🚀 Novi AI

Novi AI is an experimental AI project focused on building small, efficient language models and exploring what can be achieved at the million-parameter scale.

🧠 Models

⭐ Novi-Micro

Novi-Micro is the latest released Novi base model and the next step after Novi-Nano.

With approximately 5.04 million parameters, Novi-Micro was trained from scratch on 1 billion tokens with a 2,048-token context window.

• 🧠 ~5.04M parameters
• 📚 1B training tokens
• 🔤 16,384-token vocabulary
• 📏 2,048-token context
• 🌐 FineWeb-HQ
• ⚡ GPU-accelerated training
• 🏗️ Custom BananaMind 2-style architecture
• 💻 Designed for lightweight experimentation

Novi-Micro is a base language model, intended primarily for research, experimentation, and further training rather than instruction-following.

🔹 Novi-Nano

Novi-Nano was the first released Novi model, introducing the project at the million-parameter scale.

• 🧠 ~1.26M parameters
• 📚 300M training tokens
• 🔤 8,192-token vocabulary
• 📏 256-token context
• 🌐 FineWeb-Edu, FineWeb-HQ, and SmolLM-Cosmopedia
• ⚡ GPU-accelerated training
• 💻 Designed for lightweight experimentation

Novi-Nano remains available as an earlier-generation experimental model.

🧪 Experimental Projects

• Novi-Code — Compact text + code model
• Yuki — Image-generation research
• Super-Novi — Intentionally chaotic experiment
• Hyper-Novi — Experimental model exploring unconventional training data

These projects are experimental and may change, be renamed, or never be released.

🛠️ Project Kairo

Project Kairo is the development codename for the Novi project.

It encompasses the development and experimentation behind the Novi model series, including Novi-Micro and future models.

📜 Project History

AppleMind

Before Novi, there was AppleMind — an earlier experimental language-model project.

AppleMind is now retired, with Novi becoming the primary project.

AppleMind → Novi AI → Novi-Nano → Novi-Micro 🚀

Small models. Big experiments.

Novi AI 2026 — Project Kairo