Case study · Private AI
A private AI chat for staff, inside a closed network
An AI chat for your staff, built for a special purpose: its modes, tools and assistants are configured around your business's own tasks. It runs on an isolated server inside a closed network — queries, documents and conversations never go to any outside AI service.
0
requests sent to outside AI services — everything stays inside your network
4 modes
from an instant reply to full research with search
Minutes
to put together a new assistant for a task — no developer needed
The challenge
Staff increasingly turn to public AI services to get work done — and along with their questions, they send off contracts, customer data and internal documents. Banning AI means losing speed. Allowing it without rules means losing control over your data.
- Commercial information, personal data and documents under NDA can't go out to external clouds
- A public chat doesn't know your company's specifics: every employee writes prompts from scratch, and answer quality varies
- Cloud AI bills grow right along with headcount and usage
- What's needed is AI that follows the company's own rules and never leaves its perimeter
How it works
- 1
An employee writes to the chat
in the browser, like a familiar chat app
- 2
Picks a mode or an assistant
a quick reply, a thorough review, full research, or a task-specific assistant
- 3
The assistant plans its answer
a separate step decides whether it needs tools: search, reading web pages
- 4
The model answers on your server
the reply streams in as it's generated, with no wait for the full text
- 5
Everything stays inside
conversation history and settings are stored in your own perimeter
An isolated AI server on your network
The model runs on a dedicated server with a GPU — in your own office or in a secured data center. The server isn't visible from the internet: only your chat reaches it, over a closed channel. Everything runs inside a closed network, and your data is never handed to anyone.
- Not a single request goes to ChatGPT, Gemini or any other outside AI service
- The model, conversation history and assistant settings are stored with you and nowhere else
- No per-request cloud billing: the server's capacity is yours
- Internet access exists only as a search tool, and only in the modes that actually need it
Special modes
Each mode is a separately configured assistant with its own instructions, depth of reasoning and set of tools. An employee just picks the one that fits the task. Modes are built around your business — separate ones for sales, support, legal or analytics, for example.
Quick
The reply starts almost immediately. For everyday questions, drafts and short lookups; the assistant reaches for tools on its own when it needs to.
Careful
Before answering, the assistant decides whether it needs fresh outside data, and if so, finds it itself.
Thorough
Takes longer to think and cross-checks several sources when one isn't enough. For hard questions and decisions that matter.
Research
Studies the internet before every answer, gathering and comparing sources. Takes a few minutes — but delivers a full brief in return.
TOOL mode: assistants for repeat tasks
An employee describes the rules once — the role, the reply format, the requirements — and gets a ready-made assistant for a task that repeats. Each message is processed against those rules on its own, unaffected by earlier ones, so the result stays predictable. The assistant gets the same tools and search as the chat, with depth chosen to fit the task.
Customer reply
A customer's question goes in — a courteous reply, on-brand and to company standards, comes out.
CRM entry
From a free-form email or message to tidy, structured fields for a customer record.
Contract check
Checks the text against an internal checklist and flags the clauses that need a closer look.
Meeting summary
From notes or a transcript to decisions, action items and owners.
Product description
From a spec sheet to catalog copy that matches the rest of the listing.
Glossary-based translation
Letters and documents in another language, using the terms your company already uses.
You can create as many of these assistants as you need — one for every department and every repeat task.
Assistant tools
- Web search
- Reading web pages
- Weather forecasts
- Your systems via API
Tools are connected to fit the task: knowledge bases, CRM, accounting systems. Everything runs through the same closed perimeter.
What the business gets
- Staff use AI every day — with no risk of data leaking to outside services
- Instead of a blank chat box, modes and assistants built around the company's own tasks
- A new task-specific assistant takes minutes to set up, with no developer involved
- Predictable costs: no per-request cloud billing
- New modes and tools get added as the work grows
Who it's for
- Legal and financial firms
- Healthcare
- Manufacturing and engineering
- Sales and support teams
- HR and staff training
- Anyone whose data falls under an NDA
Questions and answers
How is this different from ChatGPT?
The model runs on your own server, so queries and documents stay with you. And it's not a blank chat box: modes and assistants are configured around your company's own tasks.
What hardware do we need?
A server with a GPU in your own office, or a dedicated server in a data center. We size the configuration during the consultation, based on your tasks and headcount.
Can it connect to our own systems?
Yes. Assistant tools are connected to fit the task — knowledge bases, CRM, accounting systems — and run inside the same closed perimeter.
Who builds the task-specific assistants?
Your own staff: they describe the rules in plain language and pick a mode. We configure the modes that need special tools or scripted scenarios.
Which models does it use?
Current open models that run on your own hardware. We match the model to your tasks and your server's capacity.
Want the same for your business?
Send a request — in 30 minutes we'll walk through your process and show you what AI can take off your plate.
Free consultation