What can private AI do for us?
Say thisEverything you'd use public AI for, on data you could never send to it: contracts, case files, citizen records, drawings, source code.
GovernmentPolicy and case searchAsk across policies, case files and decisions in local languages, with sources.
BankingCredit and KYC reviewSummarise files, flag missing documents, draft memos without data leaving the bank.
Energy and industryEngineering knowledgeSearch manuals, drawings and incident reports; help field teams find the answer fast.
Real estatePortfolio intelligence"Which leases expire in Q1 with a rent-review clause?" across thousands of contracts.
HealthcareClinical documentationDraft notes and summaries with patient data kept on site.
LegalMatter-safe researchReview and draft inside ethical walls, with an audit log of every question.
How long does setup take?
Say thisAbout 30 days from order to live on your own data. You see a working demo in the first meeting and a written sizing within days.
Day 0Unplug demo on sample data
Day 2Sizing and price in writing
Day 7Order and data plan agreed
Day 30Live on your data, first use case
Month 2We operate; your team learns
Month 6Hand-over, if you want it
What the customer provides: a rack or desk spot, power, a network segment, an admin, and the documents for the first use case. Larger rack builds take 60 days or more.
What does it cost?
Say thisA one-time appliance price, then a monthly plan if you want us to run it. Pilots start at a fixed promotional price, with the first month of operation free.
| Tier | Best for | Promotional price | Live in |
|---|
| Mini | One team, a first pilot | promo from $12,500 | ~30 days |
| Spark | A department's first production use case | promo from $24,500 | ~30 days |
| Studio | Larger models, several teams | promo from $29,000 | ~30 days |
| Station | Organisation-wide, many users | promo from $150,000 | ~30–45 days |
| Rack or hosted | Large or multi-site, or in a data centre you choose | by quote | 60+ days |
Promotional prices in USD, excluding taxes, shipping and integration. Hardware costs are changing, so every price is confirmed in the customer's written sizing. Partner pricing is shared under NDA.
Will it connect to our systems?
Say thisYes. It reads from the places your documents already live, signs people in with your existing accounts, and offers a standard API so your own apps can use it.
- Documents and data: file shares, SharePoint and OneDrive, document management systems, databases, and exports from ERPs such as Oracle and SAP
- Sign-in: your directory (Microsoft Entra ID or Active Directory, and other single sign-on), with each person's existing permissions respected
- Apps: an API in the format most AI tools already speak, so existing apps and agents can switch to private models
- Channels: web chat, Microsoft Teams, and WhatsApp for customer-facing use cases
Each connector is confirmed in the sizing. Unusual systems are handled by our integration team, usually through secure APIs.
Is it secure? Does any data leave?
Say thisNo data leaves. The models run inside your walls, and the appliance works with the internet unplugged. We wrap it in zero trust, so only the right people and devices can reach it.
- Runs on the customer's premises, or in an in-country facility they choose
- Its own network segment; every login checked by person, device and location
- Role-based access: people only see answers from documents they're allowed to open
- Every question and answer logged for audit
- Built and run by a SOC 2 Type II company; security documentation shared under NDA
Hardware is supplied subject to US export rules; we check eligibility for every order before it's confirmed.
Who maintains it after go-live?
Say thisWe do, unless you'd rather run it yourself. A monthly operate plan covers monitoring, patches, model updates and support, around the clock.
- 24/7 monitoring and support, follow-the-sun from Silicon Valley and Mumbai, with the local partner on site
- Security patches and software updates on a schedule the customer approves
- Backups of configuration and indexes; health reports every month
- Remote access only when the customer opens it, and logged; or fully on-site support through the partner
- Optional hand-over at month six, with training for the customer's team
What happens when models change?
Say thisYou're never locked in. New open models are tested in our lab, then rolled out to you when they're better. Outgrow the box, and you move up a tier.
- Open models only: Llama, Qwen, DeepSeek, Nemotron, Falcon and Jais today, swapped as better ones arrive
- Every new model is benchmarked on the customer's own questions before it replaces the old one
- Indexes and settings carry over; users see no break in service
- Move up a tier, or add a second appliance, as use grows
- Mix private and cloud models safely: sensitive questions stay private, the rest can go to approved cloud models