Langfuse
Open-core LLM observability: traces, evals, and prompt management.
About
Langfuse is the leading open-core observability platform for LLM applications: trace every prompt, completion, tool call, and agent step; score outputs with evals and human feedback; manage and version prompts; and monitor cost, latency, and quality in dashboards. It ships as a self-hostable Docker deployment or as Langfuse Cloud, with Python and JS SDKs that wrap OpenAI, LangChain, and other frameworks. The core is MIT-licensed, but enterprise features under ee/ carry a separate commercial license — and the license header names ClickHouse, Inc. as copyright holder (2023–2026), so verify current ownership terms before committing.
What it can do
- **Tracing**: full traces of prompts, completions, tool calls, agent steps
- **Evals & scores**: LLM-as-judge, human annotations, regression datasets
- **Prompt management**: versioned prompts with production/development labels
- **Dashboards**: cost, latency, quality, and user-level analytics
- **Self-host or cloud**: Docker self-host for data control; managed cloud for zero-ops
- **Broad integrations**: OpenAI, Anthropic, LangChain, LlamaIndex, Vercel AI SDK
Setup
```bash
# Self-host via Docker (see docs/deployment/self-host)
# Or use Langfuse Cloud at cloud.langfuse.com
pip install langfuse
```
```python
from langfuse import Langfuse
langfuse = Langfuse(public_key="pk-...", secret_key="sk-...", host="https://cloud.langfuse.com")
```
Framework integrations (OpenAI, LangChain, Vercel AI SDK, etc.) auto-instrument calls. Docs: langfuse.com/docs.
Prerequisites
Python 3.8+ or Node.js 18+ for SDKs. Self-host: Docker + Postgres/ClickHouse. Cloud: just API keys.
Use cases
Works with
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: 35,713 GitHub stars (log-scaled vs catalog peers); blended with ~23,563,200 monthly package downloads; last push 0d ago; 100 releases/12mo; 1043 open issues. Manual review: leading OSS LLM observability (traces/evals/prompts); 35.7k stars; self-host is Docker-operationally non-trivial; verify current ownership/licensing terms. License: MIT core + proprietary enterprise (ee/) license — open-core; copyright holder per license file: ClickHouse, Inc. (2023-2026).
★ 35,713 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.9/10
- Maintenance · recency, releases, issue load
- 8.8/10
- Openness · license + docs
- 8.6/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · 35,713 GitHub stars (log-scaled vs catalog peers)
- · blended with ~23,563,200 monthly package downloads
- · last push 0d ago; 100 releases/12mo; 1043 open issues
- · manual review: leading OSS LLM observability (traces/evals/prompts); 35.7k stars; self-host is Docker-operationally non-trivial; verify current ownership/licensing terms
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.