Langfuse

Langfuse is an open-source AI engineering platform for tracing, evaluating, and improving LLM applications and agents with observability, prompt management, experiments, and human review.

AI Agent Development AI Developer Tools Monitor & Log Management LLM Prompt

Visit Website

Open-source AI engineering platform

Langfuse is an open-source AI engineering platform for tracing, evaluating, and improving LLM applications and agents. It combines observability, prompt management, experiments, and human annotation in one workflow so teams can move from prototype to production using real usage data.

The platform is built around OpenTelemetry-compatible tracing, native SDKs, and broad integrations, which lets teams capture LLM and non-LLM activity without locking into a single framework. Langfuse also supports cloud deployment and self-hosting, with the product and core features available under an MIT license.

Core capabilities

LLM and agent tracing

Capture hierarchical traces for LLM calls, tool invocations, retrieval steps, and other application logic. Filter and inspect traces by user, session, cost, latency, or custom metadata.

Sessions, users, and agent graphs

Track multi-turn conversations as sessions and add user tracking for production debugging and usage analysis. Agents can also be represented as graphs for more complex workflows.

Prompt management workflow

Manage prompts separately from application code with versioning, deployments by label, rollbacks, prompt caching, and playground testing. Prompt history and change tracking help teams review how prompts evolve.

Evaluation and human review

Run evaluations with LLM-as-a-judge, heuristic functions, or human review on production data or experiments. Support for datasets, experiments, evaluation scores, and human annotation helps compare changes over time.

Operational metrics and alerts

Monitor quality, cost, and latency through dashboards, alerts, and trace-linked metrics. This makes it easier to understand the impact of prompt or model changes on production behavior.

Open integrations and deployment options

Connect through native SDKs, OpenTelemetry, proxy-based logging, APIs, exports, and more than 100 integrations. The platform also supports self-hosting and data portability.

Common use cases

Production observability
Instrument production LLM applications to inspect traces, sessions, and user-level behavior when debugging latency, cost spikes, or unexpected outputs.
Prompt iteration
Manage prompts as versioned assets, deploy them by label, roll back changes, and compare prompt variants in the playground before shipping updates.
Evaluation workflows
Run offline or online evaluations on datasets, then compare experiments with LLM-as-a-judge, heuristics, or human review to assess quality changes.
Human-in-the-loop review
Create human annotation queues and review traces to build golden datasets or validate model behavior with collaborators.
End-to-end LLM development
Use the same platform across prototypes and production systems to connect instrumentation, experimentation, and iteration in one workflow.

Pros and Cons

Pros

Combines tracing, prompt management, evaluation, experiments, and human annotation in one platform.
Works with existing stacks through OpenTelemetry, native SDKs, API-based access, and many integrations.
Supports both cloud hosting and self-hosting, with open-source and MIT-licensed core features.
Includes workflow features for production iteration, such as prompt versioning, rollbacks, datasets, and side-by-side comparisons.

Cons

The source does not spell out a single turnkey setup path, so implementation effort will depend on the stack and integration method you choose.
Some advanced capabilities, such as enterprise SSO, SCIM, audit logs, and dedicated support, are tied to higher-tier plans or add-ons.

FAQ

What problem does Langfuse solve?

Langfuse is built for teams that want to trace LLM and agent workflows, manage prompts, and evaluate outputs in one system. It supports tracing, prompt management, evaluation, experiments, and human annotation.

How do teams integrate Langfuse?

The source highlights native SDKs for Python and JavaScript, OpenTelemetry support, 100+ integrations, and options to capture traces through an LLM gateway such as LiteLLM. The product is designed to work with existing stacks rather than require a single framework.

What can teams do after data is collected?

Langfuse supports traces, sessions, user tracking, prompt versioning, prompt deployments, playground testing, experiments, evaluation scores, datasets, and human review workflows. The docs describe it as a connected workflow from prototype to production.

Does Langfuse offer both cloud and self-hosted deployment?

The pricing page shows a free Hobby plan, paid Cloud plans, and a self-hosted option. It also lists managed-cloud plans with longer retention, higher limits, and enterprise features such as SSO, SCIM, audit logs, and support options.

Who is Langfuse best suited for?

The product is designed for LLM applications and agents. The documentation emphasizes tracing LLM and non-LLM calls, production debugging, prompt iteration, evaluation, and quality, cost, and latency monitoring.