AI assistant for companies: definition, use cases & how to choose | meinGPT
What an AI assistant is — clearly distinguished from a chatbot and an AI agent — which tasks it takes on per department, what a company assistant needs (own knowledge/RAG, roles & permissions, GDPR, EU operation) and how to create and choose your own AI assistant without programming. With an example prompt, a selection checklist, honest limits and an FAQ.
For Managing directors, IT and digital leads, and departments introducing AI assistants.
- Who it is for
- Managing directors, IT and digital leads, and departments introducing AI assistants
- Impact
- A general language model becomes a specialised, repeatable assistant for one concrete task — built by each department itself, without code
- Task
- Understand AI assistants, use them well and choose the right one
An AI assistant is a language model configured for one concrete task. Through an instruction (what it should do), its own knowledge base (which documents it can access) and a fixed tone of voice, it turns a general chat into a specialised, repeatably usable helper — for quotes, support answers, organising applications or reporting, for instance. For companies what matters is that an AI assistant does not just answer generically but accesses released company knowledge, is operated GDPR-compliantly in the EU and can be steered centrally through roles and permissions. The difference from a simple chatbot lies in the own knowledge and the repeatable configuration; the difference from an AI agent is that an agent additionally performs actions in connected systems.
From the task to productive AI use
An AI assistant is not programmed but assembled through an interface — in three layers. First the instruction: role, task, tone of voice and working steps are described in natural language (name, description, prompt and conversation starters). Second the knowledge: relevant documents, wikis or data sources are connected, so the assistant gives answers backed by sources from your own company context instead of general internet information — access only to released content on a least-privilege basis. Third the model choice: the right language model can be set per assistant (GPT, Claude or Gemini, for instance) without changing governance or the data-protection level. At meinGPT such an assistant is created through the Assistant Creator: you describe the task, choose icon, colour and model, define conversation starters and import the finished assistant into the test environment with one click. It is then tested against real examples and the instruction sharpened until answers are reliably right, shared across the team and managed centrally — without a release cycle. The effort invested once in a good assistant pays off with every repetition, because everyone works with the same quality-assured tool instead of inconsistent individual prompts. If you additionally need actions in internal systems (creating a record, say), you upgrade the assistant into an AI agent through connectors (MCP) and an API. The choice is therefore decided less by the chat window than by knowledge connection, governance and data protection.
- Who it is for
- Managing directors, IT and digital leads, and departments introducing AI assistants
- Impact
- A general language model becomes a specialised, repeatable assistant for one concrete task — built by each department itself, without code
- Task
- Understand AI assistants, use them well and choose the right one
What Whole company gets done with AI
Concrete, repeatable flows — from the first prompt to a dependable result.
Sales — quoting assistant
An assistant takes a short description of an enquiry, pulls the right text blocks and prices from the connected product catalogue and produces a structured quote with line items, totals and a covering letter. Because context, tone of voice and knowledge base are configured once, it delivers a consistent result for every enquiry — instead of inconsistent individual prompts. Professional sign-off of the quote stays with a human.
Customer service — answer assistant from the knowledge base
A support assistant answers enquiries on the basis of the connected knowledge base (help articles, manuals, FAQs) and backs its answers with the sources used. It only accesses released content and does not draw on general internet knowledge if the company does not want it to — so answers stay consistent and traceable.
HR — application and job-ad assistant
An HR assistant summarises applications along defined criteria, drafts job ads in the company's tone of voice and answers recurring questions about processes. The assistant encapsulates the criteria so everyone in the team works to the same documented standard — the decision about candidates is still made by a human.
Knowledge search — make internal documents searchable
A research assistant searches connected file shares, wikis and manuals and delivers answers backed by sources from your own company knowledge. Access follows the permissions granted on a least-privilege basis, so each person only sees what they are cleared for — traceably, via logs.
Marketing — content and translation assistant
A marketing assistant drafts posts, reworks copy in the defined brand voice and translates content GDPR-compliantly inside the platform. Because tone of voice and guidelines are stored in the assistant, drafts come out consistent and then get an editorial review and sign-off.
Meetings & reporting — summary assistant
An assistant summarises minutes, reports or long documents in a structured way, highlights tasks and decisions, and produces recurring reports from a template. Tasks that used to take hours become a matter of minutes — the substantive review stays with the owner. Because the template is stored in the assistant once, every report looks the same regardless of who triggers it.
IT & onboarding — process assistant
An assistant walks new joiners through internal procedures, answers recurring questions about tools, policies and processes from the connected knowledge base, and points to the responsible person where there are gaps instead of guessing. Scattered process knowledge becomes centrally and consistently available, without one person having to explain it verbally again and again.
A real prompt, a real answer
Nothing hidden — you see the input and the result before you sign up.
Create an AI assistant called "Customer Service Support Assistant": it answers customer enquiries exclusively on the basis of our connected knowledge base (help articles, manuals). Tone: friendly, precise, no jargon. If the answer is not in the knowledge base, it says so openly and suggests handing over to a human rather than guessing. At the end it names the sources it used.
| Building block | Content |
|---|---|
| Role | Support assistant for customer service |
| Knowledge base | Help articles, product manuals, FAQ |
| Tone of voice | Friendly, precise, no jargon |
| Behaviour on a knowledge gap | Name it openly + hand over to a human, do not guess |
| Output | Answer + sources used |
| Model | Selectable per task (e.g. GPT, Claude, Gemini) |
| Optional tier | As an AI agent, create a ticket in the support system (MCP/API) |
Put it to work in your own company
In a short live demo we show how this solution runs in your company with meinGPT, GDPR-compliant — using your own use cases.
Or download the Choosing the right AI platform & assistants — the requirements catalogue (PDF, German) as a PDF:
A work email is enough — processed in line with the GDPR.
Built for enterprise compliance
A company AI assistant should be operated in the EU, a data processing agreement (DPA) should be standard, and input must not be used to train the models. At meinGPT the operator is SelectCode GmbH, which is ISO 27001 certified and has its security reviewed regularly through independent penetration tests (most recently SySS, 2025). The assistant's access to knowledge sources follows the permissions granted on a least-privilege basis, is logged and subject to regular access reviews; managing, releasing and revoking assistants runs centrally through user and permission management with SSO. The information security management system governs access control, logging, the handling of personal data (PII) and the secure deletion of information no longer needed; the corresponding policies and evidence are available through the Trust Center. Control over data, models and permissions therefore stays inside the company — instead of scattered across private AI accounts (shadow AI).
- Own knowledge (RAG): Can the assistant access internal documents and data sources — filtered by permissions — instead of answering only from general model knowledge?
- Without programming: Can an assistant be assembled by departments through an interface (instruction, knowledge base, tone of voice) — without a development team?
- Model choice: Can a suitable model be chosen per assistant (e.g. GPT, Claude, Gemini, complemented by European and open-source models) — instead of lock-in to a single vendor?
- Data protection & hosting: Is the assistant operated in the EU, is there a data processing agreement (DPA) and is it assured that input is not used to train the models?
- Roles & permissions: Is there central user and permission management, SSO, least-privilege scopes and logged access?
- Shareability: Can an assistant be shared across the team, reused and adjusted centrally — or does it stay tied to one individual?
- Actions (agent tier): Can the assistant trigger defined actions in internal systems through MCP and an API when needed — or does it stay chat only?
- Independent evidence: Is there evidence such as ISO 27001 certification and regular penetration tests of the operator?
What this solution cannot (yet) do
Honesty is part of the solution. These limits are known — and therefore plannable.
An AI assistant is only as good as its instruction and its knowledge base — without clean context and maintained sources, answers stay generic.
The assistant does not replace professional review: results with legal or financial effect (quotes, contracts, assessments) must be approved before use.
By default an assistant provides information but performs no actions — only as an AI agent does it trigger defined operations in systems through MCP and an API, limited by the scopes granted.
Models can be wrong or out of date; safety- and legally-critical statements need source grounding and human control.
The value drawn from internal data depends on the connection and the permissions — without released, maintained sources the assistant answers only from general model knowledge.
Frequently asked questions
An AI assistant is a language model configured for one concrete task. Through an instruction (what it should do), its own knowledge base (which documents it can access) and a fixed tone of voice, a general chat becomes a specialised, repeatably usable helper — for quotes, support answers or reporting, for instance. In a company context it is characterised by accessing released company knowledge, being operated GDPR-compliantly and being steerable through roles and permissions.
Where to go from here
- meinGPT — Assistant Creator (build your own AI assistants without code)
- meinGPT — Trust Center (ISO 27001, security & data protection)
- meinGPT — Pricing & licence model
- Model Context Protocol (MCP) — specification
- European Commission — data protection (GDPR)
Last verified: 2026-07-03T00:00:00.000Z