Repolode
Submit
aimhubio

aimhubio/aim

Self-hosted open-source experiment tracker for logging ML training runs and metadata, with a UI to compare them and a Python SDK to query them.

Open-source alternative to:Weights & BiasesTensorBoardMLflow

View on GitHubWebsite
Stars
6,289
Forks
413
Open issues
483
Last commit
20 hours ago

What is it?

Aim is a self-hosted experiment tracking tool for machine learning. You install the Python package, add a few calls to your training code to record parameters, metrics, images, text and other metadata, then run `aim up` to open a web UI for exploring and comparing runs.

The UI supports grouping, aggregation and filtering runs by their tracked parameters through Python-style query expressions. The project states it is designed to handle tens of thousands of training runs. A Python SDK exposes the same tracked data programmatically, which suits automation and Jupyter notebook analysis. Built-in converters import logs from TensorBoard, MLflow and Weights & Biases, and callbacks are provided for frameworks such as PyTorch Lightning, Hugging Face, Keras and XGBoost.

For teams, a centralized remote tracking server collects runs from multiple hosts, and an official Docker image and Kubernetes deployment guide are available.

Key features

  • 01Logs parameters, metrics, images, audio and other metadata from training runs.
  • 02Web UI for comparing, grouping and aggregating runs by tracked parameters.
  • 03Python query expressions to filter runs and metrics.
  • 04Python SDK for programmatic access to tracked data, including Jupyter use.
  • 05Converters for importing TensorBoard, MLflow and Weights & Biases logs.
  • 06Centralized remote tracking server and Docker/Kubernetes deployment options.

Use cases

  • Comparing hundreds or thousands of hyperparameter sweep runs in one dashboard.
  • Migrating existing TensorBoard or MLflow logs to a more scalable comparison UI.
  • Collecting training runs from several machines into one shared tracking server.
  • Analyzing experiment results programmatically in Jupyter notebooks.

Things to know

  • The README lists platform support as Linux and macOS only.

In collections

Data science courses & tools

Topics

#experiment-tracking#machine-learning#mlops#data-visualization#metadata-tracking#pytorch#tensorboard#python#ai#data-science

FAQ

What is aim?
Self-hosted open-source experiment tracker for logging ML training runs and metadata, with a UI to compare them and a Python SDK to query them. Aim is a self-hosted experiment tracking tool for machine learning. You install the Python package, add a few calls to your training code to record parameters, metrics, images, text and other metadata, then run `aim up` to open a web UI for exploring and comparing runs.
Is aim free to use? What license does it have?
aim is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is aim still maintained?
Actively maintained. The last commit to aim was on Oct 21, 2026.
What can I use aim for?
Comparing hundreds or thousands of hyperparameter sweep runs in one dashboard. Migrating existing TensorBoard or MLflow logs to a more scalable comparison UI. Collecting training runs from several machines into one shared tracking server. Analyzing experiment results programmatically in Jupyter notebooks.
Where can I get aim?
The source code and installation instructions are on GitHub: https://github.com/aimhubio/aim

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects