MLflow

MLflow is the open-source answer to experiment tracking and the ML lifecycle: log runs, compare them, register and deploy models, evaluate results, and — in its newer half — trace LLM applications and agents on OpenTelemetry. It runs locally or on your own infrastructure, and its adoption is measured in tens of millions of downloads a month, which is the number that matters for a tool whose value is partly that other people already use it.

The trade-off against a hosted platform is the usual open-source one, and it cuts both ways. You keep the data, you can read and modify the code, and there is no per-seat bill — but you also run the server, the storage and the upgrades, and the polished collaboration features of a commercial product are not the thing you are buying. Its other distinguishing fact is its owner: MLflow comes out of Databricks, so the direction of the project is set by a company with a commercial platform of its own, which is worth knowing when you decide how much of your workflow to build on it.