What is Orbtrace?
Read this first if you've never used an observability tool. It explains, in plain language, the problem Orbtrace solves and what you'll see in the UI.
Orbtrace is a self-hosted observability platform. It collects your logs, traces, and metrics in one place — on your own servers, never a SaaS — and uses an AI model you choose to tell you, in plain English, the root cause when something breaks. If you've used Datadog, Grafana, or SigNoz, it's that, running entirely on your hardware.
The problem in one paragraph
Modern software is made of dozens of small programs that talk to each other over the network. When something breaks at 3 AM, the question is never "is it broken?" — the dashboards already lit up red. The real question is "why is it broken, and which of the forty services is the cause?" Answering that question used to mean opening five tabs (logs in one tool, traces in another, metrics in a third, deploy history in a fourth, the on-call chat in the fifth) and stitching them together in your head while half-asleep.
Orbtrace puts all of that in one tool, on your own servers, and tells you the cause.
What does Orbtrace actually do?
You install Orbtrace on a server (or a small cluster). Your applications send three kinds of data into it:
- LogsSingle lines of text your app prints when something happens — like a diary entry. Example:
User 42 placed an order for $19.99. - TracesA map of one user request as it travels through your services. If a checkout request hits the API, then the database, then the payment provider — that's one trace with three "spans".
- MetricsNumbers measured over time, like CPU usage, request count, or queue depth.
Orbtrace stores all three together, lets you search them in one query language, draws charts on top of them, and — this is the part that competitors don't have — uses an AI model you choose (Claude, OpenAI, Azure OpenAI, or a local Ollama) to read the data during an incident and tell you, in plain English, what caused the problem and which span proves it.
Who is this for?
- DevOps / SRE / Platform engineers who run the production stack and get paged when it breaks.
- Backend developers who want to know why their service is slow before the SRE pages them.
- Engineering managers who need to answer "why was checkout down for 14 minutes?" without a war room.
- Compliance-bound teams (fintech, healthcare, public sector) who cannot ship telemetry data to a SaaS in another country.
If you've used Datadog, Grafana + Tempo + Loki, HyperDX, SigNoz, Uptrace, or Jaeger — you are in the right place. Orbtrace replaces all of those with one tool that runs entirely on your hardware.
What's special about Orbtrace?
- 1
On-prem only
There is no SaaS version. Your data never leaves your network. This is on purpose: it's why regulated industries pick us.
- 2
Up and running in minutes
From a bare server to your first trace on screen in about ten minutes. One bundle, one script, and the whole stack — UI, storage, collector — is live. No agents to wire up across ten machines, no cloud accounts to create.
- 3
Causal RCA, not just charts
During an incident, Orbtrace's AI reads the spans, logs and deploys around the anomaly and writes you a paragraph that says "the checkout-api p99 jumped at 14:02 because deploy commit a8f21b changed the SQL query in OrderRepo.findByUser — see spans X, Y, Z". It cites the spans so you can verify it isn't hallucinating.
- 4
Time-Travel Replay
Pick a moment in the past, change one fact ("what if we hadn't deployed that commit?"), and Orbtrace estimates the counterfactual outcome — a grounded probabilistic answer with a confidence interval, citing comparable past incidents, not a full simulation. Useful for post-mortems and for canary decisions.
- 5
Bring your own LLM
Pick the AI provider that fits your compliance posture. The default is Claude. Air-gapped customers run Ollama on a local GPU. Nothing is locked in.
- 6
No per-metric, per-host, per-seat tax
Add as many tags to your metrics as you want — Orbtrace never meters you by ingest volume, host count, or custom-metric cardinality the way SaaS APMs do. You pay a flat annual license by edition; within your edition, capacity is bounded by the hardware you give it, not by a usage meter. (Editions do set guardrails — service count, retention, seats — covered in Licensing.)
What is not in this product?
Setting expectations matters as much as features. Orbtrace does not:
- Run as a hosted SaaS. We will not host data for you.
- Replace your CI/CD pipeline, your APM agents, or your Kubernetes cluster — it sits next to them.
- Auto-discover services without you instrumenting them. You install an OpenTelemetry SDK (one line of config) in each app. Orbtrace does the rest.
- Send your data anywhere by default. Even the AI provider is opt-in; you pick which one to enable.
Where to go next
If you want to see it working before reading any further, jump straight to the Quickstart — it's a ten-minute path from "nothing installed" to "I see my first trace on the screen".
If you'd rather understand the words Orbtrace uses (log, trace, span, SLO, RCA, replay) before touching anything, read Concepts next.
Reading order
The docs are written to be read top-down. Each page assumes you've read the ones above it. The sidebar order is the recommended reading order.