MLflow

MLflow — Developer Documentation Highlights

Licensing & Hosting

• Apache 2.0 open-source license — free to use, modify, and distribute.
• Fully self-hostable; deploy on your own infrastructure or any cloud environment.

API & Integrations

• API available for programmatic experiment tracking, model registration, evaluation, and gateway configuration.
• First-class integrations: LangChain, OpenAI, PyTorch, OpenTelemetry — plus 100+ additional ecosystem integrations.

Observability & Tracing

• Tracing architecture is built on OpenTelemetry standards, enabling interoperability with existing observability stacks.
• Captures spans across LLM calls, agent steps, retrieval operations, and tool use.

Evaluation Framework

• 50+ built-in metrics covering accuracy, relevance, toxicity, and more.
• Supports LLM-as-judge evaluation patterns for qualitative assessment at scale.

Prompt Management

• Prompt versioning and testing workflows integrated into the MLflow tracking server.
• Supports prompt optimization pipelines.

AI Gateway

• Unified gateway interface for routing to multiple LLM providers, centralizing authentication and rate-limit management.

Model Registry

• Artifact logging, model versioning, and stage transitions (staging → production) built into the core platform.