What a private AI assistant for business does
A private AI assistant is an internal knowledge base with a chat window in front of it. It reads your policies, contracts, product sheets, procedures and past proposals, and each answer links to the file it came from. New starters stop waiting on the one colleague who knows where everything is, and nobody digs through folders for the current version of a policy. Builds start from €10,000 ex VAT.
It is what most people mean when they search for a private ChatGPT for business, an AI knowledge base for business or a way to chat with your documents. Some owners call it a company brain.
We use the same principles inside our own studio every day. Rules sort client support email and leave anything uncertain for a person, and a daily AI summary of open projects is drawn from the project records, with a guard in the code that stops the model writing any figure. Running it ourselves is where our build rules come from. On the client side, our proof is the integration an assistant depends on: the IPAV member directory, on a platform Loco Digital designed and built, stays synchronised with the Institute’s CRM. That sync carries no AI.
For document processing, quoting and inbox triage, see our AI automation services, and for workflows between your systems, automating the process around the answers. The engineering sits under our AI integration services.
How the assistant finds the answer in your documents
Before launch, each document is split into short passages and indexed by meaning, so a question about “annual leave” also finds the paragraph headed “holiday entitlement”. When someone asks a question, the system picks the passages that match best and passes only those to the language model with an instruction to answer from them. Specialists call this retrieval-augmented generation, or RAG.
Because the model works from those passages alone, every answer can show its source: the file name, the passage used and a link to open it. When no passage matches, the assistant replies that it could not find an answer in your documents.
When ChatGPT Business, Copilot, Gemini or Claude is enough
For many firms a licence is the right first step. As checked on 30 September 2026, each of these works with company documents:
- Microsoft 365 Copilot. With the Copilot add-on licence, answers draw on your emails, files, meetings and chats through Microsoft Graph. Microsoft now calls the product Microsoft Copilot (Microsoft Learn).
- ChatGPT Business. Its company knowledge feature searches connected apps such as SharePoint, Google Drive and Slack, and cites the sources behind each answer (OpenAI Help Center).
- Gemini in Google Workspace. Answers from content held inside your Workspace (Google Workspace).
- Claude Team and Enterprise. Include Claude for Microsoft 365 and connectors to other workplace tools (Claude pricing).
All four are licensed per user per month, and their prices change often. Our comparison of ChatGPT Business, Copilot, Gemini and Claude for Irish businesses sets them side by side.
A build makes sense when a licence cannot reach what you need:
- your documents live somewhere the connectors cannot reach, such as practice-management software, a CRM or an ageing file server;
- answers must come only from an approved set, such as current policies and never drafts;
- the assistant belongs inside your own customer portal or app, used by people who have no Microsoft or Google licence;
- you want to choose the model provider and the processing region yourself.
An assistant that lives inside your customer portal or mobile app is built alongside it, as part of our app development in Dublin.
An AI consult compares the options against your own documents before anything is built.
Does a private AI assistant train on your data?
Not by default on the business plans of the main providers. Microsoft says prompts, responses and Microsoft Graph data are not used to train Copilot’s foundation models. OpenAI excludes ChatGPT Business, Enterprise and API data from training by default. Google says Workspace does not train on customer data without permission, and Anthropic gives the same default for its commercial products. Free and personal plans differ, so staff should use company accounts.
In a build, we choose a provider whose API terms exclude training and write those terms into the plan.
Can a private AI assistant run in the EU?
Yes, with the right plan and provider. For EU customers, Microsoft 365 Copilot is an EU Data Boundary service, although the Anthropic models offered inside Copilot currently sit outside that boundary (Microsoft Learn). OpenAI offers European data residency to eligible ChatGPT Enterprise, ChatGPT Edu and API customers (OpenAI).
For a build, EU processing is a written requirement. The plan names the model provider, the region where your data is processed and stored, and who can see it.
Access follows your SharePoint and Google Drive permissions
Microsoft says Copilot surfaces only data a user has at least view permission for. OpenAI says company knowledge respects the permissions in connected apps, and Google says Gemini retrieves only content the user can access.
In a build, staff sign in with their Microsoft or Google account, the assistant reads each file’s sharing settings from SharePoint or Google Drive, and a passage is used only if the person asking could open the file.
That makes permissions the first job. If a salary spreadsheet is shared with the whole company, any of these assistants will find it. We review the folders to be indexed with you and fix oversharing before launch.
How we build it
- One team and one document set first. Usually a policy library, a product catalogue or a set of procedures, with the questions that team asks most.
- A test set before launch. Questions from your own staff, each with an agreed right answer, measured before launch and after every change.
- Caps and logs. A monthly cap on model usage, and a log of each question, the passages used and the answer given.
- Yours to keep. You own the code and the index, and the model provider can be swapped without a rebuild.
Live data from a CRM or accounts package is read through that system’s API, the work covered by CRM integration in Ireland and accounting automation.
