LangFuse
LangFuse is an open-source LLM observability platform that provides tracing, evaluation, and analytics for applications built on top of large language models. In Ilum, LangFuse is available as an optional module that integrates with the platform's authentication, networking, and lineage stack.
LangFuse is disabled by default in Ilum and enabled per deployment.
When to use LangFuse
LangFuse is the right choice for teams that:
- Build production applications on Spark or Trino-backed datasets that include LLM calls (RAG pipelines, classification, summarization, agentic workflows).
- Need detailed tracing of prompts, completions, and agent steps to debug and improve model behaviour.
- Want to evaluate LLM output quality with structured scores, user feedback, and offline evaluators.
- Prefer to keep LLM observability data inside the same governance boundary as their data lakehouse.
What LangFuse provides
- Traces: End-to-end records of multi-step LLM workflows, including nested calls, tool invocations, and intermediate prompts.
- Sessions: Group related traces into user or workflow sessions for behavioural analysis.
- Scores: Structured evaluation signals (user feedback, model-graded scores, custom evaluators) attached to traces.
- डेटासेट : Reproducible test sets used for regression testing and prompt iteration.
- Prompt management: Version-controlled prompts with deployment workflows.
- Cost and latency tracking: Per-trace token usage, cost, and latency dashboards.
Integration with Ilum
When enabled, LangFuse is exposed in the Ilum UI through the मॉड्यूल sidebar, gated by the LANGFUSE_READ permission. Authentication can be wired through Ilum's OAuth2 provider mode (Ory Hydra) so users sign in once.
LangFuse uses पोस्टग्रेएसक्यूएल for metadata and मिनियो (or any configured S3-compatible store) for trace storage, sharing infrastructure with the rest of the Ilum platform.
संरूपण
Enable LangFuse through the umbrella Helm chart:
langfuse:
सक्षम : सच्चा
Refer to the official LangFuse Helm documentation for the full set of values, including secrets and Postgres configuration.