MLflow
MLflow is an open-source “AI engineering platform for agents, LLMs and ML models”. Its README says teams use it to debug, evaluate, monitor and optimise production AI applications while controlling costs and managing access to models and data, and cites over 60 million monthly downloads (vendor claim).
Repository snapshot (GitHub API, 2026-11-19)
| Item | Detail |
|---|---|
| Repo | mlflow/mlflow, Python, Apache-2.0 |
| Created | 2018-06-05 |
| Activity | last push 2026-11-19; latest release v3.18.0 (2026-11-04) |
| Size | 28,899 stars, 6,637 forks |
Capabilities
For agents and LLM apps (README): production-grade tracing and observability, evaluation, a prompt registry, prompt optimisation, and an AI Gateway for managing cost and model access.
MLflow originated as experiment tracking and a model registry for classic ML; the README read puts the GenAI features first and does not itemise the classic ones, so treat the classic feature list as general background rather than something verified here.
Where it fits
MLflow is the open, vendor-neutral tracking layer: azure-machine-learning integrates with it, and it is the lineage Databricks built on (databricks). For LLM evaluation and tracing it overlaps with langsmith, braintrust and agent-evals-and-observability; its AI Gateway overlaps with litellm. See also llm-as-judge-and-evals.
Sources
- MLflow README and GitHub API (2026-11-19)