Raw Signal Technology

Raw Signal is built around a simple idea: useful AI should still work when there is no internet connection, no cloud account, and no remote provider deciding what happens next.

The AI Model

Raw Signal runs Mistral Small 3.2, a 24-billion-parameter large language model developed by Mistral AI in France. It was selected after testing multiple local AI options because it gave the best balance of practical knowledge, offline performance, and direct answers.

Why a French Model?

In testing, many U.S.-based models leaned more heavily toward refusal behavior, litigation-driven safety policies, and cloud-provider rules that can change over time. Raw Signal needed a model that could run privately on-device and answer practical reference questions without depending on a remote company's shifting gatekeeping.

Mistral Small 3.2 gave us the best fit: strong general knowledge, practical local performance, and a direct-answer profile that matched the purpose of the product.

Why Not Just Use a Bigger Cloud Model?

The largest frontier models require data-center-class hardware and are not available as self-contained offline devices. Raw Signal uses a model large enough to be useful while still practical to run locally in a compact unit.

Hardware Approach

Raw Signal is designed around AMD APU-style hardware — a CPU with integrated graphics capability — instead of a large external graphics card. That keeps the device more practical for compact, offline, limited-power use.

The system balances local language model processing with speech-to-text, text-to-speech, and voice detection so users can ask questions out loud and receive spoken answers without needing a keyboard, monitor, or cloud service.

Privacy and Memory

Raw Signal keeps questions on the device. It uses limited encrypted local conversation context so follow-up questions work naturally, then old history is cleared instead of creating a permanent cloud profile.

Locked Reference Model

Raw Signal is not designed to chase breaking news or live internet data. It is a locked reference model for stable knowledge: preparedness, first aid basics, water, shelter, food preservation, scripture, history, education, home repair, recipes, farming, and everyday problem-solving.