Running a 28.9M Parameter LLM on an $8 ESP32 Microcontroller

What happened

Developer slvDev published an open-source project on July 25 running a 28.9M parameter language model directly on an ESP32 microcontroller -- a $8 chip originally designed for IoT applications.

Context and impact

The project sparked discussion about the lower bounds of useful edge AI. It shows offline on-device AI is accessible to anyone. Potential use cases: offline voice assistants, industrial IoT sensors, home automation without cloud.

Details

• Microcontroller: ESP32 (Xtensa LX6 dual-core 240 MHz, 520 KB SRAM), ~$8 retail
• Model: 28.9M parameters -- runs entirely on-device with no network connection
• Open-source on GitHub with full source code and deployment instructions
• Hacker News score 207 -- one of the most popular submissions on July 25