MLflow

Observability

Observability for AI apps

Gain visibility into your app's logic to debug issues, improve quality, and understand user behavior.

Best-in-class tracing

End-to-end observability

Capture your app's inputs, outputs, and step-by-step execution: prompts, retrievals, tool calls, and more.

Cost and latency tracking

Track cost and latency for each step of your app's execution.

Visualize your app's execution flow

Deep dive into your app's logic and latency with an intuitive UI for debugging.

Understand many traces quickly

Review multiple traces at once with a simplified summary UI to understand how your app processes user requests.

Simple, customizable instrumentation

Automatic instrumentation

Instrument your app with one line of code for 20+ popular LLM SDKs and GenAI frameworks. Customize integrations using our APIs.

SDKs for custom instrumentation

Use decorators, context managers, and APIs to trace custom code or customize integrations.

Instrument once, use everywhere

The same trace instrumentation works in development and production, providing consistent insight across environments.

OpenTelemetry compatible

Export traces to any OpenTelemetry-compatible tool for total ownership and portability of your GenAI data.

Annotation capabilities

Attach quality feedback

Attach quality assessments from users, domain experts, or LLM judges directly to traces to understand and debug quality issues.

Attach metadata

Attach metadata like users, conversation sessions, and custom tags to traces to analyze user behavior.

Comments and notes

Add notes and context directly to traces for collaborative analysis.

Production monitoring

Automated evaluation

Schedule evaluation metrics to automatically run on a sample of production traffic. Monitoring supports the same metrics you use for offline evaluation so you have consistent online and offline quality evaluation.

Dashboards

Visualize operational and quality metrics with real-time dashboards that flag issues, latency spikes, and errors. Drill down to specific traces and steps to debug issues.