https://raw.githubusercontent.com/OrcVole/langfuse-cloudron/main/CloudronVersions.json
Description
4.67.0
Langfuse is an open-source LLM engineering platform. It gives you observability and tracing for LLM applications, evaluations, prompt management, a prompt playground, datasets, and usage/cost metrics — all self-hosted on your own Cloudron.
This package runs the open-source (MIT) edition. It integrates with OpenTelemetry, the OpenAI SDK, LangChain, LiteLLM, LlamaIndex, and most LLM tooling, so it can act as the observability sink for the rest of your stack.
On this Cloudron:
- Langfuse keeps its own login (email/password plus optional single sign-on through your Cloudron users via OIDC). It is not fronted by the Cloudron SSO proxy.
- Trace ingestion uses Langfuse project API keys (a public/secret key pair you create in the UI),
sent to the open
/api/public/*endpoints — including the OpenTelemetry endpoint/api/public/otel/v1/traces. - ClickHouse and MinIO are bundled and managed for you; PostgreSQL and Redis come from Cloudron addons. All trace data and object storage live in the app's backed-up data directory.
The enterprise (/ee) features (RBAC, audit logs, data retention policies, SCIM) are intentionally
not included.
Recent Changes
Fix ClickHouse merges failing in a loop on larger installs. Langfuse's
events_fulltable carriesfull-text indexes that are rebuilt for the whole merged part during every merge, and with up to 32
merges at once the bundled ClickHouse ran into its 2.5 GiB memory cap (MEMORY_LIMIT_EXCEEDED, about
1.4 times a second on a 400,000-event install), so parts piled up and the 30-day retention could not
drop old rows. The ClickHouse cap is now 4 GiB, background merges are limited to 4 threads, and the
app's memory limit rises from 6 GiB to 8 GiB to keep the same headroom for the web and worker
processes. If you changed this app's memory limit by hand, check it is at least 8 GiB.
Langfuse itself is unchanged (4.67.0).