Open-source AI infrastructure

Everstack vs Langfuse

Compare Everstack and Langfuse on licence terms, self-hosted feature gating, evaluations, gateway routing and agent runtime, with working code for both.

How a trace gets created

Langfuse is not a gateway. Its own documentation states that your requests never pass through Langfuse on the way to a provider, so traces come from client-side instrumentation: changing an import, wrapping calls in observations, or shipping OpenTelemetry spans yourself. Everstack routes the request, so the trace already exists before any SDK is added.

Licence and self-hosted feature gating

Langfuse's core is MIT and free to self-host without usage limits, but everything under ee/, web/src/ee/ and worker/src/ee/ carries a separate Langfuse Enterprise Licence. Project-level RBAC, audit logs, data retention policies, server-side data masking, SCIM, and the Instance Management API need a paid key when self-hosting. Everstack ships under one Apache-2.0 licence with no enterprise directory.

Where Langfuse is the better choice

Langfuse has a far larger community, over 33,900 GitHub stars as of August 2026, mature prompt management, a well documented OpenTelemetry ingestion path, and backing from ClickHouse, which acquired it in January 2026. If you want tracing and evaluations for an application whose model calls already route somewhere you control, Langfuse is the simpler adoption.