# ESP-Claw

A C-language AI agent framework for ESP32-series IoT devices that defines device behavior through chat and dynamic Lua loading, running the full sense-decide-execute loop locally on a few-dollar chip.

ESP-Claw is Espressif's agent framework for IoT devices, letting a chip's behavior be defined through conversation rather than firmware reflashing. It runs the full loop of sensing, decision-making and execution locally on Espressif chips, turning a connected device from a passive command executor into an active decision-making one, and any incoming event can trigger the same agent loop with millisecond response times.

Behavior is defined through IM-style chat combined with dynamically loaded Lua scripts, so non-programmers can configure device behavior, and structured memory keeps device state organized while staying off the cloud for privacy. It communicates over the Model Context Protocol as both a server and a client, supporting standard MCP devices, and supports both OpenAI-style and Anthropic-style model APIs, with native support for OpenAI's GPT models and Alibaba's Qwen models.

Supported boards, including ESP32-S3, ESP32-P4 and ESP32-C5 hardware, can be configured and flashed entirely in the browser with no local toolchain, or built and flashed locally for boards not yet listed. It is written in C and released under the Apache-2.0 license.

Website: https://esp-claw.com/