espressif/esp-claw

ESP-Claw is an AI agent framework for IoT (Internet of Things) devices, written in C and designed to run on Espressif's ESP32-series chips, inexpensive microcontrollers that cost just a few dollars. The core idea is turning a simple connected device from a passive executor (one that just follows commands) into an active decision-maker that can sense, think, and act on its own. The framework implements what it calls "Chat Coding," users define what a device should do by chatting with it rather than writing code. Behavior can be programmed via messaging apps like Telegram, QQ, Feishu, or WeChat. Under the hood, the device uses an LLM (large language model, the same type of AI behind chat assistants) to interpret instructions and decide actions. It supports OpenAI-style and Anthropic-style AI APIs, including models from OpenAI, Alibaba Cloud, Anthropic, and DeepSeek. Key features include event-driven responses (any sensor trigger can kick off an AI decision loop in milliseconds), structured on-device memory that keeps data off the cloud, and support for MCP, a standard communication protocol for connecting AI models with external tools. The architecture is modular, so components can be trimmed down or extended with custom integrations. To get started, multiple ESP32-S3-based development boards are supported and can be flashed directly from a browser with no local install required. Local builds for other board variants are also documented. This framework is made by Espressif, the manufacturer of the ESP32 chip series.