Private AI Platform
Ask your data in plain language. It never leaves your building.
We design and implement private AI platforms on your own infrastructure. Your people get answers from company data and documents without writing SQL or digging through folders. Your developers get a coding assistant. Your security team gets the access control and audit trail they will ask about first.
- Runs on your infrastructure
- Your data stays inside
- No per-seat licences
AI the way most companies get it
- Staff paste internal data into public chatbots anyway
- Reports wait in a queue for the one person who knows SQL
- Text-to-SQL demos impress, then return confident wrong numbers
- Nobody can say who asked what, or which data an answer came from
- Per-seat licences grow faster than adoption
A platform you own
- Models run on your infrastructure. Prompts and data stay inside your network
- People ask in plain language and get a clear answer, a table or a chart
- Answers come from definitions your data owners approved, not improvised SQL
- Every answer shows where it came from
- Open-source foundations with no per-seat licence fees
One platform, many uses
A private AI foundation your company owns. Start with one use case and add the next on the same base.
Answers from your data
- Plain-language questions over your business systems
- KPIs, trends and comparisons from approved definitions
- Answers as tables and charts you can export and share
- Fresh data on a schedule you set
Answers from your documents
- Policies, manuals, contracts, specifications and wikis
- Every answer links to the passages it used
- People only see documents they already have access to
- Kept current as documents change
A coding assistant for your team
- Write, explain, document and review code
- Works in the editors your developers already use
- Your source code never leaves your network
- Optional awareness of your own repositories
AI inside your own tools
- One internal endpoint for every app that needs AI
- Budgets and limits per team or application
- Every request logged in one place
- Swap or upgrade models without touching your apps
How a question becomes a number
The model interprets. Governed code does everything else.
- Deterministic code
- AI, with a narrow job
- People
You
Ask a question
"What were our ten best-selling products by margin last month?" You are signed in with your normal company account.
AI model
Map it to approved definitions
The model translates your words into metrics, dimensions and filters from a catalogue your data owners approved. It never writes SQL.
Governed code
Enforce the rules and run it
The analytics layer checks your permissions, applies the data rules for your role and runs a read-only query against the right source.
Governed code
Return the answer with evidence
You get the result plus the definitions, sources and data freshness behind it. The audit record is written before the answer reaches you.
The design rules we build by
Every platform we build follows these rules. Your security team can hold us to them.
Your data stays home
Models, prompts, answers and logs stay inside your network. If your environment is isolated from the internet, the platform can run that way too.
The model chooses. Code executes.
The AI picks from definitions your data owners approved. Governed code builds and runs the actual query.
Permissions follow the person
Your identity travels with every request. The AI never sees more than the person asking would see.
Every answer is traceable
Sources, definitions and data version are recorded with each answer. You can always explain where a number came from.
Models chosen by test, not by brand
Open models improve every few months. We pick them by testing candidates on your questions, in your languages, on your hardware. Upgrading later means re-running the same test.
Sized to you, not to a brochure
We start with the simplest setup that does the job well. Capacity and complexity grow only when your usage needs them.
From first call to everyday use
Start small, prove it on real questions, then grow
Discovery
We map your systems, data, identity setup and infrastructure, and agree the first use cases with the people who will use them. Together we write a set of real questions with agreed answers.
Pilot
A first version for one team and one use case, checked against those reference questions. You see real answers on your own data before committing to more.
Roll out
More teams, sources and use cases, plus monitoring, tested backups, security documentation and training for users and administrators.
Operate and extend
Your IT team runs the platform after handover. We stay on for support, model upgrades and new use cases when you need them.
Every organisation is different. We agree a realistic plan after discovery, once we know your data and infrastructure.
Is this right for you?
A good fit if
- Your data can't go to public AI services because of confidentiality, regulation, IP or an isolated network
- Your knowledge lives in databases, business systems and documents nobody can search properly
- You want your own IT team to own the platform after handover
- You'd rather invest once in a platform than pay forever per seat
Probably not a fit if
- Cloud AI is allowed for your data. A SaaS tool may be cheaper, and we will tell you so
- You want a black-box product with a vendor roadmap. This is a platform you own
- Nobody has defined your key metrics yet. That is the real first project, and we can help with it
Our acceptance bar
Before go-live we agree a reference set of real business questions with your data owners, each with an expected answer. The platform passes only when it answers them correctly or refuses explicitly. A confident wrong answer counts as a failure, whatever the overall score. These are the terms we sign up to.
Frequently asked questions
Does any of our data leave our network?
No. The standard design runs entirely on your infrastructure, so models, prompts, answers and logs stay inside your network. It can also run in an environment with no internet access at all. Calling an external model is possible for less sensitive use cases, but only if you choose it and your data policy allows it.
Which AI models do you use?
Open-weight models with permissive licences that are free for commercial use. We don’t lock the platform to a brand, because the best open model changes every few months. We define the capability each job needs, then test candidates on your own questions, in your languages, on your hardware. The winner goes into production. Upgrading later means re-running the same test.
How accurate are the answers?
Accuracy comes from design, not from a bigger model. The AI never writes raw SQL against your databases, which is where most text-to-SQL projects fail. It maps your question onto definitions your data owners approved, and governed code builds the query. Before go-live we test against a reference set of real questions with agreed answers. When the platform can’t answer reliably, it says so instead of guessing.
What hardware do we need?
It depends on how many people use it, for which jobs, and which models perform best on your questions. That can mean anything from a single GPU server for one department to a larger setup for company-wide use. We deliver a full sizing covering GPU, CPU, memory, storage and network. We can work with hardware you already own, or help you buy the right hardware.
Can it use our existing data warehouse or BI model?
Yes, and that’s the best case. If you already have a data warehouse, a BI model or a KPI catalogue, we build on it. That reduces the modelling work significantly. If you don’t, we set up a small analytical store that refreshes from your source systems.
How do you handle access control?
Users sign in with your existing identity provider, so roles and groups come from where you already manage them. The user’s identity travels with every request, and data and document permissions are applied by code before anything is returned. The AI never has more access than the person asking.
What does it cost?
The software foundations are open source, so there are no per-seat licence fees. You pay for implementation, any custom development, and optional support and maintenance. After discovery we quote clear phases with the assumptions written down. Hardware is separate, if you need new hardware at all.
Who runs the platform after go-live?
Your IT team, with full documentation, administrator training and a proper handover. We offer ongoing support for updates, model upgrades and new use cases. You can run it without us. When you want the next use case, the people who built the platform are one call away.
Want AI on your own data, without your data leaving?
Tell us where your data lives and who needs answers from it. We'll tell you honestly whether a private AI platform makes sense for you, what it would take and roughly what it would cost.
Discuss your AI platformFree 30-minute call · No sales pitch · Honest feasibility check