AI agents

Agents that take on the repetitive work

An agent understands what you ask, looks things up in your tools and handles what your team does by hand today: reading, sorting, preparing and notifying. Here’s how they work, the kinds there are and how we put them to work in your business, always with a person in charge.

Tell us about your project See the kinds of agents

What an agent is

Software you hand work to.

An agent is an AI program that takes on a task and carries it out: it understands what needs doing, uses your tools to do it and tells you the result. Unlike a chat, it doesn’t stop at the answer: it finds the data, fills in the form or prepares the document for you.

01
Understands
Reads the request, the email or the document and knows what needs doing.
02
Looks it up
Finds what it needs in your software: the customer, the order, the invoice, the calendar.
03
Acts and reports
Prepares, records or notifies, and notes down what it did and why.

Kinds of agents

Two ways of working.

Depending on the task, the agent works beside you when you call it, or on its own in the background. Many businesses end up using both.

Local

Beside you, when you need it

You launch it from your own computer. It works with your files, your email, your chats and the web apps you already use, and bridges them with your services. You see what it does and decide at every step.

For example:

“Sort the invoices in this folder and send them to the accountant.”
“Reply to this email with the order status.”
“Copy the data from this spreadsheet into the supplier’s form.”

Remote

In the background, without you having to be there

It runs on a server, automatically: at a set time or as soon as something happens. It analyses the data, makes the routine decisions for you and keeps processes moving without getting stuck. It only calls you when a person is needed.

For example:

Every morning it reviews outstanding payments and prepares the reminders.
It sorts issues as they come in and notifies the right person.
It spots a stalled order and unblocks it, or tells you if it can’t.

Local and remote at a glance
	Local	Remote
Where it runs	On your computer	On a server, yours or ours
When it works	When you launch it	On its own: on schedule or when something happens
What it works with	Your files, email, chats and websites	Your systems, databases and APIs
Who decides	You, step by step	The agent on routine matters; you on what matters
Best for	One-off, varied tasks	Processes that repeat every day

AI models

The right brain for each task.

Every agent runs on an AI model, and not every task needs the most powerful one. Using a huge model to sort emails is like hiring an engineer to tidy the mailbox.

Frontier

They reason and plan long, difficult tasks. They’re the most expensive, so we keep them for when they’re truly needed.

For example: Claude Opus, OpenAI Astra

Mid-range

Nearly as capable for a good deal less. They power most agents day to day.

For example: Claude Sonnet, OpenAI Sol

Lightweight

Fast and cheap, for short tasks that repeat a lot.

For example: Claude Haiku, OpenAI Luna, Gemini Flash-Lite

Classification

They read something and decide what it is, instantly and at almost no cost. They don’t act alone: they’re combined with other models or programs.

For example: email type, urgency, language, spam

How we combine them

They work as a team.

One model is rarely enough. Here’s how the email that comes in every day gets shared out, for example.

Complexity: How much reasoning it takes to do it well.
Volume: Ten cases a day is not the same as ten thousand.
Cost: The cheapest model that does it well.
Speed: Whether someone is waiting for the answer.
Privacy: Open models on your servers, such as Qwen, GLM or Gemma, if the data can’t leave.

We choose the best model for each situation, and review it when a better or cheaper one comes out.

Connected with MCP

A standard plug for your systems.

MCP (Model Context Protocol) is an open standard for connecting agents to software. We build an MCP server for your ERP, CRM or database and, from then on, any agent, local or remote, can look things up and act on your real data.

One connection serves all your agents and assistants.
You choose what each one can read and what it can change.
If you switch assistants tomorrow, the connection still works.

What they can do

Repetitive work you can take off your plate.

Some tasks that suit an agent well. Yours may be different: we start from how your team works.

Admin and finance

Reading supplier invoices, preparing bookkeeping entries, reconciling bank transactions and chasing outstanding payments.

Customer service

Sorting emails and messages, drafting replies to the usual questions and passing the rest to a person.

Operations

Logging issues, ranking them by urgency and notifying the right supplier without waiting for someone to notice.

Sales

Keeping the CRM up to date, following up on quotes sent and preparing you for every meeting.

Documents

Filling in forms, generating contracts or reports from templates and filing everything in its place.

Reports and alerts

A Monday summary of what matters, and an alert as soon as something looks out of the ordinary.

Under control

A person in charge, always.

We automate what makes sense, not everything. And whatever is automated can be reviewed.

You set the limit

Together we decide what it does on its own, what it leaves ready for review and what it never touches.

Just enough access

Each agent can reach only what its task needs. Nothing more.

Everything is logged

Every lookup and every action is recorded: what it did, when and why.

Your data, where you decide

If you need it, the agent runs on your own servers. And we choose with you which AI services it uses and on what terms.

How we start

We start with one task.

1
We pick the task
The one that repeats the most and takes up the most time.
2
We run a pilot
A focused agent, with fixed scope and budget and a result you can measure.
3
We grow what works
With what we’ve learned, we take it to more tasks or more people on the team.

FAQ

Common questions.

Do I have to change my software?

Usually not. The agent can work with most software that allows connections: email, spreadsheets, an ERP or CRM with an API, web apps. Before we start we review yours and, if one doesn’t allow it, we find a way around it.

What if it gets something wrong?

AI models are more reliable all the time, and we still leave nothing to chance: we add checks that verify its work before it counts as done, we monitor and log every result, and for the most critical tasks it doesn’t act without your approval.

Do I need a local or a remote agent?

It depends on the task. If you launch it and it changes every time, local. If it repeats on its own every day, remote. We work it out together in the first conversation.

What about my data?

The agent only reaches what it needs, with permissions you control. If you need it, it runs on your servers, and we choose the AI services with you based on their privacy terms.

Which AI model do you use?

Whichever fits each task best. We work with several providers, such as Anthropic, OpenAI or Google, and with open models that can run on your servers. We choose by capability, cost, speed and privacy, and review the choice when a better or cheaper model comes out.

How much does it cost?

It depends on the task and the systems to connect. We start with a pilot with fixed scope and budget, so you know the cost before we begin.

What task would you give an agent?

Tell us what keeps repeating in your day-to-day and we’ll suggest where to start.

Tell us about your project