# MLflow / Aim

Experiment tracking you host. MLflow is Apache-2.0; Aim is Apache-2.0 — Weights and Biases without the hosted run meter.

**Replaces:** Weights and Biases

**Tier:** Medium

## Problem

Weights and Biases is a hosted experiment tracker. You wanted runs, params, and artifacts on a server you control. MLflow is Apache-2.0 (mlflow/mlflow). Aim is Apache-2.0 (aimhubio/aim). This is not W&B's every artifact SKU, Reports, or a hosted model registry SaaS. LLM traces are Langfuse — /replace/langsmith. You still pay disk.

## Stack

- [MLflow](https://mlflow.org) — Apache-2.0 experiment tracking, models, and evaluation
- [Aim](https://aimstack.io) — Apache-2.0 experiment tracker if MLflow is not the shape

## Estimated SaaS cost saved

| You stop paying | Typical / year | Self-host instead |
|---|---:|---|
| Weights and Biases (small team) | $1,200 | MLflow or Aim + artifact storage |

**About $1,200 / year** on a small team, before you count lock-in.

## Deploy on Railway

Placeholder: MLflow tracking server plus a volume or S3-compatible store for artifacts. Aim if you want that UI. No one-click template in this repo yet — Start on Railway. LLM traces stay on /stacks/langfuse-observe.

[Start on Railway](https://c17.io/go/railway/mlflow-tracking?from=%2Fstacks%2Fmlflow-tracking.md)

[Langfuse if this is LLM traces, not training runs](https://c17.io/stacks/langfuse-observe)

## AGENT_PROMPT

```
Replace Weights and Biases with self-hosted MLflow (preferred) or Aim.

- Official tracking server. Persist the backend store and artifacts. Log one run you already have. Do not claim W&B Reports.
- LLM traces stay on /stacks/langfuse-observe. Write AGENT_PROMPT.md.
```

## FAQ

### MLflow or Aim?

MLflow if you want the common Apache-2.0 tracker (the published W&B stack). Aim if you want that UI — /compare/mlflow-vs-aim.

### Is this LangSmith?

No. LLM traces are /replace/langsmith (Langfuse). This page is training and eval runs.

### Are these free?

Both catalog rows are Apache-2.0. You pay disk and compute.

Source: https://c17.io/stacks/mlflow-tracking