We already trace everything in Langfuse.
Keep it. Omnia’s traces go to Langfuse as standard OpenTelemetry. What Langfuse leaves to you is acting on what it sees: holding a version back, rolling one back, refusing a tool call.
That version was on part of the traffic, as every new version is. The same metrics stopped its promotion and rolled it back, and its failing sessions became regression scenarios. docs
on langfuse Online evaluation scores live traces as they arrive. Langfuse describes its role as ex-post evaluation, and points to other libraries for run-time measures. their docs
At a glance
| Topic | Omnia | Langfuse, per its docs |
|---|---|---|
| Open source | yes PromptArena and PromptKit, Apache 2.0. docs | documented MIT outside the enterprise folders. source |
| OpenTelemetry | yes Emits it, and ingests it from agents elsewhere. docs | documented Ingests OTLP over HTTP. source |
| CI gate | yes The suite gates the release. docs | documented A GitHub Action fails on a score threshold. source |
| Simulated users | yes Built in, against real tools. docs | partly Cookbook. source |
| Voice testing | yes Barge-in, overlap, dead air. docs | not documented Not documented. source |
| Runs and rolls out agents | yes Yes, with automatic rollback. docs | not documented Observes only. source |
| Runtime enforcement | yes Per-call policy and redaction. docs enterprise | not documented Not in scope. source |
Common questions
- Can I keep Langfuse and use Omnia?
- Yes. Omnia emits standard OpenTelemetry, which Langfuse ingests.
- Who owns Langfuse?
- ClickHouse acquired Langfuse in January 2026; the core stays MIT licensed.
Put your own agent through the gate.
Install Omnia in your cluster and run it free while you build.
Langfuse claims are from Langfuse, now part of ClickHouse’s published documentation, checked 22 October 2026. Spot something out of date? Tell us.