Machine sizes

Every harness runs on a dedicated microVM with its own kernel. Memory and disk are dedicated to you; CPU is a burst ceiling, so when your agents compile, test, or index, they take the full headroom instantly.

Sizes

	Starter	Pro	Power
Price	$49/mo	$99/mo	$199/mo
CPU	2 vCPU boost	4 vCPU boost	8 vCPU boost
Memory	4 GB dedicated	8 GB dedicated	16 GB dedicated
Disk	64 GB NVMe	128 GB NVMe	256 GB NVMe

The disk is persistent: it survives restarts and node failures, and it holds everything, from the harness’s database to your agent workspaces.

What runs where

Concurrent agent runs are the main thing that separates the sizes. Each agent run (a Claude Code session, a QM run, a Hermes job) holds its own working memory on top of the harness itself.

Starter fits a personal setup or a light team: a Hermes assistant, a QM workbench with occasional runs, or a Paperclip board running 2 to 3 agents at a time. This is the right default if you are trying Harnesser out.

Pro fits a working team: a Paperclip board with around 6 concurrent agents, heavier QM usage, or agents that build and test real codebases where the extra memory and disk keep npm installs, git checkouts, and builds fast.

Power fits an agent operation: a Paperclip company running 12 or more concurrent agents, several long-lived always-on agents plus burst work, or large codebases and datasets that want the full 16 GB and 256 GB.

Two practical notes

• Memory is the limit that bites first. CPU bursts, but memory is fixed at the size you chose. If agents start failing under load, run fewer of them concurrently or use a bigger machine.
• Watch your disk. Agent work accumulates: repositories, dependencies, build artifacts. A full disk stops the harness’s database, so clean up old workspaces on smaller machines.

The size is chosen at creation and cannot be changed on a running harness yet. If you outgrow a machine, create a new harness at the bigger size.