Services
AI agents for businesses
AI agent development: agents that work inside your systems, with human approval and a full log.
An AI agent carries out tasks on its own inside your systems: it answers questions from your ERP and documents, keeps data up to date and creates drafts. We develop AI agents for businesses that only work with the permissions you grant. Anything that changes something is approved by a person, and every step is logged.
Typical triggers
Typical triggers
Clients get in touch when one of these situations arises.
- 01Staff gather information from the ERP, file storage and emails before they can reply to a customer.
- 02Your team already uses ChatGPT or Claude, but the assistant does not know your company data.
- 03Draft quotes, master data upkeep or follow-ups are left undone because nobody has time for them.
- 04A fixed workflow is not enough, because every request is a little different and involves several systems.
AI agents for businesses
What you get
A clear brief for the agent
We define which tasks the agent takes on, which systems it may read and where it only creates drafts.
Secure connection to your systems
ERP, CRM, file storage and mailbox are connected through dedicated interfaces with precisely defined permissions, on request via the open MCP standard.
Approval by a person
Anything that changes data, costs money or goes to customers is created by the agent as a draft. A person approves and sends it.
A log of every action
Every query and every change is recorded with time and trigger, so it can be traced later.
Testing with real tasks
Before go-live, we test the agent on real cases from your daily work. You see results, not promises.
Data protection concept
Set out in writing: model, processing region, retention, no use for training.
AI automation or AI agent?
AI automation fits if
- the workflow is always the same, for example for receipts, incoming mail or quotes
- there are many similar cases per month
- the result always ends up in the same system
An AI agent fits if
- every request is different and the way to the answer has to be worked out first
- information comes together from several systems and documents
- staff want to work in conversation with the assistant
- the agent prepares tasks that a person completes
MCP: the open standard for AI access to company systems
MCP (Model Context Protocol) is an open standard through which AI assistants such as Claude or ChatGPT access company systems. For each system, an MCP interface describes which tasks the assistant may carry out, for example «search customer» or «create draft quote». Everything else stays locked.
The benefit for you: your staff work in the assistant they already know, and you decide what it may see and do. If you switch AI models later, the interface stays in place.
We use this ourselves. Through our own MCP interfaces, AI assistants create draft quotes and invoices in our business software and maintain the content of this website as drafts. The interfaces only allow drafts. Finalising, sending and publishing are done by people only.
From our own practice: a draft quote via AI assistant
01
Request in the chat
We describe the customer and the service to the AI assistant in our own words.
02
Look up the data
Through the MCP interface, it finds the address, earlier documents and items in our business software.
03
Create the draft
It creates the quote as a draft. It cannot finalise or send it.
04
Review and send
A person reviews the draft in the software, adjusts it and sends it.
You decide what an AI agent may do
| Level | Example | Who decides |
|---|---|---|
| Read only | Look up stock levels, open items or contract details | the agent answers directly |
| Create drafts | Quote, reply to a customer, draft invoice | a person approves |
| Change data | Correct an address, enter an appointment, set a status | with or without approval according to your rules, always logged |
| Not provided | Trigger payments, delete documents, send to customers without approval | the interface does not offer it at all |
AI agents for businesses
Approach
01
Choose the task
Together we pick a task that comes up often, involves several systems and takes up time today.
02
Define permissions
For each system, we define what the agent may read, create as a draft or change.
03
Build the interfaces
We connect your ERP, file storage and mailbox through secure interfaces, via MCP on request.
04
Test with real tasks
Your team tests the agent on real cases. It only goes live once the results are right.
05
Operation with a log
Every action is logged. We evaluate where the agent helps and where it needs fine-tuning.
Effort
Price range
- Price range
- CHF 12'000 to 60'000
- Duration
- 4 to 12 weeks
Non-binding estimate, excl. VAT.
What drives the effort
An agent that answers questions from one system sits at the lower end. Each additional system, write permissions with approval levels and a dedicated user interface increase the effort. Ongoing model costs for most SMEs are in the low hundreds of francs per month.
Technologies
- Model Context Protocol (MCP)
- Claude
- Azure OpenAI
- n8n
- TypeScript
- C# / .NET
- PostgreSQL
AI agents for businesses
Frequently asked questions
What is the difference between an AI agent and a chatbot?
A chatbot answers questions, an AI agent gets tasks done. It looks things up in your systems, combines the results and creates, for example, a draft quote or a reply to a customer. What it may do is defined per system, and anything that changes something is approved by a person.
Can an AI agent change data in our ERP on its own?
Only if you allow it. By default, the agent reads and creates drafts that a person approves. For simple changes such as correcting an address, writing directly can make sense, always with a log. The agent cannot trigger payments or delete documents, because the interface does not offer these functions at all.
What is an MCP server and do we need one?
An MCP server is an interface based on the open Model Context Protocol standard. Through it, AI assistants such as Claude or ChatGPT access a company system in a controlled way, for example your ERP or document storage. You need one if your staff should work with company data inside the AI assistant. We develop MCP servers for existing systems and use our own every day.
Where does our data go when AI is involved?
We set that out in writing for each project. The default is processing in a European region, under contracts that rule out any use of your data for training. For sensitive data, self-hosted models are an option. Hosting and database are located in the EEA, and processing is designed around the GDPR and the Swiss Federal Act on Data Protection.
How do you stop the AI from making mistakes?
By never letting it decide alone where mistakes are expensive. Every extraction is given a confidence score, uncertain cases go to a person for approval, and every decision is logged. Before go-live we measure the accuracy rate on your real data.
Do you work with n8n or build your own software?
Both, depending on the case. n8n suits manageable workflows between existing systems, and your team can help maintain it. As soon as audit trails, permissions, high volumes or a dedicated user interface are needed, a purpose-built application is more robust. We often combine the two.
Let us talk about your case.
20 minutes are enough to find out whether a custom solution is worth it for you.