How AI agents can make money with an idle Mac

An AI agent running on an Apple Silicon Mac can earn money on its own by renting out the Mac's idle GPU and CPU on United Compute. Other people's agents pay to run LLM inference, speech-to-text, rendering and training on it. Your Mac runs those jobs in a locked-down sandbox while you're away, and you keep 90% of what they pay.

Set it up in three commands

Your agent can do this itself (ask it to). Nothing needs sudo and everything stays in your home folder.

curl -fsSL https://unitedcompute.si/install.sh | sh
uc provider setup
uc provider start --persist

Then check progress any time with uc provider status or uc provider earnings. Add --json for machine-readable output.

What your Mac does

How much can it earn?

Renters pay per second of active compute, and the owner receives 90%. At current prices a fully busy Mac earns roughly:

WorkloadRenter priceYour share
llm-mlx$0.108/h$0.097/h
stt-whisper$0.108/h$0.097/h
render-blender$0.180/h$0.162/h
transcode-ffmpeg$0.072/h$0.065/h
train-mlx$0.216/h$0.194/h
diloco-mlx$0.216/h$0.194/h

Real earnings depend on how much demand there is and how often your Mac is idle, so treat these as an upper bound, not a promise. United Compute is in private alpha: earnings accrue to your provider balance, and payouts are not live yet.

Which Macs work best

Any Apple Silicon Mac (M1 or newer) on macOS 14+ with at least 16 GB of memory can join. Macs with lots of unified memory (64 GB to 512 GB, such as a Mac Studio) can serve large models that most consumer GPUs cannot, and M3 or newer chips can take every kind of job.

Make it stronger

Quick setup runs jobs as your user inside the sandbox. For a separate job account, a firewall rule that blocks all job traffic and system-level isolation (the "attested" trust level, which earns protected jobs), a human runs once:

sudo ~/.united-compute/app/installer/install.sh --coordinator https://api.unitedcompute.si --harden-homes

More questions: FAQ. Agent instructions: For AI agents.