rather than answered with something plausible:
  • No nested expressions. sum(rate(m[5m])) is a 400. It used to be read as a metric name spelled sum(rate(m[5m])), match nothing, and return an empty series — a blank panel with no error on it.
  • No binary or set operators between series — no dividing one metric by another to compute a ratio, and no and / or / unless.
  • by is supported, one label at a time; without, on, ignoring, group_left and group_right are not. See Grouping below.
  • No sub-queries, no histogram_quantile, no recording rules.
  • Only equality label matchers. !=, =~ and !~ are refused.
Five labels can be matched, and nothing else. service_name, service_version, host_name, environment and metric_type — the same five GET /api/v1/labels advertises. Any other matcher — {pod="…"}, {instance="…"} — returns a 400 naming the labels that are available.A matcher outside that set is refused rather than dropped. It was previously accepted and then ignored, which returned data for every value of that label under a heading that named one — a wrong answer rather than a missing one.An empty matcher value is the one exception, and it is reported rather than applied: {service_name=""} covers every service, and the response carries a warnings entry saying so. Prometheus would instead select only series that carry no such label. This keeps a Grafana panel written against an unset dashboard variable populated rather than blanking it on the next refresh.

​
Grouping

by groups the result by one of the five matchable labels, in either the prefix or the suffix spelling:
Two things differ from Prometheus and both are deliberate:
  • The default grouping is service_name, not “collapse to one series”. In PromQL sum(x) returns a single series; here it returns one per service, and has since this datasource existed. Every panel built against it is drawn that way, so changing the default would silently redraw all of them.
  • To get PromQL’s collapse, name no label: sum by () (metric_name) returns exactly one series for the whole account, published with no grouping label.
One label at a time. sum by (service_name, host_name) (x) is a 400; without is too, because it needs a series’ full label set, which a column-shaped store does not have.
You do not need Grafana to write one. The product has a PromQL console at Metrics → Query, which runs against the same endpoints with the same subset, lists what is and is not supported beside the box, and completes metric names as you type.

​
Metric names

An underscored Prometheus-style name also matches the dotted OpenTelemetry spelling of the same metric, so an existing Prometheus dashboard finds data that arrived over OTLP:
The mapping runs one way. Underscores are expanded to dots; dots are not contracted to underscores. A query written as http.requests.total matches only that exact spelling, so where you have a choice, write the underscored form.

​
When to use this, and when not to

A mixed setup is entirely reasonable. Keep your Grafana dashboards where they are, and use aiAxonIQ for the investigation surfaces — logs, traces, the service map and the AI features — that Grafana was never going to serve from a Prometheus datasource anyway.

​
Other Prometheus clients

Anything speaking the Prometheus HTTP API can read from these endpoints under the same limits. The rate-function caveat applies equally — that is the one construct that returns an answer instead of an error, so it is the one a client assuming full PromQL will be misled by.

​
The other direction

To send metrics from an existing Prometheus server into aiAxonIQ, use remote write — a configuration change on one side and nothing on your services. See Prometheus remote-write.

​
Next

Prometheus remote-write

Getting metrics in.

Exploring metrics

The native metrics surface.

API reference

The full API, including these endpoints.