What is Langfuse?
Langfuse is an open-source LLM engineering platform for tracing, evaluating, monitoring, and improving large language model (LLM) applications and AI agents. It connects production observability with prompt management, datasets, experiments, and human feedback.
Key Capabilities
- LLM observability: Captures hierarchical traces of LLM calls, tool invocations, and retrieval steps, with filtering by user, session, cost, latency, and custom metadata.
- Prompt management: Separates prompts from application code and supports versioning, deployment, and rollback.
- Evaluations: Runs LLM-as-a-judge, heuristic, and human-review evaluations against production data and experiments.
- Experiments and datasets: Defines test cases, compares model or prompt results, and creates regression datasets from reviewed traces.
- Monitoring: Provides dashboards and alerts for LLM cost, latency, and quality.
- Developer integrations: Supports OpenTelemetry instrumentation, native Python and TypeScript SDKs, APIs, a command-line interface, and an MCP server.
Audience & Use Cases
- Audience: AI engineers, application developers, ML engineers, and platform teams building LLM applications or AI agents.
- Use cases: Debugging production agent behavior, testing prompt and model changes, managing prompts, measuring quality and cost, and creating evaluation workflows from production data.
Technical Specifications
- Deployment: Cloud service or self-hosted deployments using Docker Compose, Kubernetes/Helm, or Terraform for AWS, Google Cloud, and Azure.
- License: Product features are available under the MIT License.
- Architecture: Uses OpenTelemetry-compatible tracing and supports storage and querying components including ClickHouse, Redis, and object storage.
Categories & Use Cases
Technical Details
| Mobile Application | No |
|---|
FAQs
What is Langfuse?
Langfuse is an open-source LLM engineering platform for tracing, evaluating, monitoring, and improving large language model (LLM) applications and AI agents. It connects production observability with prompt management, datasets, experiments, and human feedback.