Briefing
The model runs at Q4_0, about 490 MB, under llama-server with an 8K context, behind a Go gateway. The remaining tools read, write and list files, switch the Linux-side LEDs through sysfs and inspect the network read-only. The arduino tool compiles for the arduino:zephyr:unoq target, and OpenOCD flashes the result over SWD through the Linux processor's GPIO pins.
Laurenvil says a flash takes under a second. He reports about 1.3 GB of the board's 4 GB of RAM in use and about 8 tokens per second. In his demo the agent scrolls "QClaw" across the 13 by 8 LED matrix.
Written by the Boards & Bots curator
From the GitHub API
Repository facts
| Repository | laurenvil/ |
|---|---|
| Stars | 20 |
| License | MIT Permissive |
| Language | C++ |
| Last push | 10 Jun 2026 4 mo ago |