How to run AI agents from your home computer

Why run agents at home

Running agents on your own machine buys you three things a metered cloud cannot: no per-token bill for the bulk work, data that never leaves the box, and a computer that is yours to point at any job. The tradeoff is that you own the setup — keeping it on, keeping it safe, keeping it supervised. This guide covers both halves: the machine, and the layer that makes it useful from anywhere.

The two setups

There are two honest ways to do this. The first is fully local and do-it-yourself: install a runner like Ollama, pull an open model, and drive a coding agent from your terminal. You control everything and you maintain everything. The second is to attach your machine to a workspace that handles dispatch, supervision, and memory for you, so you send a task from any device and the machine does the work. The first is more control; the second is more reach. They are not exclusive — the same machine can do both.

Step 1 — Attach your machine

The reach setup starts by pairing your machine once. With MachineSync, you open the Attach machine screen in The Cloud, get a one-time code, and run the installer on the machine:

curl -fsSL https://www.thecloud.so/machinesync/install.sh | sh -s -- --code <pairing-code>

It installs a small daemon in your home directory — no sudo — that starts at login, keeps itself alive, and survives restarts. Today the machine you attach is a Mac, Apple Silicon or Intel, with a signed and notarized app.

Step 2 — Dispatch work from any device

Once attached, hand an agent a task from your phone, your laptop, or the web. The work runs on your home machine — a real coding agent working autonomously through to the pull request it opens on a connected repo, with merging always left to you. Runs default to a hosted frontier model, or run free on the machine's own local model when you want no metered ceiling.

Step 3 — Watch it, and keep it bounded

Never run an agent you cannot see. Every run shows up in the Agents view, tagged with its machine and current task; open one to steer it, stop it, or read the full transcript. Safety comes from bounds: agents open pull requests instead of pushing to main, destructive tools are marked so they confirm first, and nothing goes public without your explicit say-so. Visibility plus limits is what makes home compute safe.

Step 4 — Make it always-on

The point of home compute is work that happens while you are away. Recurring jobs become a Loop: a prompt plus a cadence — interval, cron, or when the machine is idle — aimed at a repo, a Space, a Studio app, or a task, each with its own model, write policy, and budget. For the full pattern, see the guide on hosting an always-on AI dev environment.

Which machine to use

Any computer that stays on works, but the sweet spot for a home agent server is a quiet, low-power machine you can leave running. A Mac mini is the common pick — small, silent, and efficient enough to sit on a shelf for months. If that is the route you want, the Mac mini as your AI machine guide covers the specs that actually matter.

Start here

Get a pairing code from the Attach machine screen in MachineSync, run the installer on the Mac you want to use, and it appears live in the Agents view, ready for its first agent:

curl -fsSL https://www.thecloud.so/machinesync/install.sh | sh -s -- --code <pairing-code>

From there, dispatch a task from any device and watch it run on your own hardware.

Frequently asked questions