Arboris (family)

This repository is the family hub for Arboris. It carries the model card and license. Trained variants, once they exist, will each live in their own repository under the CoastalDigitalResearch organization so that from_pretrained(...) resolves a single model per repo.

Status: architecture preview, pre-alpha. Arboris is in active development. No trained weights are published, and no variant repositories are reserved yet. Source code lives on GitHub, not here — see Source below.

What Arboris is

Arboris is a graph-based, dendritic model exploring an alternative approach to world models. Where a conventional world model compresses state into a single latent that rolls forward in time, Arboris represents state as a graph and grows its predictive structure dendritically, along the relations that carry signal. It is an early research direction, and the design is expected to change.

Intended use

Research into graph-native world models and dendritic representation learning. This is an experimental architecture preview, not a production model.

Source

The training/eval/research code is not distributed through this model repository. It is published as source on GitHub: https://github.com/AdamPippert/Arboris.

Partnership

Arboris is developed by independent researcher Adam Pippert. Coastal Digital Research trains and publishes the models.

License

Released under the Apache License 2.0. See LICENSE and NOTICE.