Custom AI agent development
for your business.

A colleague asks a question.
The answer is spread
across five places.

We build custom AI agents for research, customer questions and sales preparation. An agent can choose steps and use approved tools to complete a task. You decide what it can read, what it can do and when a person must approve.

Getting ready for a meeting or answering a customer often starts with a search: the CRM, an old email, a shared folder, someone else’s notes.

01 / Where it helps

Where could an AI agent
help your team?

Start with one recurring task. We test it with the documents, questions and languages your team actually uses.

01

Find an internal answer

Help a colleague find the relevant policy, procedure or project detail, with a link to the source.

If the information is missing or disagrees, the question goes to its owner.

02

Get ready to help a customer

Bring together earlier conversations, product details and account history, then prepare a suggested reply.

Your team reviews sensitive replies and any promises to the customer.

03

Prepare research

Gather the agreed sources, organise the findings and help update a working memo.

Your team’s conclusions stay separate from the AI’s research.

These are examples of possible projects. We agree the tools, access and checks with you before building.

02 / How it could work

“Can you help me
get ready for this meeting?”

Here is how an internal assistant could prepare a meeting brief. It uses the information and tools you have chosen for that job.

  1. 01

    Ask

    A colleague requests a brief for a meeting.

  2. 02

    Check

    Check which records that colleague can access.

  3. 03

    Find

    Gather relevant account details and notes.

  4. 04

    Prepare

    Show the brief with sources and open questions.

  5. 05

    Review

    Ask for any required approval and save the agreed result.

When it cannot find the answer, it should tell you.

Missing documents, conflicting information and broken connections need a clear explanation. The agent stops or asks the responsible person. It cannot give itself more access to finish the job.

Reading an email, drafting a reply and sending it are separate permissions. We agree them before connecting live tools.

03 / Planning the first version

How much should
the agent do for you?

We first check whether your existing software or a straightforward automation can do the job. Custom development makes sense when the task needs different steps, company information and tools that an existing solution cannot bring together.

Example permissions to agree together
TaskWho stays in control
Read informationUse the records available to that person and show the sources.
Prepare a draftLabel it as a draft, with the facts behind it.
Send a message or change a recordShow what will happen and ask for approval where needed.
Move money or grant accessKeep outside the agent’s default job, with separate authorization and checks.
05 / How we work together

What your first AI agent
project includes.

We agree what the first version should do, how to check it and who will look after it. Then we work through it with your team.

  1. 01

    A brief you can approve.

    We write down one task, its users, sources, connected tools and permitted actions. The brief names what stays outside the project and which decisions need a person.

  2. 02

    Examples and acceptance criteria.

    Together we choose everyday and difficult requests from your work. We agree how to judge useful answers, supporting sources, correct actions and cases the agent must pass to a person.

  3. 03

    A connected agent, tested with your team.

    We build the agreed task and check results against those examples. Tests include missing information, denied access, failed tools and misleading instructions. We verify saved results in the connected system.

  4. 04

    Documentation and an operating owner.

    You receive the build, setup documentation, approval rules and test findings. We show your team how to correct or pause the agent, and agree who monitors it and what support is included.

06 / Access and upkeep

Useful help,
with someone in charge.

Keep access tied to the person.

We agree which records each role can use and what happens when a document is removed or someone’s access changes.

Know where information goes.

We identify the AI providers, connected tools and storage used for documents, conversation history and activity records.

See what happened.

Keep a record of proposed actions, approvals and results, with a way to pause the agent or deal with a failed step.

Own the build and plan its upkeep.

You own what we build from day one. We agree who manages accounts, reviews answers and watches usage costs. Provider subscriptions and ongoing support are set out separately. Changes to models, instructions or sources need testing.

07 / Questions before you start

Questions about AI agents.

Do we need an agent or an automation?

If the steps are always the same, an automation may be enough. An agent can help when the next step depends on the question or the information it finds. We compare options around the job.

Can everyone see all our connected documents?

They should only see what their role allows. We agree and test access for each group, including what happens when permissions change.

Can it show where an answer came from?

We can build answers with source links and test whether those sources support the answer. We also check how it handles missing, conflicting or old information.

What will it ask us to approve?

We agree that with you. Reading, drafting, sending and changing records can each have different rules. A reviewer needs to see the proposed action before approving it.

How do you check that it stays within its job?

We try normal tasks, misleading documents, requests it should refuse and failed connections. Then we check what it actually did, including whether it respected access and approval rules.

Can it work in our languages?

We test German, French, Italian or English against the work you need it to do. The examples should reflect your team’s terminology and documents.

Where are our documents and conversations kept?

We explain which providers process them, where they are stored, who can access them and how long they are kept. We check the connected services as part of the same setup.

What does a first project cost?

First projects start from CHF 4,800 for one agreed task. Scope depends on source quality, integrations, permitted actions, languages and testing. Before building, the proposal confirms deliverables, timing and acceptance criteria. AI usage, hosting, subscriptions and ongoing support are shown separately where needed.

What would you like
your team to have help with?

Bring one recurring task, the tools and information it uses, and a few real examples. We’ll discuss what an agent could deliver, what needs approval and whether an existing tool would be enough.

Book a first call