A local AI assistant built with a small CPU-only model
Bantz is a personal that runs directly on a local computer. It uses Gemma 4b, a small model, and can run on a CPU without needing a GPU.
Its planned and working features include reading and summarizing Gmail by category, connecting to , doing multi-source web research in the background, and watching CPU, RAM, and swap use with alerts. It also supports scheduled tasks and autonomous instructions, and desktop control on Wayland is still being built.
The main concern behind it is that relying on outside can create risk when a company, policy, or government decision cuts off access. Making a small local model useful is difficult, but can reduce exposure to sudden shutdowns and outside pricing.
Key points
- Bantz is a local AI , not a cloud-only service.
- It runs Gemma 4b on a CPU, so it does not require a GPU.
- It aims to handle Gmail summaries, calendar access, web research, system alerts, and scheduled tasks.
- The project is motivated by concern over depending on outside .
- can lower , but reliability and maintenance are the hard parts.