Build a tiny AI agent that reads sensors and controls hardware directly on a microcontroller.
Chat with a physical device over Telegram to trigger real world actions.
Add new skills to a deployed device at runtime without reflashing firmware.
Run scheduled or event triggered tasks entirely on-device with no cloud dependency.
| jetpax/pycoclaw | frectonz/electionwatch.et | andreicristi88/vexyn | |
|---|---|---|---|
| Stars | 134 | 15 | 9 |
| Language | Astro | Astro | Astro |
| Setup difficulty | moderate | easy | easy |
| Complexity | 4/5 | 2/5 | 1/5 |
| Audience | developer | general | general |
Figures from each repo's GitHub metadata at analysis time.
Requires a specific ESP32 board with at least 8MB flash and 4MB PSRAM.
PycoClaw turns a cheap microcontroller chip, the ESP32, into a small AI agent that can plan, remember things, and act on the physical world, all running directly on the chip itself rather than in the cloud. The hardware costs around five dollars and uses tiny amounts of power, measured in milliwatts instead of watts. The agent runs on a MicroPython based core called PFC, which aims to match the features of a project called OpenClaw but built to run on a microcontroller instead of a full computer. It gives the device a hybrid memory system that combines simple keyword search with vector based search, both running on the device itself. You can chat with it through Telegram or a companion app called Scripto Studio, with Discord support planned. It can also read and write files, run commands, browse the web, read sensors, generate images, run scheduled tasks, connect to other tools through MCP, and control real world hardware over CAN, I2C, SPI, GPIO, USB, or a small display. Skills, meaning extra abilities the agent can learn, are loaded at runtime from an online marketplace called ScriptoHub, so the device does not need to be reflashed every time you want to add a new capability. The device can also spawn smaller helper agents to handle delegated tasks, wake itself up on a schedule, and save its state to flash memory if the battery is about to die. To get started, you need a supported ESP32 board with enough flash storage and memory, then flash the firmware through a one click browser tool, and manage the device afterward through the Scripto Studio web app. It currently runs in production on the ESP32-S3, ESP32-P4, and ESP32-C6 chips, with support for another chip family still in progress. The project is released under the MIT license.
A MicroPython based AI agent that runs entirely on a five dollar ESP32 chip, no cloud required.
Mainly Astro. The stack also includes MicroPython, ESP32, C.
You can use, modify, and share this freely, including commercially, as long as you keep the copyright notice.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.