Self-Hosted / Projects / Langfuse
Langfuse
StableSpecs verified September 16, 2026 · against vv4
Overview
Self-hosted LLM observability and evaluation platform for tracing, prompts, costs, and quality.
Langfuse is a self-hosted observability and evaluation platform for applications that use large language models. It collects traces and generations so teams can inspect requests, latency, token usage, costs, and model behavior across development and production.
Why people choose Langfuse
The broader search intent is a self-hosted alternative to hosted LLM observability tools such as LangSmith, with the added need for prompt management and evaluation. Langfuse connects traces to prompts, datasets, experiments, annotations, and feedback so an engineering team can investigate a failure and measure a fix in the same workspace.
What is included
Official product and documentation pages cover observability tracing, metrics dashboards, prompt management, datasets, experiments, evaluations, human annotations, API/SDK integrations, and an OpenTelemetry-based integration path. The platform supports common LLM frameworks and exposes APIs for sending observations and retrieving project data.
Requirements and tradeoffs
Self-hosting Langfuse is a multi-service deployment rather than a single lightweight container. Follow the current self-hosting documentation for the supported application, PostgreSQL, ClickHouse, and object-storage configuration, then plan persistent volumes, secrets, backups, upgrades, and network access. Cost and feature behavior can differ between the open-source self-hosted edition and Langfuse Cloud.
Best fit
Choose Langfuse when you operate an LLM or agent application and need first-party visibility into traces, prompt versions, evaluation results, and spend. A simple log viewer is easier for occasional experiments, while a hosted service may be preferable when you do not want to operate analytical storage and upgrades.
Interface previews
Screenshots of the Langfuse interface.
Feature support
| Feature | Support |
|---|---|
| LLM and agent tracingOfficial documentation positions tracing and observability as core Langfuse capabilities. | Supported |
| Token, cost, and latency analysisThe product site describes tracing LLM calls with cost and latency and provides metrics views. | Supported |
| Prompt management and versioningPrompt management is a documented platform area for storing and working with prompts. | Supported |
| Datasets and experimentsOfficial pages connect datasets and experiments with the evaluation workflow. | Supported |
| LLM output evaluationsLangfuse documents evaluations, scores, and quality-improvement workflows. | Supported |
| Human annotations and feedbackAnnotations and user feedback are included in the documented observability loop. | Supported |
| OpenTelemetry integrationThe official docs provide an OpenTelemetry integration path for instrumenting applications. | Supported |
| SDK and API integrationsThe project provides SDKs and API documentation for sending and managing observations. | Supported |
| Self-hosted deploymentThe official docs provide a self-hosting path with application and analytical data services to configure. | Supported |
| Cloud/self-hosted feature parityThe open-source self-hosted edition and Langfuse Cloud can have different limits, operational requirements, or feature availability. | Partial support |
Community signals
- Stars
- 34.7k
- Forks
- 3.8k
- Open issues
- 960
- Last commit
- 3h ago
- Latest release
- v4.36.1
- Repo created
- May 2023
Includes open pull requests
21h ago
Refreshed nightly from the GitHub API.
Questions
What is Langfuse used for?
Langfuse is used to trace LLM and agent applications, inspect latency and cost, manage prompts, run evaluations, and analyze feedback and quality.
Is Langfuse a self-hosted LangSmith alternative?
Langfuse can serve as a self-hosted alternative for teams that want LLM tracing, prompt management, and evaluation under their own infrastructure. Compare integrations and feature availability with the hosted options you use.
Can you self-host Langfuse?
Yes. Langfuse publishes official self-hosting documentation. The deployment requires configuring multiple data and application services rather than only starting a single stateless container.
Does Langfuse support prompt management?
Yes. Prompt management is a documented Langfuse capability and is connected to tracing and evaluation workflows.
Can Langfuse evaluate LLM outputs?
Yes. The official platform includes evaluations, datasets, experiments, scores, and human annotations for assessing LLM application behavior.
What database does self-hosted Langfuse need?
Follow the current official self-hosting guide for the supported service configuration. Langfuse deployments use PostgreSQL and analytical storage such as ClickHouse, plus object storage where required by the selected version and setup.
Is Langfuse free when self-hosted?
The software is open source, but self-hosting is not operationally free: the operator supplies compute, storage, backups, monitoring, and maintenance. Cloud plans and self-hosted capabilities should be checked separately in the current official documentation.
support // the lab
Found this write-up useful?
If it saved you time or a rebuild, you can support more practical homelab guides.