Langfuse
LLM engineering platform for tracing agents, managing prompts, running evals and tracking cost and latency.
Trace agent calls, version prompts, run LLM-as-a-judge evals and watch cost, on cloud or your own servers.
Langfuse is an open-source AI engineering platform for agent evaluation and observability. It ingests traces from your LLM app, manages prompts and runs evaluations. It suits developers building LLM apps who want to see what each call did and what it cost.
What it is
You instrument your app with the Python or JS/TS SDKs, OpenTelemetry or integrations such as OpenAI, LangChain, LlamaIndex or LiteLLM, and traces appear in Langfuse. From there you version prompts, run LLM-as-a-judge or manual evaluations, and build dashboards for latency, cost and quality. A beta AI Gateway routes model calls through one endpoint.
Who it's for
It suits teams shipping LLM features that need tracing and evals beyond logs. It is more than a small hobby project needs if you only call a model occasionally, and self-hosting adds infrastructure you have to operate.
Key features
Tracing for LLM apps
Track LLM calls, retrieval, embedding and agent actions as traces.
Prompt management
Version, collaborate on and cache prompts, and change them without redeploying.
Evaluations
LLM-as-a-judge, code evaluators, user feedback and manual labeling.
AI Gateway (beta)
One endpoint for provider calls, with central keys and cost tracking.
Dashboards and metrics API
Track latency, cost and quality with custom dashboards and alerts.
OpenTelemetry-based ingest
Any OpenTelemetry-instrumented app or collector can send traces.
Up and running in 3 steps
Our honest verdict
What's great
- Built on OpenTelemetry, so existing instrumented apps can send traces without a custom SDK.
- Prompt versions and labels can change without changing application code.
- Many integrations are listed, including OpenAI, LangChain, LlamaIndex, LiteLLM and several agent frameworks.
- A Cloud Hobby plan is free and needs no credit card.
Where it falls short
- Self-hosting is a real operations job: Docker Compose suits a single VM, but the README calls Kubernetes with Helm the preferred production deployment.
- Enterprise features live in separate ee directories under a different licence from the MIT core.
- Skill Management and the AI Gateway are labelled beta.
- Cloud usage is metered in units, with the free Hobby plan limited to 50k units per month and 30 days of data access.
Pricing & license
The core is MIT-licensed and free to self-host. Langfuse Cloud (https://langfuse.com/pricing) shows Hobby free with 50k units a month, Core $29 per month, Pro $199 per month and Enterprise $2499 per month, with extra units at $8 per 100k.
Open remote AI-training roles
768 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Can I self-host Langfuse?
Yes. The README gives Docker Compose for a single machine, Helm for Kubernetes and Terraform templates for AWS, Azure and GCP.
What does Langfuse Cloud cost?
The pricing page shows Hobby free, Core at $29 per month, Pro at $199 per month and Enterprise at $2499 per month.
Does it work with LangChain or OpenAI?
Yes. The README lists integrations for OpenAI, LangChain and LangGraph, LlamaIndex, LiteLLM, the Vercel AI SDK and others.
