WireClaw - AI That Lives on the Wire

What if an AI agent could live inside the wall?

Most AI agents live in the cloud. They continuously depend on large language models to make decisions.

That works for chatbots.

But what about physical systems?

Should turning off a heater depend on a remote server? Should automation stop working if WiFi drops?

This project explores a different model.

LLM-assisted, locally autonomous

WireClaw runs on an ESP32 microcontroller (~520 KB RAM).

It uses an LLM (via OpenRouter, Ollama, etc.) to interpret natural language and generate automation rules.

Here’s the key:

Once the rule is created, it runs entirely on-device.

No model inference is needed at runtime.

The ESP32 executes deterministic logic locally - even offline.

How it works

• You send a message in natural language (Telegram, Serial, or NATS)
• WireClaw forwards the request to an LLM (cloud or local - your choice)
• The LLM returns structured intent
• WireClaw converts that into a deterministic rule
• The rule is stored in persistent memory
• From then on, the microcontroller runs it locally

No cloud dependency after rule creation. Reboot-safe. Offline-capable.

What I show in the demo

In the video, I:

• Talk to the ESP32 in plain English
• Register a temperature sensor in one sentence
• Create an automation rule: “tell me when it gets too hot”
• Trigger it live with a soldering iron
• Reboot the device
• Show that the agent remembers

The LLM is used only to understand intent - not to continuously control hardware.

Talk once, automate forever

This is different from “AI controlling hardware.”

Instead, AI compiles human intent into embedded logic.

After that:

• execution is deterministic
• no inference latency
• no token cost
• no cloud dependency
• no hidden decision-making

It combines the flexibility of natural language with the reliability of embedded systems.

The full source code is available on GitHub: https://github.com/M64GitHub/WireClaw

Try it yourself!

All you need is an ESP32 and a browser. No toolchain, no compiling.

Flash the Firmware

Go to wireclaw.io/flash and click "Install WireClaw." Select the serial port for your ESP32. The flash process takes about 2 minutes and includes the filesystem.

Connect to WireClaw-Setup

After flashing, the device reboots into AP mode. The LED pulsates light blue. Connect to the WireClaw-Setup WiFi network. A captive portal opens automatically - if it doesn't, open http://192.168.4.1 in your browser.

Configure

Enter your WiFi credentials, API key, and model.
Note: "API Base URL" is only needed for local LLMs (Ollama, llama.cpp, etc.) - leave it blank when using OpenRouter.

Save & Reboot

Hit save. The device reboots, connects to your WiFi, and starts the dual-loop agent. You're done - talk to it via Telegram, serial, or NATS.