Local Model Stack
Pre-configured open-source models ready to run: Llama 3, Mistral, and Phi-3 variants selected for your GPU's VRAM. Swap models from Hugging Face at any time.
Software · beta
A private AI workstation
on your own hardware.
Delivered on a single mid-range GPU, fully offline-capable from day one. When we configure Sovereign AI for your team, you get the model stack, the UI, and a documented configuration you can fork.
Sovereign AI turns a single mid-range GPU into a private AI workstation: local models, a private web UI, and tool access — all running on your hardware, offline by default. Configured, documented, and handed over by 3D3D.
What is Sovereign AI?
A self-hosted stack that turns one mid-range GPU into a private assistant: local models, web UI, and tool access, configured and handed over.
beta Working and evolving. Real users, honest rough edges. Email for access.
Open Sovereign AI What do you get with Sovereign AI?
A named system of 4 components — not a list of features, a description of what the software actually does.
Private Web UI
A clean chat interface that runs entirely on your machine. No requests leave your network unless you explicitly enable a tool that reaches outside.
Agent Tool Access
Browse, search, file read and write, and code execution — all scoped to your machine and audited through ORA's governance layer if you choose.
Documented Configuration
Every configuration decision is documented in plain language. Hand it to your IT team, modify it yourself, or run it exactly as delivered.
Who built this, and why does it work?
Delivered to clients on a single mid-range NVIDIA GPU. Fully offline-capable from the first boot. The configuration is documented so any IT person can understand and modify it.
— Randall Marshall · 3D3D · New Brunswick, Canada · info@3d3d.ca
How much does it cost to build something like Sovereign AI?
Three fixed tiers for scoped work. All include full source handover, your hosting, your keys — no ongoing seat licence required.
Starter build
$1,500
One-time fixed-scope deliverable. Best for single-feature tools, internal utilities, and clearly defined automation.
Mid-scope build
$2,500
Discovery, design, build, and handover. Best for client-facing apps and systems that need a defined user experience.
Full custom build
$5,000
Complete scope, full source, and production deployment. Best for commercial apps and anything shipping to paying users.
All prices CAD. We scope before we quote — describe the problem at info@3d3d.ca and we will respond within 24 hours.
Questions about Sovereign AI.
What hardware does Sovereign AI require?
A single mid-range NVIDIA GPU — RTX 3060 or better — is the minimum. We configure the model stack to match your GPU's VRAM and document every choice so you can modify it.
Can Sovereign AI run without internet access?
Yes. Once set up, the stack runs fully offline. External network access is disabled by default and enabled only for tools you explicitly configure and approve.
What models are included?
Configuration includes Llama 3, Mistral, and Phi-3 variants selected for your hardware. You can add or swap models from Hugging Face at any time — the stack is not locked to any specific model provider.
What does a Sovereign AI setup cost?
Setup starts at $1,500 for a single-user workstation. Multi-user enterprise deployments are $3,500. Monthly care plans — monitoring, updates, and model management — start at $99/mo.
Owned software
Want one of these for your business?
Describe the problem. We design it, build it, and hand over the keys. No seat licences, no lock-in — you own what we build.