Skip to main content
New: LLM Observability is now GA

Full-stack observability.
Open by design.

Monitor metrics, traces, and logs across your entire stack. Built on OpenTelemetry and Prometheus. No vendor lock-in, ever.

Explore a live demo instantly — no signup · Or start free for up to 3 services
app.aiaxoniq.com/overview
Live
J
aiAxonIQ
Overview
Services
Metrics
Traces
Logs
Alerts
Settings

Production Overview

Last 24 hours · All services

24h
↻ Live
Requests/s
42.3k
+12% vs last hour
Error Rate
0.12%
-3% vs last hour
P99 Latency
145ms
+2ms vs last hour
Request Rate
api-gatewayauth-service
Error Distribution
Service Health
api-gateway
auth-service
payment-svc
Built in the open — on standards and infrastructure you already trust
OpenTelemetryClickHouseApache KafkaPrometheusOpenSearchPostgreSQLRedisQdrantDockerKubernetesOpenLLMetrygRPCOpenTelemetryClickHouseApache KafkaPrometheusOpenSearchPostgreSQLRedisQdrantDockerKubernetesOpenLLMetrygRPC

The problem

The bill keeps growing.
The outage still takes all night.

Most teams did not choose a fragmented stack. It accumulated — a tool per signal, a vendor per contract, an agent per host — and every one of them is now load-bearing.

Every incident starts with a tab hunt

Metrics in one tool, traces in another, logs in a third. Correlating them is manual, it happens at 3am, and it is the part of an outage that takes the longest.

The bill scales faster than the traffic

Per-host and per-seat pricing turns growth into a procurement conversation, and proprietary agents mean the instrumentation you paid to write does not leave with you.

AI spend is invisible until the invoice

Model calls are the fastest-growing line item in most engineering budgets and the least instrumented. A runaway prompt is discovered by finance, not by monitoring.

IngestOTLP + Prometheus

HTTP, gRPC and remote-write. Instrument with the open SDK you would have used anyway.

TransportApache Kafka

One durable, replayable pipeline for every signal, with dead-letter queues.

StoreClickHouse + OpenSearch

Columnar storage with materialized-view rollups; BM25 full-text search over logs.

Run itCloud or your cluster

The same stack, self-hosted with Docker Compose or Kubernetes whenever you want it.

The answer

One pipeline. Every signal. No black boxes.

Logs, metrics, traces and LLM calls travel the same durable path into the same store, so correlating them is a query rather than a project. Every layer is a technology you can read about, benchmark, and run yourself.

Correlation is built in

A trace links to its own logs and metrics because they arrived together, not because someone configured a join.

Leaving costs you nothing

Open SDKs in, open formats out. The instrumentation you write here keeps working anywhere it is pointed.

The product

Three jobs, done properly

Tracing, log search and AI spend are where platforms are usually slow, expensive or missing — and under them, eleven surfaces reading the same data.

Explore the full platform

Keep every trace, not a sample

Store spans in ClickHouse columnar storage with materialized-view rollups, so dashboards stay fast over high-volume telemetry — and the trace you need at 3am is still there, fully correlated with its logs.

Explore APM & Tracing

Search logs like a search engine

BM25 full-text search backed by OpenSearch finds the needle, while automatic log pattern clustering groups the haystack — recurring errors collapse into patterns instead of pages of noise.

Explore Log Management

See what your LLM spend buys

Track token cost, latency, and a per-model breakdown for every AI call — OpenLLMetry-compatible, so a runaway prompt shows up in your dashboards before it shows up on your invoice.

Explore LLM Insights

Real-time metric pipelines

Ship metrics over OTLP or Prometheus remote-write into ClickHouse columnar storage. Materialized-view rollups keep aggregations fast over high-volume telemetry.

Learn more
  • Prometheus remote_write compatible ingestion
  • ClickHouse storage with 1-minute and 1-hour rollups
  • Multi-dimensional labels and aggregations
  • Kafka-backed pipeline: durable and replayable
CPU Utilization
All nodes · Last 1 hour
node-01node-02
Avg
62%
Max
90%
P95
85%

AI that shows its working

It tells you why, not just that

Detection is the easy half. The expensive part of an incident is the twenty minutes spent deciding which of forty graphs mattered — so that is the part the models work on.

Root cause, ranked and evidenced

Signals that moved together at onset are correlated into candidate causes, ordered by confidence — each one linked back to the traces, logs and metrics it came from. Never a single unexplained verdict.

Anomalies found statistically

Z-score and IQR detectors over your own baselines, plus Prophet forecasting for capacity planning. Methods you can name, argue with, and tune.

Ask in English, get SQL

Natural-language querying compiles to a query you can read before it runs, and retrieval over your own runbooks suggests the next step rather than inventing one.

Root cause analysis
INC-4182 · checkout error rate · 3 signals correlated
Analysed 12s ago
Connection pool exhaustion — payments-db86%
pool_wait_ms p99 ×147 CrashLoopBackOffdeploy 4m before onset
Upstream provider latency41%
charge_latency p95 ×2.1
Node memory pressure18%
ip-10-0-3-08 at 88%

Ranked hypotheses with the signals behind each. Every candidate links back to the traces, logs and metrics it was derived from.

Integrations

It already works with your stack

If it speaks OpenTelemetry or Prometheus remote-write, it works — every major language, framework and cloud, with no proprietary agent to install.

Explore all integrations
OpenTelemetry
Prometheus
Grafana Agent
Kubernetes
Docker
AWS
GCP
Node.js
Python
Java
Go
.NET
PostgreSQL
Redis
Kafka

Tenant isolation at ingest

Separation is enforced where data arrives, not filtered at query time. Per-tenant rate limits included.

RBAC with four roles

Owner, admin, editor and viewer, with team invites. SSO and SAML on the Enterprise plan.

Immutable audit log

Every tenant action recorded append-only, so a reliability or security review has something to read.

Runs in your own cluster

The same stack via Docker Compose or Kubernetes. Data residency becomes a deployment choice, not a negotiation.

For the enterprise

Built multi-tenant. Not retrofitted.

Isolation, roles and audit trails are properties of the architecture rather than a tier that unlocks them. The controls a security review asks about are the same ones the free plan runs on.

Certifications, sub-processors and the current security posture are documented rather than badged.

For developers

Your first trace, in about two minutes

Open standards all the way down. Instrument with the OpenTelemetry SDK you would use anyway, point it at aiAxonIQ, and keep the instrumentation if you ever leave.

Install

npm install @opentelemetry/api @opentelemetry/auto-instrumentations-node

Configure

export OTEL_EXPORTER_OTLP_ENDPOINT="https://app.aiaxoniq.com/otlp"
export OTEL_EXPORTER_OTLP_HEADERS="x-license-key=YOUR_LICENSE_KEY"
export OTEL_SERVICE_NAME="my-service"

Run

node --require @opentelemetry/auto-instrumentations-node/register your-app.js

It is --require rather than an import at the top of your entry file: the hook has to install instrumentation before your application modules resolve. An import runs after the modules it needs to patch are already loaded, and produces an app that starts cleanly and emits nothing.

Full instrumentation guides, ingest endpoints and troubleshooting live in the SDK documentation.

Pricing

Priced on data, not on headcount

No per-seat fees and no per-host fees, so adding an engineer or a node does not reopen the contract. Start free, self-host at any point.

Free

$0/mo

3 services, 7-day retention, full OpenTelemetry ingest.

Get started

Growth

Most Popular
$49/mo

Unlimited services, 30-day retention, AI anomaly detection.

Get started

Enterprise

Custom

SSO/SAML, RBAC, audit logs, dedicated cluster, custom SLAs.

Contact sales