MLflow Tracing
mlflow.org
- Free plan
- Open source
- Linux
- Web
- API
- Self-hosted
What MLflow Tracing is
MLflow Tracing is a free, open-source tool for recording what happens during LLM application and agent requests. It captures inputs, outputs, and metadata for each step, then presents trace details such as a span tree, per-span latency, tool calls, exceptions, token counts, cost breakdowns, feedback, and ground-truth annotations. Automatic tracing is listed for more than 40 LLM and AI-agent libraries and frameworks, including OpenTelemetry, OpenAI, Anthropic, LangChain, LangGraph, LlamaIndex, DSPy, and Vercel AI. Python and TypeScript SDKs also provide manual tracing APIs. The service supports OpenTelemetry and GenAI Semantic Conventions, with hybrid deployment and data hosted on your infrastructure. Custom span processors can mask sensitive data before transmission. The mlflow-tracing package has a documented 95% smaller footprint than the full mlflow package. Production guidance recommends asynchronous logging and sampling.
Who it suits
It suits teams diagnosing LLM or agent behavior across development and production. It is also useful for organizations needing self-hosted trace data, token and cost tracking, or integrations with supported frameworks.
Good to know, and what to watch
- Free open-source plan
- Automatic tracing for 40-plus libraries
- Detailed span, token, and cost views
- OpenTelemetry-compatible trace export
- Sensitive-data masking before transmission
- TypeScript SDK excludes browser and Edge Runtime
- OTLP ingestion needs MLflow 3.6.0 or later
- Production TLS and authentication need proxy or VPN guidance
Plans
Every plan and priceMLflow Tracing
Free plan
The verdict
MLflow Tracing combines broad framework coverage with detailed request-level diagnostics. Its self-hosted model and production guidance fit teams that need control over trace data.
What the maker publishes
- Automatic instrumentation
- MLflow offers one-line automatic tracing integrations for supported LLM and agent libraries, including OpenAI, LangChain, and DSPy. source
- Automatic tracing
- The integrations page lists automatic tracing for more than 40 LLM and AI agent libraries and frameworks. source
- Browser limitation
- The TypeScript SDK works in Node.js backend environments and is not compatible with frontend browser or Edge Runtime environments. source
- Company founded
- Databricks says it was founded in 2013 by the creators of Apache Spark, Delta Lake, MLflow, and Unity Catalog. source
- Company headquarters
- Databricks says it is headquartered in San Francisco. source
- Dataset creation
- Production traces can be used to build evaluation datasets that retain details about components such as retrievers and tools. source
- Developer support
- The repository directs users to documentation, GitHub issues, a mailing list, and Slack for help and project discussions. source
- Documentation
- MLflow documentation has separate sections for LLM and agent development, tracing, and evaluation, and for traditional machine-learning workflows. source
- Evaluation
- MLflow connects trace inspection with evaluation results to help assess application quality. source
- Human feedback
- Users can attach feedback to traces with metadata such as the user and timestamp. source
- Integrations
- Listed integrations include OpenTelemetry, OpenAI, Anthropic, LangChain, LangGraph, LlamaIndex, DSPy, Vercel AI, and other frameworks, providers, gateways, and tools. source
- Integrations
- The integrations directory organizes automatic tracing support across frameworks, coding agents, and model providers. source
- Licence
- The public MLflow repository identifies its licence as Apache-2.0. source
- Lightweight SDK
- The production tracing package, `mlflow-tracing`, is documented as having a 95% smaller footprint than the full `mlflow` package. source
- Maker
- The MLflow project is developed at Databricks, according to TrustRadius. source
- Maker and community
- The MLflow site describes the project as backed by the Linux Foundation and committed to open source for more than five years. source
- Manual instrumentation
- The Python and TypeScript SDKs provide manual tracing APIs, and the Python SDK supports synchronous and asynchronous functions with the documented version requirements. source
- Manual instrumentation
- Python developers can instrument custom functions with the `@mlflow.trace` decorator or create spans directly. source
- OpenTelemetry
- MLflow Tracing is OpenTelemetry-compatible and supports GenAI Semantic Conventions for trace export and ingestion. source
- OpenTelemetry
- MLflow Tracing is compatible with OpenTelemetry and supports GenAI semantic conventions for trace export and ingestion. source
- OTLP ingestion limit
- OpenTelemetry trace ingestion requires MLflow 3.6.0 or later and a SQL-based backend store. source
- Pricing
- The MLflow Tracing page describes the open-source product as free and says trace data can be hosted on the user's infrastructure. source
- Production monitoring
- Tracing records latency, token usage, and quality metrics to help monitor deployed LLM applications and agents. source
- Production package
- The product documentation recommends the `mlflow-tracing` package for production deployments and says it has a 95% smaller footprint than the full MLflow package. source
Checked against the maker's pages on 21 Oct 2026.