Skip to main content
Short answers with a link to the long one. If you are debugging something right now, Troubleshooting is the faster page.

​
Getting data in

No. There is no aiAxonIQ agent and no aiAxonIQ SDK. You use the upstream OpenTelemetry tooling for your language or platform with two settings changed — which also means nothing here locks you in. See Installation overview.
A Collector, in almost every case. It gathers host and container telemetry your SDK cannot see, holds the credential in one place instead of in every service, batches so you stay inside the rate limit, and gives you one place to filter and redact.Direct export is reasonable for a single service or a local experiment.
Yes, in both directions. Remote-write pushes its metrics to aiAxonIQ with a config change on one side, and a Prometheus-compatible query API lets Grafana read aiAxonIQ back. See Prometheus remote-write and Grafana.
eBPF instrumentation observes it from the kernel — no code change, no restart. It is early access; see Zero-code instrumentation.
HTTP unless you have a reason not to. It survives proxies and TLS-terminating load balancers that gRPC does not, and you can reproduce any failure with one curl. See Exporters.

​
My data is missing

202 means accepted, not stored. A batching consumer flushes at 50,000 rows or five seconds — on a quiet account you are waiting out the five seconds.Then, in order of how often each is the answer: the time range (narrow to 15 minutes), a hardcoded timestamp copied from an example, or the wrong organization. See Verify your data arrived.
service.name is not set, and its default is unknown_service. Every service that omits it collapses into one entry. Set OTEL_SERVICE_NAME — see Resource attributes.
A fresh record is in the raw table before the rollups covering it are complete. Wait a minute; do not widen the range. See Data retention.
Either context propagation broke — the missing service appears as its own root trace — or sampling is inconsistent, and it disappears entirely. See Traces with a hole in them.
The logger is not trace-aware. A print() or a bare file logger produces correct, searchable lines that are permanently disconnected from every trace. Use your language’s OpenTelemetry logging bridge.

​
Retention and cost

Logs, traces and full-resolution metrics for 30 days. Minute-resolution metric rollups for 90 days, hour-resolution for 365 days. See Data retention.
Volume allowances do not stop your ingest. Telemetry keeps arriving and stays queryable; overage is handled commercially. Which also means you will not be warned by a failure — watch Settings → Usage.Seats are the exception and are enforced. See Plan limits.
In order of return: drop health-check and metrics-scrape traffic at the Collector, tail sample traces, raise log levels in noisy services, and check no metric label carries a user or request id. The first usually beats all the others combined.
There is no self-service per-record delete. Contact support for a credential committed to a log or a mistaken import.The durable fix is redaction at the Collector, before it leaves your network — see Collector configuration.

​
Access and organizations

No. Every member of an organization sees all of its telemetry. Roles control what you can do, not what you can see.If some telemetry must be invisible to some people, the boundary you need is a separate organization. See Organizations and multi-tenancy.
A deployment.environment resource attribute, in one organization, is what most teams want. Two organizations is the answer only when the requirement is access control.
No. Your identity provider asserts identity only. A user with no existing membership is refused 403, however well their authentication worked. Invite first, then have them sign in. See Single sign-on.
Support can be granted time-limited access to investigate a ticket, and every such session is recorded in your audit log — who, when and for how long. See Audit log.
Key creation requires Admin. Ask your organization’s Owner or an Admin. See Roles and permissions.

​
Alerting

Most often an unimplemented comparator. Only GT, LT and EQ exist — a rule saved with GTE, LTE or NE is created without complaint and never fires, with no error anywhere.Second most often: an anomaly or synthetic rule created as an ordinary rule. Those are fired by different components, not by the evaluation loop.Use Test after saving any rule. See Alerting.
Rules evaluate on a fixed 60-second loop, which is the floor on detection latency. Setting a shorter evaluation window does not make it check more often.
Generic outbound webhooks are disabled by default, pending request signing and protection against server-side request forgery. Slack, PagerDuty, Teams, OpsGenie and email all work. Your operator can enable it.

​
About the product

Not by design, and it is testable. Everything speaks standard OTLP — point the same exporter at a different backend and it works. You can even export to both during an evaluation. See Exporters.
Self-hosted deployments exist and are arranged with your account contact. There is no public download or self-service installer today, and the operational material ships with the deployment rather than being published here. See Onboarding: self-hosted.
During the private beta, by invitation. See Create your account.

​
Next

Troubleshooting

The pipeline stage by stage.

Glossary

Terms used throughout, defined once.

Choose an onboarding path

Six end-to-end journeys.