Coding & Development

MLflow

The open-source AI platform for tracking, evaluating and deploying models and agents

Free and open source (Apache 2.0). Self-hosting costs only your own infrastructure. Managed hosting is offered by Databricks; its pricing could not be verified today.
Visit MLflow →
Pricing
Free and open source (Apache 2.0). Self-hosting costs only your own infrastructure. Managed hosting is offered by Databricks; its pricing could not be verified today.
Best for
Good for ML engineers, data scientists and platform teams who can run a server and want one system for training experiments and LLM traces.
Official site
mlflow.org
Last checked
2026-10-21

MLflow is a strong pick for engineering teams that want free, self-hostable experiment tracking and LLM observability without vendor lock-in. Skip it if you want a polished hosted product with support and no ops work, or if you only need a quick prompt playground. It started as a classical ML lifecycle tool (tracking runs, registering models, deploying them) and has grown to cover LLM and agent work: OpenTelemetry-based tracing, evaluation with built-in metrics and LLM judges, a prompt registry, an AI Gateway for routing between model providers, and an Agent Server for deployment. The site calls itself Apache 2.0 licensed and Linux Foundation backed, and lists 30M+ monthly package downloads, 20K+ GitHub stars and 900+ contributors (vendor-reported figures).

Best for

Good for ML engineers, data scientists and platform teams who can run a server and want one system for training experiments and LLM traces. Not for non-technical teams, for people who want a fully managed service with an SLA out of the box, or for anyone who only needs lightweight prompt testing.

Key features

01

LLM and agent tracing

Records traces of LLM apps and agents using OpenTelemetry, with production monitoring of quality, cost and safety.

02

Evaluation and issue detection

Runs evaluations with 50+ built-in metrics and LLM judges, and can flag problems found in traces.

03

Prompt registry and optimization

Versions prompts, tracks lineage, and offers automatic prompt optimization.

04

AI Gateway

A single OpenAI-compatible API in front of several LLM providers, with rate limits, fallbacks and cost control.

05

Agent Server

A FastAPI-based way to deploy agents as a production endpoint.

06

Experiment tracking and model registry

The original core: log parameters, metrics and artifacts per run, compare runs, version and deploy models.

Pricing breakdown

MLflow (open source)

Free
Apache 2.0; you pay only for your own hosting
  • All features above
  • Self-hosted, e.g. via the quickstart command shown on the site
  • Python, TypeScript/JS, Java and R support (per the previous listing; not re-verified on the homepage)

Managed MLflow (Databricks)

Pricing could not be verified
The Databricks pricing page could not be fetched today; previously described as billed through Databricks platform usage
  • Hosted by Databricks
  • Check databricks.com for current terms

Pros and cons

Pros

  • No license cost and no usage meter; the only spend is the servers and storage you choose.
  • Open standards (OpenTelemetry tracing, OpenAI-compatible gateway) reduce lock-in.
  • One tool for both traditional ML runs and LLM/agent traces, so teams do not need two systems.
  • Large community and many integrations, so most frameworks already have a connector.

Cons

  • Self-hosting means you handle upgrades, scaling, backups, authentication and access control.
  • The scope is wide; the UI and concepts can feel heavier than dedicated LLM-tracing tools if that is all you need.
  • Managed hosting is a separate Databricks product, and its pricing was not verifiable from the public MLflow site.
  • Popularity figures on the site are self-reported; no independent review score was verified.

Frequently asked questions

Is MLflow really free?

The software is free under Apache 2.0 per mlflow.org. You still pay for wherever you run it, such as servers, a database and storage.

Does MLflow work for LLM apps, or only classical ML?

Both. It began with experiment tracking and model registry, and now includes tracing, evaluation, a prompt registry and an AI Gateway for LLM apps and agents.

Do I need Databricks to use it?

No. The open-source version runs on its own. Databricks sells a managed version, but its current pricing could not be verified here.

How do I try it?

The site points to a live demo and a local quickstart; check mlflow.org for the current command.

Ready to try MLflow?

Head to the official site to explore pricing and start a free trial where available.

Visit MLflow →