Forge Your AI. Your Way.

Modular AI ecosystem with specialized expert models

Why use one bloated 31GB generalist when you can forge a personalized AI from specialized 3GB experts? Choose language-specific code models, creative models, gaming assistants—download only what you need. Start at 6GB. Build your perfect AI toolkit.

Start Training View Projects

Why Ash Forge?

🎯 Specialized Models

Expert models for specific domains: Python, Rust, JavaScript, creative writing, gaming, and more. Each optimized for its task.

🔥 CPU Training

Train models on CPU with our optimized LoRA fine-tuning pipeline. No expensive GPUs required. One-command setup on Linux or Windows.

⚡ Lightweight

3GB expert models beat 31GB generalists at their specialty. Start with 6GB minimum, add models as needed. Your hardware, your rules.

🔓 Fully Open

Apache 2.0 licensed. Fork it, modify it, train your own models. Community-driven model marketplace coming soon.

🚀 Production Ready

C++ inference engine (ash.cpp), Redis job queue, PostgreSQL tracking, systemd services. Built for real workloads.

🤝 Community Driven

Join our Discord to contribute models, datasets, and feedback. Build the AI ecosystem you want to see.

50+

Planned Models

3GB

Per Model

100%

Open Source

0$

API Costs

Get Started

1. Install Infrastructure

curl -sSL https://github.com/ash-forge/forge-train/raw/main/setup.sh | sudo bash

One command installs everything on Ubuntu 26.04 LTS.

2. Submit Training Job

forge-train submit --model ash-code:python --dataset datasets/python-expert

Queue your model for training (48-72 hours on CPU).

3. Use Your Model

ollama run ashforge/ash-code:python

Deploy locally or integrate into your applications.