AI agent setup for small teams

Your own AI team.

Two to four agents, each with a fixed job and rules in plain text, running on your own hardware. Set up in three weeks, and then the team is yours.

Book a 30-minute intro call€4,900 net, fixed price. The intro call is free.

01The starting point

You use ChatGPT or Claude every day. Your knowledge still gets lost in chats, content happens when there is time, and everything operational runs across your desk.

What is missing is a division of labour with clear ownership. Each agent gets its own territory, a memory and one page of rules you can read and change yourself. Which roles your team gets, we decide together.

Built for owner-led service firms and teams of up to 20 people, in consulting, agencies, e-commerce or SaaS. For corporations with an IT approval process it is cut too small.

What comes in, such as emails, chats, notes and appointments, goes to your AI team on your hardware. That is two to four agents, each with its own rulebook. They write one daily report, you read it and approve. Sending, posting and deleting only happen with your OK.

02Possible roles

01

Captures what comes up in chats and notes and files it in order. Your knowledge then lives in one place.

02

Turns your work into drafts for posts and articles, every claim backed by a source. Nothing is published without your OK.

03

Writes you one file a day with what happened. One file instead of four chats.

04

Builds recurring workflows from your field. It starts with an appointment-request agent that takes a request from your inbox to your calendar.

03Run in-house

I sell the setup I run myself. My own team of four agents ran productively for five working days, and all I did was read the reports.

5/5productive days on which all four agents delivered
16work results in those five days
5daily reports, one file each
0interventions from me during the run

What went wrong in the pilot

  • One day was lost because the schedule only ran while a desktop app was open. The trigger now runs as a background service.
  • The report agent twice flagged a file as missing that was there. Fixed in version 2 of the report playbook.
  • On day one an agent broke off because it had started its own background process. It did not happen again for the rest of the pilot.

You get the full case study with all the numbers in the intro call.

04Three weeks

01

Week 1 · Scoping

A 90-minute discovery session. Afterwards the roles of your team are set, and each agent has a written rulebook covering its job, its data boundaries and its hard prohibitions.

02

Week 2 · Build

The agents run on your computer or server, with model fallback, playbooks and a daily routine. Sending, posting and deleting only happen with your approval, and a documented kill switch stops everything.

03

Week 2 · Starter kit

A tested appointment-request agent from inbox to calendar, with runbooks for deployment, testing, backup and updates. We adapt it to your first real use case.

04

Week 3 · Handover

A test run, a 60 to 90-minute handover session, a written operating guide and two weeks of support by email or chat. After that the team is yours.

05Price

Setup

€4,900one-off, net
  • Three-week fixed scope
  • Two to four agents with rulebooks
  • Starter kit with runbooks
  • Two weeks of support after handover

Half on order, half on handover.

Care

€490per month, net, optional
  • Model and rule maintenance
  • One iteration call per month
  • Cancel any month

For anyone who would rather not tune models and rules themselves after handover.

In operation the base models run on free tiers, and the framework needs no licence. More demanding tasks can run through a Claude subscription you already have.

All prices are net, plus VAT where applicable. This offer is for businesses only.

06Questions

A Mac or Linux computer, or a small server, with admin access. If you have none, I set up a small server in your name for about €10 a month. The contract and the data stay with you. On top of that, two 90-minute sessions and about 30 minutes for the test run.
The agents and everything they store live on your hardware, and I run nothing for you. What an agent may send to a language model is written in its rulebook. Which systems the agents can access, such as email or calendar, we decide together during discovery.
Hosting and operation by me, software development beyond the starter kit, data migration, legal or compliance opinions, and licences for third-party services. Further agents after handover are billed by effort or covered by Care.
The team keeps running on your side. Two weeks of support are included. After that you run it yourself or book Care for €490 a month, cancellable monthly.
Yes, in one move. The kill switch is documented, and every rulebook is plain text you can read and change yourself.
30 minutes, free of charge. Afterwards you know whether an agent team makes sense for you or whether a single good workflow is enough. If it fits, you get a fixed-price proposal with suggested roles.

07First step

The first step takes 30 minutes.

In the intro call we work out whether an agent team fits how you work. If it does not, I will tell you that too.

Book the intro callor write to hello@hungmai.io