ESP-Claw Lets You Build IoT Projects via Chat

Are you still spending hours writing firmware for your embedded systems projects? If you are using an ESP32 microcontroller, you might not need to anymore — at least in some cases. Espressif has just released ESP-Claw, a chat coding AI agent framework for IoT devices. You simply flash the ESP-Claw firmware to your development board and wire it to any additional hardware you need, then send a message to ESP-Claw via a messaging app like Telegram explaining how it should work. Within seconds, it just works.

Users can define device behavior through natural language, with the system dynamically generating Lua scripts to control hardware like LED strips, displays, or motors. Once validated, those scripts can be stored locally, ensuring reliable operation even without a connection to the Large Language Models (LLMs) that created them. This hybrid model — LLM-generated logic combined with deterministic Lua execution — offers both flexibility and stability.

ESP-Claw is built around an event-driven architecture. Rather than waiting for explicit user commands, devices can react proactively to real-world triggers such as sensor readings or system events. For time-sensitive applications, local rules execute in milliseconds, bypassing any need for cloud interaction. If no rule applies, the system can escalate to an AI model for deeper analysis or even offload complex tasks like image recognition.

Unlike typical AI assistants that lose context between sessions, ESP-Claw organizes long-term data into categories such as user preferences, device events, and behavioral rules. These are stored locally with lightweight tagging, allowing the system to retrieve relevant context efficiently without heavy database overhead. Over time, the device can identify patterns and suggest new automations.

Interoperability is addressed through MCP, a unified protocol that allows ESP-Claw devices to act as both clients and servers. This enables direct communication between devices, software services, and cloud tools without relying on fragmented standards. Hardware capabilities are exposed as AI-friendly “tools,” making it easier for agents to understand and interact with physical systems.

With browser-based flashing, modular components, and support for common peripherals like cameras and microphones, getting started is straightforward. Developers can extend functionality through reusable components, while beginners can experiment with interactive chat-based control. If you’d like to learn more, there is plenty of documentation available.