Use cases
Internal AI agent: ask your own systems in plain language
Order data sits in your ERP, customer history in your CRM and complaints in your service desk. An internal AI agent answers questions across all of them and shows each employee only what they are allowed to see.
Short answer
An internal AI agent lets employees ask questions in plain language to your own business systems, such as ERP, CRM and service desk. BeeManaged builds it in app.beemanaged.com: read-only by default, rights per role, changes only after approval and an audit log for every question. It is custom work, priced in consultation, and it runs in the EU.
What is an internal AI agent?
An internal AI agent is an AI assistant for your staff, connected to your business systems. Employees ask a question in plain language; the agent looks up the data in the systems it may read and answers with the source. Unlike a customer-facing agent on your website, it serves employees only.
It is also known as an enterprise AI assistant or AI knowledge base, although it reads current data from your systems as well as documents. As with every BeeManaged agent, the agent prepares and your team decides.
Why can't ChatGPT answer questions about our own orders?
Because it cannot see them. ChatGPT Business and Enterprise only know your company data through apps you connect, and the built-in AI of your ERP or CRM only looks inside its own package. In a manufacturing company the answer to one question is often spread over three or four systems.
A sales rep asks: how are things with customer X? The orders are in the ERP, the visit report in the CRM, an open complaint in the service desk and the drawing in SharePoint. Today that means four screens or a call to the office.
Connecting is not a given either. As of September 2026, Exact has an official MCP server on its roadmap without a date, and AFAS has no official MCP server.
Sources
- Exact: roadmap preview (Dutch) (exact.com)
- AFAS: AI integration (Dutch) (klant.afas.nl)
How does an internal AI agent work?
Employees type a question in app.beemanaged.com. The agent checks their role, reads the connected systems it may read and names the source of each answer. It is read-only by default: changes go through an approval queue, and every question and action is recorded in an audit log.
01
Ask in plain language
About an order, customer or product, in app.beemanaged.com.
02
Rights are checked
The agent only uses data that fits the employee's role. A service engineer sees tickets, not margins.
03
Systems are read
Through their official API, with read access only by default.
04
Answer with source
Each answer names the system the data came from, so it can be checked there.
05
Changes need approval
A proposed change, such as updating a ticket, waits in the approval queue until a person approves it. Everything is logged.
What can employees ask an internal AI agent?
Anything that is in a connected system and fits their role. In a manufacturing company that usually means order status, delivery times, customer history, open tickets and simple management questions. The table shows typical questions per department and the systems the answer comes from.
Answers depend on which systems are connected and what their API exposes.
| Department | Example question | Answer comes from |
|---|---|---|
| Sales | What did customer X order this year, and is there an open quote? | ERP and CRM |
| Inside sales | Which orders for customer X are still open, and what is the delivery time? | ERP |
| Service | Which tickets this week are about product Y? | Service desk |
| Planning | Which orders due next week are still waiting for material? | ERP |
| Management | How many brochure requests came in per country this quarter? | Conversation data and CRM |
Which systems can an internal AI agent connect to?
An internal agent can connect to any system with a usable API, such as Exact Online, AFAS Profit, Microsoft Dynamics, Mendix, TOPdesk, HubSpot, Salesforce and Microsoft 365 including SharePoint, each through its official API.
HubSpot and Salesforce also offer an official MCP server as of September 2026. That makes connecting easier, but MCP does not set rights across systems or approval of write actions.
Sources
- HubSpot: remote MCP server generally available (developers.hubspot.com)
- Salesforce: hosted MCP servers generally available (developer.salesforce.com)
Is an internal AI agent safe for company data?
It is when access is limited and traceable. The agent and your data run in the EU, models are used under business terms and your data is not used for training. The agent gets only the rights it needs, reads by default and changes nothing without approval. A data processing agreement is standard.
Vendors warn about the uncontrolled alternative. AFAS calls uncertified MCP servers and AI agents uncontrolled and risky, and advises minimal token rights, limited token validity, EU processing and a data processing agreement. The MCP specification warns against over-broad scopes. When employees switch on their own connectors, it is often unclear who checks those settings.
- Least privilege: a dedicated account per system, read-only by default, and rights per role.
- An approval queue for write actions and an audit log per question and action.
Sources
- AFAS: AI integration and its advice (Dutch) (klant.afas.nl)
- MCP specification: security best practices (modelcontextprotocol.io)
Internal AI agent or ChatGPT Business, Enterprise or Microsoft 365 Copilot?
ChatGPT Business and Microsoft 365 Copilot are general assistants with a licence per user, strong for writing, summarising and searching connected sources. A managed internal agent from BeeManaged is built around your own systems, with one role model across them, write actions through approval and ongoing management.
Choose ChatGPT Business if your team mainly needs help with writing and research and your key systems already have a working app. It is cheaper and quicker to start. OpenAI does not train on business data by default; data residency is limited to Enterprise and Edu.
Choose Microsoft 365 Copilot if most knowledge sits in Outlook, SharePoint and Word and your IT team wants to build connectors or agents in Copilot Studio.
Choose a managed internal agent if answers sit in ERP, CRM and service desk together, changes need approval and you would rather not maintain it yourself.
| ChatGPT Business / Enterprise | Microsoft 365 Copilot | Managed internal agent from BeeManaged | |
|---|---|---|---|
| Which data it reaches | Connected apps and custom MCP apps | Microsoft 365; other systems via Copilot connectors or agents (Copilot Studio) | Your ERP, CRM, service desk and documents, plus conversation data |
| Rights across systems | Respects each connected source's permissions | Set per connector for data outside Microsoft 365 | One role model across all systems, read-only by default |
| Write actions | Via custom MCP apps, in beta | Federated connectors read-only by default; otherwise build agents | After approval in the approval queue |
| Management | Your own team | Your own IT team | BeeManaged; you decide on roles and approvals |
| Pricing model | Per user per month: Business USD 20 (annual, minimum 2 users); Enterprise on request | Per user per month: USD 30 (annual); Copilot Business USD 21 up to 300 users; Copilot Credits for advanced agents from October 2026 | Custom work, priced in consultation; monitoring and dashboard €250 per month |
Sources
- OpenAI: ChatGPT Business and Enterprise pricing (openai.com)
- OpenAI: company knowledge in ChatGPT Business, Enterprise and Edu (help.openai.com)
- OpenAI: enterprise privacy (openai.com)
- Microsoft: Microsoft 365 Copilot pricing (microsoft.com)
- Microsoft: Copilot connectors overview (learn.microsoft.com)
Internal agent in the Unica pilot
In the Unica pilot we also explored an internal agent.
What does an internal AI agent cost?
An internal agent is custom work, priced in consultation. The price depends on which systems we connect, which roles need access and whether write actions are needed. Monitoring and the dashboard in app.beemanaged.com cost €250 per month. We agree the price with you before we build anything.
One conversation is usually enough to tell whether an internal agent is worth it for you, or whether ChatGPT Business or Copilot will do.
What are process dashboards in app.beemanaged.com?
Process dashboards combine conversation data from your agents with data from your connected systems. They are live in app.beemanaged.com, next to the internal agent, conversation monitoring and the monthly report, and show how requests and tickets develop across both sources.
Updated · As of September 2026 · Author Peter Davelaar, AI marketing and business intelligence
Frequently asked questions
What is the difference with ChatGPT Business or Enterprise?
ChatGPT Business and Enterprise are general assistants, licensed per user, that know your data only through connected apps; write actions are in beta as of September 2026. A BeeManaged internal agent is built on your own systems, with rights per role, approval for changes and ongoing management.
Which systems does it work with?
It works with Exact Online, AFAS Profit, Microsoft Dynamics, Mendix, TOPdesk, HubSpot, Salesforce and Microsoft 365 including SharePoint, each through its official API. Other systems with a usable API we assess together.
Do employees only see what they are allowed to see?
Yes. Rights are set per role and the agent only reads data that fits that role, with read-only access by default. Every question is recorded in the audit log, so you can check who asked what.
Can the agent change something in our systems?
Only after approval. The agent can prepare a change, such as creating a task, but it waits in the approval queue until a person approves it. Otherwise the agent only reads.
Where is our data stored?
In the EU. Models are used under business terms, your data is not used for training and a data processing agreement is standard.
What does it cost?
An internal agent is custom work, priced in consultation, because it depends on your systems and roles. Monitoring and the dashboard cost €250 per month.
Find out which questions your systems can answer
Tell us which systems you use and which questions cost your team the most time. We will tell you whether an internal agent makes sense, or whether ChatGPT Business or Copilot is enough.