Start here
Try the playground
A workspace that already has a fortnight of telemetry in it — no account, no
install, no card. Three demonstration estates, each a different kind of
system with a different thing going wrong.
Quickstart
From an account to data on screen in about ten minutes, using nothing but
curl. The shortest path that proves every hop.Choose an onboarding path
Six end-to-end journeys — SaaS, Docker, Kubernetes, developer, enterprise,
self-hosted — with what to do on day two, not just day one.
How it works
The path your data takes, and the two properties of it that explain most of
what you will run into later.
Verify your data arrived
Telling “accepted” apart from “queryable” — the distinction behind most
first-install confusion.
Product tour
Every section of the dashboard, what it shows, and where its guide is.
Your first hour
Data is flowing — now a dashboard, an alert, an incident and a first
investigation.
Send data from where you run
Kubernetes
A DaemonSet collector for node telemetry, and the second collector most
clusters end up needing.
Any application
Auto-instrumentation for Node.js, Python, Java and .NET; an explicit setup
for Go.
Prometheus
Remote-write from an existing Prometheus server. A config change on one
side and nothing on your services.
Services you cannot instrument
Zero-code instrumentation with eBPF
Traces and RED metrics from services you will never add an SDK to — vendor
images, legacy containers, anything nobody owns. One agent per host, observing
from the kernel: no code change, no rebuild, no restart.
AgentSight: Kernel AI Agent Monitoring
Observe what autonomous AI agents actually do on your hosts — spawned
subprocesses, socket connections, network calls and file access — directly
from eBPF.
Then use it
Search logs
Field filters, booleans, wildcards and numeric comparisons, with the full
field list.
Explore traces
Where the time went, which hop failed, and the two reasons a trace has a
hole in it.
Service map
The dependency graph, discovered from traffic rather than maintained by
hand.
Alerting
Rule types that actually evaluate, the three comparators, and which
channels work out of the box.
Service level objectives
Error budgets, burn-rate alerting, and gating a deploy on remaining budget.
Synthetic monitoring
HTTP, SSL, DNS, TCP and multi-step checks run against your endpoints from
outside.
APM
Requests, latency, errors and Apdex per service, grouped into projects.
Incidents
What an alert becomes, and how to acknowledge, assign and resolve it.
AI Assistant
Investigate, ask your data and root-cause analysis — and what each needs.
LLM & GenAI Observability
Token counts, model latency, cost attribution, prompt evaluations and
error rates across OpenAI, Anthropic, Bedrock and self-hosted models.
AI Features & Governance
Anomaly detection, capacity forecasting, natural-language query and agent
governance with strict tenant isolation.
Understand it
Platform architecture
Why there are four stores, and which one answers each question you ask.
Organizations
Multi-tenancy, isolation, and where the boundary genuinely is.
Data retention
One window per plan, 30 days by default, applied to rollups as well.
Security overview
Credentials, roles, tenant isolation and what happens to your data.
Projects, services and environments
How telemetry is grouped, and which grouping is a boundary.
Integrations
Slack, PagerDuty, Opsgenie, Teams and email channels, set up once.
The three things you need
1
A license key
Created in Settings → API keys, sent as the
X-License-Key header.
See Create a license key.2
An endpoint
The receiver’s OTLP address, which differs per deployment. Get Started
in the dashboard shows the one that is correct for yours.
3
An exporter
Any OpenTelemetry SDK, or the OpenTelemetry Collector. aiAxonIQ speaks
standard OTLP, so there is nothing vendor-specific to install and nothing
to rewrite if you later point the same exporter elsewhere.
Not sure where a problem is?
Troubleshooting splits the pipeline into four
stages and tells you which one you are in — usually in about a minute, which
eliminates most of the list before you start reading. The
FAQ is the faster page when you are not debugging.
Build on it
API reference
Everything the product does, programmatically — generated from the API’s
own route table and checked against it on every build.
Grafana and Prometheus clients
Use aiAxonIQ as a Prometheus datasource with no adapter — and the PromQL
limits to know before you rely on it.