NMAI — New Millennium Artificial Intelligence
NMAI is the AI line of the Klondike Software Project (KSP). It covers all of KSP's machine learning work: language models, and any future NMAI efforts outside that scope.
Structure
NMAI is organized as a family of separately-named models rather than one
single line. Each model gets its own name, its own repos, and its own
versioning — a model's version number (e.g. 1.3) tracks that model's own
history and does not correspond to any other NMAI model's version number.
Everything under the NMAI name is built the same way: trained from scratch, independently, on hardware KSP owns or has direct access to — no third-party training infrastructure, no outsourced training runs.
Organization
- Repositories live under the
KSP-NMAIHugging Face org, one repo per model or model variant. - Model cards are the source of truth for any individual model's architecture, training data, and benchmark results — this README does not duplicate that information.
- Datasets used across NMAI models are drawn from existing public corpora (e.g. FineWeb-Edu, DCLM-baseline) rather than NMAI-original data collection, unless a specific model's card says otherwise.
License
NMAI's models and code are released under the Apache License, Version 2.0. Copyright and license headers are included per-repository; see individual model cards for the specific training datasets and their own licenses.
KMOS, the Klondike Software Project's operating system effort, is a separate project under KSP and uses the separate Kunix License, not Apache 2.0. See kunix.org for KMOS specifically — NMAI's licensing does not apply there.
Links
- Models: huggingface.co/KSP-NMAI
- Klondike Software Project: kunix.org
Copyright 2026 Joseph Jones. Part of the Klondike Software Project.