Project Alpha
Your Private, Offline AI Platform
Project Alpha is a fully custom, high-performance AI system running on my local server. It combines multiple open-source Large Language Models (LLMs), Retrieval Augmented Generation (RAG), and secure access tools — all designed for privacy, control, and offline capability. Built as a functional proof of concept, it shows what's possible when you take AI out of the cloud and run it on your own hardware.
How It Works
I built the entire stack from the ground up: a powerful local AI server running Pop!_OS, Ollama, and OpenWebUI in isolated Docker containers. Models run locally on dual liquid-cooled RTX 4090 GPUs with fast storage, giving strong performance without relying on external services.
Everything stays containerized, so I can take the system fully offline while keeping full functionality. For controlled external access, I added Cloudflare protection, and a custom domain with encryption.
See my blog for the full build journey — Linux, bash scripting, multi-GPU setup, prompting, RAG, and more.
What You Get
- Model Selection & Switching: Choose from several base models and my personally tuned custom variants (including RAG-enhanced ones for higher accuracy and reduced hallucinations).
- Flexible Chats: Text or voice input, interactive mode, and the ability to bring in additional models mid-conversation for different perspectives.
- Advanced Tools: Document uploads, web search for fresh information, image generation via ComfyUI workflows, and code interpretation.
- RAG-Powered Responses: Some models can pull from your own documents or knowledge base for more relevant, trustworthy answers.
Trial users can access multiple base models plus several custom-tuned ones. Everything runs in a clean, GPT-like interface that's easy to use.
Why Build Custom Instead of Using Commercial AI?
Commercial platforms are powerful, but they come with trade-offs: ongoing changes that break custom workflows, privacy concerns, data usage policies, and dependency on internet connectivity.
If you need full control, true offline operation, or the ability to deeply tune models for your specific needs, a self-hosted setup like Project Alpha delivers where cloud services fall short.
Current Offering
Right now, Project Alpha is available for limited trial access to a small, trusted group for experimentation, peer feedback, and demonstration purposes. I'm also open to discussing consultation services for individuals or organizations who want to build their own secure, offline AI platforms.
I have no plans to compete with big commercial providers — their scale, speed, and resources are in a different league. Project Alpha exists to prove the concept, share the knowledge, and support real builders who value privacy and customization.
Ready to Try It?
If you're interested in trial access, custom model tuning, or consultation, head to the Contact page. We'll review terms, discuss your needs, and get you set up with a secure URL and temporary credentials.