BuildBB-0143

Hermes Agent on a 1 GB Raspberry Pi 3B+ uploads a sketch to the bench board, retrying on its own

Steve from IQ Workshop ran Hermes Agent on a 1 GB Raspberry Pi 3 Model B Plus, using OpenAI's Codex model and Telegram for chat. He asked it to write and upload a countdown blink sketch to an ESP32 on his workbench. The first upload failed; the agent diagnosed the error, retried and saved the fix to its memory.

Open original source youtube.com · IQ Workshop

Shown working
Why this evidence label?

The 12-minute video shows the Pi setup and the upload attempts on the bench; a companion blog post from 10 August describes the same run.

How we check entries
Play videoHermes Agent on a 1 GB Raspberry Pi 3B+ uploads a sketch to the bench board, retrying on its ownThe player loads from youtube-nocookie.com when you press play.
The source video, on YouTube.

Briefing

The companion post of 10 August gives the setup. Steve flashed Raspberry Pi OS Lite 64-bit to a USB drive, ran the Pi headless and installed Hermes over SSH. He picked a cloud model so the Pi did no heavy compute, and set up a Telegram bot that answers only him. The agent, named Gort, kept forgetting its identity until he found he had edited a SOUL.md file in the wrong place.

He then gave Gort SensorsIot's Universal Embedded Workbench repository and had it write a skill for it. The test job was an ESP32 sketch that counts an LED down from five to one, pauses five seconds and repeats, laid out as a PlatformIO project. Telegram voice dictation also worked, slowly, on the Pi.

Tags

Report a problem with this entry