Project Venus
Private AI for your business
Research · In development
Make the whole Twin yours.
Project Venus is our open weight model research initiative, which gives you sovereignty over your entire AI stack.
- Fast
- Draft model predicts multiple tokens ahead to increase speed
- Task-adapted
- Adapter trained on real task trajectories to optimise for correct outcomes
- All you can eat
- No metering on token usage. Use as much as you like.
- (Subject to concurrency thresholds)
Train on how the business works. Keep live facts in the Twin.
A useful answer depends on what your business means by revenue, a completed project or an overdue commitment. We are researching how a private model can learn those definitions and your policies, while current figures and source records stay in your Twin.
- 01
Run it privately, at speed
The private model already serves from dedicated UK or EU infrastructure. We measure and tune the whole serving path (hardware, checkpoint formats and speculative decoding) so private does not mean slow.
- 02
Teach it how your business works
Use reviewed examples to teach it your terminology, policies and when it should say it cannot answer, while current figures stay live in your Twin.
- 03
Evidence stays attached
Answers still show their sources, explain the calculations used and make missing evidence clear, whichever model or infrastructure answers.
Bring the model closer to your business.
Today your Twin can answer through an external provider, or through Venus, our private model already serving from dedicated UK or EU infrastructure. Both run under the same rules for sources and access, and the region is your choice.
A fully UK-hosted Twin.
The goal is to run your server, storage and private model together in the UK: one national boundary around the whole stack. Not yet a standard deployment.
What runs where: today and what we are building
Part
Today
In progress
Server and storage
TodayA dedicated server and storage for your Twin, in your chosen region.
In progressA UK datacentre for the server and its storage.
Your source systems
TodayYour Twin reads your existing systems without changing the original records. Keep managing those records where they are.
In progressUnchanged.
The model
TodayRequests go to the AI provider configured for your organisation, or to Venus serving privately from dedicated UK or EU infrastructure.
In progressA private model on UK GPUs, inside the same boundary, so prompts and business context never leave it.
Sources and checks
TodayAnswers from the checked Twin identify their sources and the version used. Questions it cannot support are refused.
In progressUnchanged, whatever model answers.
Is a private model right for you?
Three quick questions and your email: click your way through. We read every answer and will tell you honestly whether we think a private model fits.