Trust boundary
The model, vector database, documents, conversation logs and agent tools stay inside your company network. No data crosses the boundary. It also works on networks closed to the internet.
Service
Private AI
Every question your employees ask a public AI tool carries a piece of your company knowledge outside. We build the same experience inside your company. The model runs on your infrastructure; documents, questions, answers and logs never leave your company network.
External servicesdata doesn't go here
COMPANY NETWORK · TRUST BOUNDARY
Employees142 people
InterfaceChat
GatewayIdentity · Permissions · Logs
AgentPlan · Tools · Approval
Local modelOpen-source LLM
Knowledge baseDocuments · DB · API
D
What approvals are needed for a €40,000 purchase from a foreign supplier?
For foreign purchases over €25,000, approval is needed first from the budget owner, then from the finance director . Above €35,000, approval from the general manager is also required.
Purchasing Procedure · p.3
Signature Circular 2026 · p.1
AI that stays inside your company network
Four layers, adapted to every company.
We don't reinvent private AI for every client. We have a proven model and adapt it to your data, your permission structure and your workflow.
Knowledge
Procedures, contracts, reports and past work are connected to the model. Answers come with sources: which document, which page. It doesn't make up what it doesn't know; it says so.
Agent
The layer that gathers data, compares periods, spots deviations, writes the report and fills in approved templates. Actions such as sending, deleting and publishing stay subject to human approval.
Interface
Employees ask both general questions and internal business questions from one screen. The same chat interface they're used to; the only difference is that it knows your company.
Your data stays with you, and so does control.
Public AI tools are fast, but they don't know your company and they carry information outside. Private AI offers the same convenience under your company's rules, inside your own boundaries.
Public AI toolsPrivate AI
DataGoes to the provider's serversNever leaves your company network
Company knowledgeDoesn't know your documentsAnswers from your procedures and reports, with sources
CostGrows with users and usageInfrastructure is yours; unit cost drops as usage grows
Check-upThe provider changes the modelYou choose and pin the model
Access and logsNo visibility into who asked whatRole-based permissions, every action logged
EnvironmentRequires an internet connectionAlso runs on closed networks
Data protectionCross-border data transferData stays on your own servers
Let's work out together which tasks can move in-house. Free intro call
Which work does it take over?
These are tasks found in almost every company, regardless of sector. Below them are concrete examples from several sectors.
Operations and reporting
Periodic reports, comparisons, management summaries
HR and administration
Policy questions, standard correspondence, document preparation
Quality and compliance
Document search, action tracking, audit files
Training and onboarding
Q&A from procedures, training materials
Sales and proposals
Searching specifications, preparing proposals and presentations
Management decision support
One-page summaries from different sources
ManufacturingTrend summaries from shift reports, failure history and quality records
HealthcareProcedure and regulation questions, without patient data ever leaving
FinanceAudit file preparation, tracking regulatory changes
LogisticsShipment and volume analysis, periodic management reports
Public sector and defenseA knowledge assistant running on an air-gapped network
Metrics
Not by guesswork, by looking at the dashboard.
- Time savedMinutes cut from reporting, search and document preparation
- Automation rateShare of tasks the agent can prepare end to end
- Source citation rateHow many answers were based on a document
- Correction rateHow many outputs came back at human review
- AdoptionActive users and usage by department
What's delivered
- Mapping the data inventory and access policies
- Security architecture and defining the trust boundary
- Model comparison and selection
- Building the company knowledge base
- Employee chat interface
- A first task agent chosen together
- Role-based permissions and logging infrastructure
- Approval flow for critical actions
- Dashboard for usage, accuracy and time saved
- Team training and an admin guide
- Handover of source code, model and indexes
- 3 months of free bug fixes after launch
About private AI
How much infrastructure is needed?
It varies from company to company, so a single answer up front wouldn't be honest. What matters is how many people will use it, which model can do your work and how many documents need indexing. In the scoping phase we measure these and give you a concrete infrastructure and cost table. If you already have a suitable server, we build on it; new investment isn't a must.
Are open-source models as good as the big commercial ones?
For general chat and creative writing, the big commercial models are still ahead. But if what you need is answers drawn from your company documents and preparing specific reports, today's open-source models handle that more than well. Before deployment we compare candidates on your own questions and report the results.
What does it cost in the long run?
With public tools, every user and every question adds to the bill, and that line grows as your team grows. With private AI the infrastructure is set up once, and the cost per question drops as usage grows. In our call we put both calculations side by side for your user count.
What about data protection rules?
Because data never leaves your company network, the question of cross-border data transfer disappears. Access follows your existing corporate permissions and every access is logged, giving you a concrete record you can show in an audit.
Will it talk to our existing systems?
Yes, that's the core of the work. It connects in a controlled way to your ERP, accounting, document management, databases and file servers. You decide which system an agent can access and with what permissions.
You may also want to see
Let's map the scope first.
We work out together which data lives where and which work can be handed over. Then the timeline and price come to you in writing.