Odahu splits the ML/AI model lifecycle into three phases:

  1. Train
  2. Package
  3. Deploy

Applications and tools can further automate each phase by implementing pluggable extensions as

  1. Trainer
  2. Packager or
  3. Deployer

Trainers and Packagers can be registered as components of the Odahu Platform using:

  1. Trainer Extension
  2. Packager Extension

When registered, these components can use Odahu Trainer Metrics and Trainer Tags.

Users are encouraged to integrate third-party Trainer Extensions and Packager Extensions.


Taken together a Trainer, Packager, and Deployer comprise a Toolchain that automates an end-to-end machine learning pipeline.

Ready to use

Odahu provides a Trainer Extension and a Packager Extension

  1. MLflow Trainer
  2. REST API Packager

These power the default Toolchain.

Model storage

Odahu Platform stores models in Trained Model Binaries for different languages.

Presently, Odahu Platform supports only:

  1. General Python Prediction Interface

Users are encouraged to provide additional formats.