
AI agents that work inside your systems
An agent is not a chat window. It knows your data, follows your rules and works where your team already works: in the ERP, in the CRM, in SharePoint, in the mailbox. And someone on your side decides what it is allowed to do.
- Level 1: Personal assistance
- Level 2: Domain agents
- Level 3: Process agents
What an AI agent takes on — and what it does not
Not every company has to start at level 3. Most begin at level 1 or 2, and that is exactly right.
Personal assistance
Helps with writing and searching — your team leads
- Preparing and summarising meetings
- Researching across approved documents
- Drafting proposals, emails and reports
- Supporting engineering and documentation
Domain agents
Know one domain and work on your data
- Sales agent: CRM, references, proposal history
- Service agent: documentation, tickets, fault analysis
- Back-office agent: policies and forms
- Engineering agent: repositories and architecture
Process agents
Take over individual steps — with approval by a person
- Checking and routing incoming documents
- Pre-qualifying enquiries
- Sorting tickets and proposing solutions
- Checking data and flagging what needs attention
At every level the same rule applies: people decide.
What an agent may do on its own is your call — not ours, and not the model’s.
Five agent packages from our own business — and yours
We built these agents and run them every day in our own house. That is why the building blocks already exist: the connectors, the permissions model, the approval logic, the reporting. What differs in your company are the data sources, the vocabulary and the approval rules — that is what we work on with you.
Sales agent
Prepares customer meetings, finds matching references and products, summarises conversations, drafts follow-ups. Reads your CRM, proposal history and product material — writes nothing without your approval.
More about this agentService agent
Answers questions about technical documentation, helps with fault analysis, pre-sorts tickets and suggests solutions. Particularly valuable where service knowledge sits with a few experienced people.
More about this agentKnowledge agent
Makes company knowledge from SharePoint, file shares, email and your DMS findable — with a source for every answer. Everyone only sees what they are already allowed to see.
More about this agentQuotation agent
Drafts quotations from templates, previous quotes and price lists. Sales reviews and approves. That can cut the time from conversation to a quote ready to send from days to hours.
More about this agentManagement assistant
A daily briefing on numbers, appointments and open decisions. Pulls together from ERP, calendar and mailbox what belongs on the desk in the morning.
Custom agents
Engineering, procurement or process agents for workflows that exist only in your company. Not a package, but built on the same foundations — and therefore faster than starting from zero.
How an AI agent comes about in your company
- 01
Sharpen the task
What should the agent deliver, for whom, in which situation? And just as important: what must it never do?
- 02
Connect the data
We establish which sources are authoritative, how they can be reached and which permissions apply. Two or three good sources usually beat ten mediocre ones.
- 03
Prototype
A working agent on real data, anonymised where necessary. Your team tests it against actual cases, not invented ones.
- 04
Put it into production
Roles, approvals, logging, monitoring. From here it is no longer an experiment but a tool in daily use.
- 05
Hand it over
Your AI people learn to maintain the agent, take in feedback and make small adjustments themselves.
What companies ask us about AI agents
What is the difference between an AI agent and ChatGPT?
ChatGPT is a chat window: it knows what was on the internet, but nothing about your company, and it only works when someone types something in. An agent knows your data, works inside your systems — ERP, CRM, SharePoint, mailbox — and carries out a defined task, such as drafting a quote from an inquiry. What it may and may not do is your call.
How long does it take to get the first agent?
That depends mainly on how accessible your data is and how many systems the agent needs. You get a reliable estimate after the workshop — by then we know which data is needed and where it lives.
What may an agent do on its own?
Whatever you allow it to. To start with, we recommend: the agent prepares, a person decides. Quotes, customer emails and system entries only go out after approval. Where an agent may take on more later is your decision, once it has proven itself in daily work.
Do we need our own AI specialists to run an agent?
No. We build the agent, put it into operation with you and show someone on your team how to maintain it and make small adjustments themselves. We are still there for the hard cases.
Which agent fits your company?
That is best settled against a concrete example from your business — starting with a free 30-minute call. And if your data cannot go to the cloud, the same agent runs on your own servers.