DETAIL 03Self-hosted AI assistant100% LOCAL INFERENCE
JavierJavier
A personal assistant running entirely on a Raspberry Pi 5 — local quantized LLM, self-growing MCP architecture, infrared eye tracking, OS and smart-home control, 24/7 via Telegram.A personal assistant running entirely on a Raspberry Pi 5 — local quantized LLM, self-growing MCP architecture, infrared eye tracking, OS and smart-home control, 24/7 via Telegram.
ROLE
Creator
YEAR
2026
PART
BOARD
STACK
Raspberry Pi 5 · Qwen (local) · MCP · OpenCV
100% local
Inference
24/7
Availability
3.1
The premiseThe premise
No cloud, no subscription, no data leaving the room: a personal AI assistant that lives on a Raspberry Pi 5 on my desk, running a quantized Qwen model locally.No cloud, no subscription, no data leaving the room: a personal AI assistant that lives on a Raspberry Pi 5 on my desk, running a quantized Qwen model locally.
Asset pending — Photo of the physical Raspberry Pi 5 setup with the IR camera
FIG 3.1
3.2
The architectureThe architecture
A self-growing MCP architecture: Javier can be granted new tools at runtime, each one gated behind human approval. An infrared camera adds eye tracking — look left and the active window snaps left. It controls the OS, the smart home, and answers on Telegram around the clock.A self-growing MCP architecture: Javier can be granted new tools at runtime, each one gated behind human approval. An infrared camera adds eye tracking — look left and the active window snaps left. It controls the OS, the smart home, and answers on Telegram around the clock.
Asset pending — Short clip of the eye-tracking window snap working
FIG 3.2
3.3 — Assembly sequence
Raspberry Pi 5 assembly — IR camera module, heatsink, local inference stack · loading