Langfuse is an open-source LLM engineering platform that provides observability, analytics, and monitoring tools for AI applications built on large language models. It enables developers to trace LLM calls, evaluate model outputs, manage prompts, and debug production issues through a comprehensive dashboard and SDK integrations. The platform supports popular frameworks like LangChain, LlamaIndex, and OpenAI, and can be used as a cloud-hosted service or self-hosted deployment.
Founded
2023
Company Size
11-50 employees
Headquarters
Berlin, Germany
Funding
Seed
Public materials reference SOC 2 Type II and ISO 27001 reports, as well as GDPR-related data processing agreements. HIPAA alignment is also mentioned.
Hierarchical traces capture LLM calls and related application logic, including tool invocations, retrieval steps, and multi-turn sessions. This is the core observability feature for debugging and understanding production behavior.
Provides a central UI to manage prompts, versions, releases, and prompt experimentation. Teams can iterate on prompts without redeploying application code.
Supports online and offline evaluations using LLM-as-a-judge, heuristic functions, and human review. Evaluations can be run on production data or during experiments.
Allows teams to test application behavior before deployment using datasets and experiments via the UI or SDK. This helps compare prompt or model changes against known cases.
Tracks key metrics such as cost, latency, quality, and user behavior to help teams monitor performance over time. The UI is designed for inspecting logs and debugging issues quickly.
Collects user and employee feedback to help score and improve LLM outputs. Feedback can be incorporated into evaluation workflows.
Supports multi-turn conversations as sessions and includes user tracking. This makes it easier to analyze behavior across repeated interactions.
Langfuse Cloud is available as a managed SaaS offering. AWS Marketplace references monthly usage billing through AWS for the cloud service.
Langfuse can be self-hosted as an open-source deployment. This gives teams control over infrastructure and data residency.
Native Python SDK for instrumenting LLM applications and sending traces, prompts, and evaluation data. It is one of the primary integration paths for developers.
Native TypeScript SDK for integrating Langfuse into JavaScript and TypeScript applications. It supports tracing and observability for modern web and agent stacks.
Integrates with popular LLM frameworks and tools including LangChain, LlamaIndex, OpenAI, Dify, and LiteLLM. Documentation also references 50+ library/framework integrations.
Can capture traces via OpenTelemetry for broader observability integration. This enables instrumentation beyond native SDKs.
Supports tracing via an LLM gateway such as LiteLLM. This helps teams observe requests routed through gateway layers.
Self-hosting documentation lists authentication and SSO support, indicating enterprise authentication capabilities are available in the platform.
Self-hosting documentation and pricing references automated access provisioning and SCIM API support for user management.
Pricing information references project-level role-based access control. This helps limit access by team or project.
Supports client-side data masking to help protect sensitive information in traces and logs. This is useful when handling personal or confidential data.
Common questions about Langfuse features, pricing, and capabilities
Langfuse provides detailed distributed tracing that visualizes the entire execution flow of your AI application. You can inspect every step of a nested chain, including the exact inputs, outputs, latency, and token usage for each individual LLM call, making it easy to identify where a logic error or bottleneck occurred.
Yes, Langfuse includes a dedicated Prompt Management system. This allows you to decouple prompts from your application code, version them, and instantly deploy updates to production. You can also link specific traces to the prompt version used, enabling better evaluation of how prompt changes affect output quality.
Langfuse offers comprehensive evaluation features, including manual human annotation and automated model-based evaluations. You can define custom scores for metrics like relevance, toxicity, or accuracy, helping you systematically measure and improve the performance of your AI models over time.
Migration is seamless; you simply update your API keys and environment variables to point from your local or staging environment to your production project in the Langfuse dashboard. Because Langfuse is open-source, you can start with a local container and move to our managed cloud whenever you are ready to scale.
Yes, the Hobby plan is completely free and includes all core observability features, making it perfect for proof-of-concepts and personal projects. It allows you to explore the dashboard, test integrations, and set up your first traces without any financial commitment.
Langfuse offers native SDKs and integrations for popular frameworks including LangChain, LlamaIndex, and the OpenAI SDK. It is designed to be framework-agnostic, meaning you can also use our Python or TypeScript SDKs to manually instrument any custom LLM implementation or proprietary model.
Absolutely. While we offer high-level integrations for common frameworks, our flexible API and SDKs allow you to send traces from any environment. You can wrap your custom functions with our decorators or use our low-level logging methods to capture data from any model provider or internal setup.
On the Hobby plan, data is accessible for 30 days, while the Core plan extends this to 90 days. Once the retention period expires, older traces and logs are automatically purged from the system. If your project requires long-term historical analysis, we recommend the Pro plan which offers 3 years of data access.
Starting from the Core plan at $29/month, Langfuse allows for unlimited users. This ensures your entire engineering and product team can collaborate on debugging and prompt management without worrying about per-seat licensing costs, which is ideal for growing startups.
Langfuse is headquartered in Germany and adheres to strict European data privacy standards. For our cloud-hosted service, we use secure infrastructure with encryption at rest and in transit. For organizations with high security requirements, we also offer a self-hosted deployment option to keep all data within your own VPC.
The Enterprise plan is designed for large-scale teams and includes enterprise-grade security features, dedicated support, and custom compliance assistance. This tier is tailored to meet the rigorous auditing and data handling requirements of heavily regulated industries and large corporate environments.
Pro plan users receive priority email support to help resolve technical issues quickly. Enterprise customers benefit from a dedicated support channel, faster response time SLAs, and direct access to our engineering team for architectural guidance and advanced troubleshooting.
Great for hobby projects and POCs. 30 days data access.
Starting at
$0.00/month
Free plan
First 50,000 units included
For 0-50,000 units
50k units per month included
First 2 users included
For 0-2 users
Up to 2 users
For production projects. 90 days data access and unlimited users.
Starting at
$29.00/month
Monthly base subscription
First 100,000 units included
For 0-100,000 units
First 100k units included
For 100,001-1,000,000 units
$8 per 100k units
For 1,000,001-10,000,000 units
$7 per 100k units
For 10,000,001-50,000,000 units
$6.5 per 100k units
$6 per 100k units
For scaling projects. 3 years data access, high rate limits.
Starting at
$199.00/month
Monthly base subscription
Optional add-on for SSO and RBAC
First 100,000 units included
Standard graduated unit pricing applies after 100k included units
For large scale teams. Enterprise-grade support and security.
Starting at
$2499.00/month
Starting price for Enterprise
Custom volume pricing available
User reviews coming soon
We're building our review system to help you make informed decisions.
Performance data coming soon
We're collecting uptime and performance metrics to provide comprehensive insights.