Dacular is an experiment in running AI agent workloads entirely on local hardware — specifically the Mac mini. The goal is to give users a self-contained AI stack they fully control, with no data leaving their machine.

The stack combines zeroclaw (an agent framework) with Modular Max running Gemma3 as the local inference engine. zeroclaw handles orchestration and tool use; Max provides fast, on-device LLM inference.

The setup is intentionally isolated: a dedicated macOS user account runs the AI stack, keeping it separate from personal data. Data from the main account can be mounted when Claude is not running, but the mount is removed before Claude starts — so private data is never accessible during an agent session.

Installation is handled by the install-dacular Claude skill, which guides users through account creation, software installation, and configuration in a single session.