Private ChatGPT for your business
Ask questions of your own documents.
With the source alongside.
For companies whose knowledge sits in manuals, contracts, procedures and old emails, and where colleagues ask each other the same questions all day. I build a private ChatGPT environment on those documents: it runs inside the EU, shows everyone only what that person may see, and names the document every answer comes from.
The knowledge is there, it just cannot be found.
Everyone asks the same colleague
How does that warranty arrangement work again, which version of the procedure applies now. One person knows, and that person is interrupted all day.
Searching SharePoint turns up nothing
The document exists, but it is called something else than you think and sits in a folder from three years ago. After ten minutes of searching you ask anyway.
Staff paste pieces into ChatGPT
It works, so it happens. Only nobody knows which contracts and client details now sit in a personal account.
Not everyone may see everything
Salary agreements, client contracts and board papers sit between the work instructions. An assistant that shows everything to everyone is not an option.
An answer without a source is worthless
If you cannot check where an answer comes from, you still have to look it up yourself. Then the AI has saved nothing.
No idea what it costs per month
A subscription per user is clear. Custom work with usage per question feels like an open-ended bill.
How I set up a private ChatGPT environment.
Decide sources and permissions
Which folders and systems go in, what stays out, and which role may see which documents. This conversation comes before the technology.
Connect to where the documents live
SharePoint, Google Drive or your own system. The documents stay where they are; the assistant reads along and keeps up when something changes.
Answers with a source
Every answer names the document and the passage it comes from, so you can check it. If it finds nothing, it says so.
Limits and tracking
A daily limit on usage, an overview of what it costs and a list of questions it could not answer. That last one is your list of missing documentation.
An AI knowledge base on your own documents
A private ChatGPT for your business is an assistant that searches your documents before it answers. It does not read the whole internet but your manuals, contracts, quotes and work instructions, and it only answers with what is in them.
That makes it suitable as an internal AI helpdesk: new staff finding their way, the sales desk checking what was agreed with a client, the shop floor looking up which procedure applies. If you want visitors to your website to get answers instead, an AI assistant for your website is the better fit.
Internal AI helpdesk
Questions about HR arrangements, IT, procedures and agreements, answered from the documents that cover them, with a link to the original.
Checking contracts and agreements
What was agreed with this client about delivery terms or warranty. The answer with the passage from the contract alongside.
Onboarding
A new colleague asks the assistant instead of the busiest person in the office, and gets the document where it is written down.
Technical documentation
Search manuals, specifications and fault procedures in plain language, even when you do not know what the document is called.
Where your data lives
The application, the database and the searchable copy of your documents run inside the EU, on hosting in your name. You get a data processing agreement.
For each question, only the question and the few passages needed for the answer go to the language model, at the provider you choose. That can be Anthropic, Google or OpenAI, or a European party if your sector requires it. Conversation content is not stored by default.
- Application and database inside the EU
- Only the passages needed go to the model
- Conversation content not stored by default
- Keys on the server, never in the browser
- Data processing agreement
- Code and hosting in your name
Who may see what
This is where most off-the-shelf solutions go wrong. For every document it is recorded who may see it: per role, per department or per document. When someone asks a question, the server only searches that user's documents, before anything goes to the language model.
That happens on the server, not in the prompt. An instruction like 'do not show this to staff' can be talked around; a document that is never fetched cannot. If the permissions are already right in SharePoint or Google Drive, I take them over where possible instead of setting everything up again.
Connecting ChatGPT to SharePoint, Google Drive or your own system
The documents stay where they are. The assistant reads them through the integration, keeps a searchable copy and updates it when a document changes or disappears. Your team does not have to move anything or work differently.
If part of the knowledge lives in a system of your own, such as an ERP, a CRM or a database of projects, I connect that the same way. What such an integration looks like is on system integrations. If the assistant should not only answer but also prepare something, such as a quote or an email, it becomes an AI agent.
The proof: the assistant in Avitly
There is no client case of exactly this shape yet, so I point to what I use daily myself. Avitly is my own platform for AV and IT installers. The assistant in it searches projects, quotes and equipment through tools, with the user's permissions, and only answers with what it finds.
The same rules apply here: look it up first, then answer, and say it does not know when there is nothing. On demos there is an assistant that shows how that works. It searches a database rather than documents, but the principle is the same.
What it costs, and what it costs per month
The build is a fixed amount you hear up front, excluding VAT. What that amount is depends mostly on the number of sources and how fine-grained the permissions are. How pricing works is on pricing.
On top of that there are running costs: language model usage, which you pay directly to the provider, and hosting for the application and database, which I pass on at cost. Usage depends on the number of questions and the model you choose, so I can only give an honest monthly figure once I know how many people will use it. There is always a daily limit, so it has a ceiling you set yourself. If you want to outsource the upkeep, that is possible through maintenance and support.
Knowledge base on one document source · € 2,950 to € 5,500
One SharePoint environment or Google Drive, permissions per role, answers with a source.
Knowledge base on several systems · € 4,500 to € 8,500
Documents from several sources, or data from an ERP or CRM alongside. First the integrations, then the assistant on top.
New application with AI built in · € 6,500 to € 12,000
The knowledge base is part of a larger system, with its own screens and processes around it.
Clients rate me 5 stars on average.
“I am extremely happy with the result of my new website for DoubleStreet Australian Cobberdogs, made by Webframer. From first contact, Webframer radiated expertise and creativity.”
“Webframer helped me enormously with the Veldvisuals website. It is not only good to spar with Bas about building the site, but also in the aftercare and whenever changes have to be made. Recommended!”
“Great to work with. Listens carefully to what we want. When we ask for changes, they are made expertly.”
“Really pleasant to work with and short lines of communication!”
“The website looks great!! Definitely recommended!”
“Excellent! Fast service and good contact!”
Translated from the Dutch originals on Google.
I am Bas Voets. Since 2021 I have been building websites, client portals and business software for small and mid-sized companies, from Berkel en Rodenrijs in the Netherlands. From first call to launch you talk to me, not to an account manager.
Are my documents safe?
What if it gives a wrong answer?
What does it cost per month?
How is this different from ChatGPT Team, Enterprise or Microsoft Copilot?
Do we have to move our documents?
Which AI model are we tied to?
You don't have to take my word for it.
Three pieces that show up in most websites and applications, as working examples instead of screenshots. Click into them, they actually run.
Price calculator
Visitor drags the size, picks options and sees what it costs right away.
Multi-step request
The same questions, but in three steps with a check per step. This one gets finished.
Dashboard
The numbers you now paste together from three systems, on one screen with filters.
One hour of looking before any building.
In the AI scan we spend an hour together on one process in your business. Afterwards you get a short plan: what AI gains you there, what it costs and how long building takes. Free, and you are not committed to anything.
Tell me what you want to build.
A half-hour call. You get an honest answer about what it costs, how long it takes and whether building it is smart at all.