Find your fit

Which private AI is right for your business?

Four quick questions. No email required. We will point you to the tier that matches your size, your data, and your risk.

Read the fine print

“Private AI” vs. a routing gateway.

Some tools that promise you “own your AI” are really a gateway — a router that forwards your prompts out to OpenAI, Anthropic, or Google and bills you per token. Your data still leaves the building, the meter is still running, and if a provider revokes access — see Fable 5 — so are you.

A routing gateway

Your data leaves

Prompts are forwarded to third-party model providers. Metered per token. Runs on access that can be cut off.

Stavryn · private AI

Nothing leaves

The model runs on hardware you own, air-gapped, inside your network. Flat cost, no meter. Nobody can switch it off.

One test for any “private AI”: ask where the model runs and where your prompt goes. If the answer is “we route it to a provider,” it isn’t private — it’s just a nicer bill.
The models

Open models, sized to the job.

You run open-weight models on hardware you own — fine-tuned on your data, never shared. Here is the spectrum we deploy, from a lean everyday assistant to a sovereign-tier frontier model.

Qwen 3 · 32B
Parameters32B dense
Context128K tokens
Min hardware1 × H100 · 80 GB
Best forAssistant · drafting · RAG · support
Mistral Large 3 · 675B
Parameters675B MoE · 41B active
Context256K tokens
Min hardware4 × H200 · 564 GB (4-bit)
Best forCoding · analysis · Apache 2.0
GLM-5.2 · ~750B
Parameters~750B MoE · 40B active
Context200K tokens
Min hardware8 × H200 · 1.1 TB
Best forFrontier-grade reasoning
How many people will use it?
What kind of data and rules apply?
What will you mainly use it for?
Do you need it air-gapped?