I launched an AI agent on a Raspberry mini computer worth $15. People bought Mac Minis for the same job.

The photo shows the board plugged into my MacBook. A green LED is on. The terminal shows the agent's command line interface with a Telegram channel configured and a skills command.

No local models

To be clear, the people with Mac Minis did not run local models either. They configured everything on a cloud API: Claude, ChatGPT.

What this means

An agent that calls a cloud model API does the heavy work on the provider's servers. The device only runs the agent loop, talks to the network and sends messages to Telegram. That takes almost no hardware.

So the machine under the agent can cost $15. A Mac Mini is a fine computer, but for this setup it is more than the agent needs.

If you do want to run local models, the hardware question changes. This post is about the other case, and it is the common one.

Originally posted on LinkedIn.