ESP-Claw 🦞 AI Agent Framework for IoT Devices

💬 Chat as Creation · 🚀 Millisecond Response · 🧩 Smart and Extensible · 😋 Grows with You

Home | Docs | Online Flashing | Build from Source | 简体中文

**ESP-Claw** is Espressif's **Chat Coding** AI agent framework for IoT devices. It defines device behavior through conversation and completes the full loop of sensing, decision-making, and execution locally on Espressif chips. Inspired by the OpenClaw concept and reimplemented in C, ESP-Claw is lightweight, intelligent, and continuously evolving. With just an ESP32-series chip that costs only a few dollars, you can experience what makes ESP-Claw so nimble.

## 🌟 Key Features Traditional IoT usually stops at connectivity: devices can connect to the network, but they cannot think; they can execute commands, but they cannot make decisions. ESP-Claw brings the Agent Runtime down onto Espressif chips, turning them from passive executors into active decision-making centers.

💬 Chat as Creation	⚙️ Event Driven
IM chat + dynamic Lua loading Ordinary users can define device behavior without programming	Any event can trigger the Agent Loop and more Response can be as fast as milliseconds


🧬 Structured Memory	📤 MCP Communication
Organize memories in a structured way Privacy stays off the cloud	Supports standard MCP devices Works as both Server and Client


🧰 Ready Out of the Box	🧩 Component Extensibility
Quick setup with Board Manager Supports one-click flashing	Every module can be trimmed as needed You can also add your own component integrations

## 📦 Quick Start

ESP-Claw now supports a wide range of development boards based on ESP32-S3, ESP32-P4, ESP32-C5, and ESP32-S31, including breadboards, M5Stack CoreS3, and many others. Supported boards in [`./application/edge_agent/boards/`](./application/edge_agent/boards/) can be flashed online directly: configuration and flashing are done entirely in the browser, with no need to compile firmware locally or install a development environment first.

You can also build ESP-Claw locally. Please refer to the [local build documentation](https://esp-claw.com/en/tutorial/) for board adaptation, building, and flashing. Boards not listed above, as well as chips like the ESP32-P4, can also be supported through local builds and flashing. You can find practical examples in our [documentation](https://esp-claw.com/en/tutorial/). ### Supported Platforms

**LLM**: ESP-Claw now supports both OpenAI-style APIs and Anthropic-style APIs. It natively supports GPT models from OpenAI, Qwen models from Alibaba Cloud Bailian, Claude models from Anthropic, DeepSeek models from DeepSeek API, and also supports custom endpoints. > [!TIP] > > ESP-Claw's self-programming capability depends on models with strong tool use and instruction-following ability. We recommend `gpt-5.4`, `qwen3.6-plus`, `claude4.6-sonnet`, `deepseek-v4-pro` or models with comparable capability. **IM**: ESP-Claw supports Telegram, QQ, Feishu, and WeChat, and can be extended further. ## Development Plan ESP-Claw is still under active development. Feel free to open an issue to report problems or request features. You can also share your ideas through our [online survey (in Chinese)](https://fcn5wbhnyubf.feishu.cn/share/base/form/shrcndYcjbGFY1ymttTSyYoGIPh). [Click here to view our TODO List (in Chinese)](https://fcn5wbhnyubf.feishu.cn/wiki/SRlgwWUYei4WmykU8uMcUtzTnFf?table=tblWSgzWcyW7jv7B&view=vewaP9B0KX) and vote for the features or issues you care about. That helps us prioritize them sooner. ## 📷 Follow Us If this project helps you, please consider giving it a star. ⭐⭐⭐⭐⭐ ### Star History

## Acknowledgements Inspired by [OpenClaw](https://github.com/openclaw/openclaw). The implementation of Agent Loop, IM communication, and related capabilities on ESP32 also draws on [MimiClaw](https://github.com/memovai/mimiclaw).