Run it yourself. Or don't.
Same Tracium.

OpenTelemetry-native traces and metrics on a core you can self-host today, with a managed cloud for when you'd rather not. Own your data either way.

Start free trial Self-host it
Terminal
  • $ export OTEL_EXPORTER_OTLP_ENDPOINT=https://app.tracium.ai
  • $ export OTEL_EXPORTER_OTLP_HEADERS="Authorization=Bearer trc_..."
app.py
  1. # Already exporting OpenTelemetry?
  2. # Point it at Tracium. No code changes.
  3. >_
How it works?
ExportEmit OpenTelemetry from your app
ConnectPoint it at your instance or ours
OptimizeTrack cost, calls, and errors live

Currently observing AI agents' behavior and costs at

Aqtos logo IWM Network logo xo cluster logo it.mk logo

"One line of code gave us real visibility into AI costs, performance, and errors. I might be biased, but it’s a great tool."

Damjan Dano

— Damjan Dano
founder @ aqtos.com

AI observability, simplified.

tracium.ai dashboard

Made for developers

Tracium is an open-source, OpenTelemetry-native observability layer for AI systems. Track costs, trace requests, debug failures, and detect anomalies automatically, across every model, tool, and agent, in one place. Run it on your own infrastructure or let us host it.

tracium.ai dashboard
Costs

Track token spend and infra cost across models, agents, and workflows in real time.

tracium.ai dashboard
Errors

Capture, classify, and replay failures across model calls, tools, and agent steps.

tracium.ai dashboard
Calls

Trace every request end-to-end, including tool hops, agent steps, and model invocations.

tracium.ai dashboard
Per Tenant Analytics

Slice usage, cost, latency, and outcomes by customer, workspace, or environment.

tracium.ai dashboard
Prompt Capture

Inspect the exact prompts and completions behind every call, and switch capture off when the data is sensitive.

tracium.ai dashboard
Anomaly Detection

Automatically surface unusual spikes in cost, latency, and errors before they become incidents.

Transparent pricing for every team

Self-host the open-source core for free, forever. Or choose a managed cloud plan and let us run it for you, with full observability from day one.

Developer

$29 / month

For builders running real AI systems.

  • 50,000 traces / month
  • 5 workspaces
  • 30-day trace retention (metrics kept longer)
  • Alerts (Slack/email)
  • Per-tenant analytics
  • Automatic anomaly detection
  • Team collaboration
  • API access
  • Custom integrations
  • Priority support
Start 14-day free trial

Startup

$99 / month

For small teams scaling their AI systems.

  • 200,000 traces / month
  • 10 workspaces
  • 90-day trace retention (metrics kept longer)
  • Alerts (Slack/email)
  • Per-tenant analytics
  • Automatic anomaly detection
  • Team collaboration
  • API access
  • Custom integrations
  • Priority support
Get started

Scale

$299 / month

For teams running complex production systems.

  • 1,000,000 traces / month
  • Unlimited workspaces
  • 6-month trace retention (metrics kept longer)
  • Alerts (Slack/email)
  • Per-tenant analytics
  • Automatic anomaly detection
  • Team collaboration
  • API access
  • Custom integrations
  • Priority support
Get started

Integrations

tracium integrations tracium integrations

Questions & Answers

If your app already exports OpenTelemetry, there's nothing to change: point your OTLP exporter at Tracium, add your workspace API key, and traces start flowing. Not instrumented yet? Drop in OpenLLMetry with a single init call and you're sending spans in minutes.

Every LLM call, end to end: cost, token usage, latency, and errors, plus full request traces across models, tools, workflows, and agents. Tracium reads standard OpenTelemetry GenAI conventions, so any attribute you attach (team, environment, customer) becomes a dimension you can slice cost and usage by.

Most tools hand you a proprietary SDK to wrap your code. Tracium doesn't wrap OpenTelemetry, it is an OpenTelemetry backend, so there's no vendor agent and no lock-in. It's open-source (Apache-2.0), runs on your own infrastructure or our cloud, and stores traces in ClickHouse to stay fast from the first span to hundreds of millions.

Self-host and your telemetry never leaves your infrastructure. Ingest is key-only: every request needs a per-workspace API key, and unknown or revoked keys are rejected before anything is stored. Prompt and completion capture can be switched off if you don't want that text retained, so sensitive prompts stay under your control.

Yes. Tag spans with a customer or workspace id and Tracium breaks down cost, tokens, latency, and outcomes per tenant, so you can see exactly what each customer costs you and bill accurately.
To top