Espressif Systems ESP-Claw framework builds local AI agents for ESP32 devices

We had already seen OpenClaw-like AI agents for ESP32 targets such as Mimiclaw and PycoClaw, but Espressif Systems has released its own ESP-Claw framework for building local AI agents capable of LLM-driven interaction and execution on ESP32 devices.

ESP-Claw enables ESP32 boards to respond to events, work with LLM-driven decisions, retain useful context, and take actions locally without connection to the cloud, unless required. The agent can control sensors and device state, and perform real-world actions such as controlling an RGB LED strip.

ESP-Claw

Some highlights of ESP-Claw include:

  • Chat coding – Define device behavior through natural conversation. The LLM handles dynamic decisions, and local Lua scripts execute deterministically, even when offline.
  • Quick response time – Devices react to real-time events instead of polling. A local event bus drives Lua rules for sensors and triggers, enabling millisecond-latency response on or offline.
  • Plug and Play with MCP –  ESP-Claw acts as both MCP Server and Client, exposing hardware to agents while calling external services.
  • On-chip private memory – Structured long-term memory lives on-chip. Preferences and routines are auto-extracted from conversations and events never leave the device.
ESP-Claw ESP32 RGB LED Strip demo
Initial demo in Chinese, and Espressif Systems added English translation on top.

The project’s website features several demos to control an RGB LED strip, create a game inflation balloons while blowing into a microphone, online/offline operation, and more.  There’s also a tutorial to get started. Note that ESP-Claw requires at least 8MB Flash and 8MB PSRAM, and it currently only supports ESP32-S3, but ESP32-P4 is coming soon. All you need to do is to get a compatible board, a breadboard, and a few components, and flash the firmware through the Web Flasher, or if you prefer, build it from source.

ESP-Claw Tutorial hardware setup
Hardware setup for ESP-Claw tutorial

ESP-Claw supports Telegram, QQ Bot, Feishu, and WeChat ClawBot messaging applications, as well as Tavily for web search if needed. You’ll also need to configure an LLM, and current options are OpenAI, Qwen, ChatGPT, and “Custom” for advanced users. Once everything is configured, you can control the ESP32-S3 board through your messaging applications.

ESP-Claw event sources action targets
Event sources and action targets

The video demo below automatically generates the code for a rainbow effect on an RGB LED strip with 40 LEDs connected to IO 14. Espressif-owned M5Stack also posted its own ESP-Claw video demo for StickS3 & CoreS3 hardware.

Share this:

Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress. We also use affiliate links in articles to earn commissions if you make a purchase after clicking on those links.

Radxa Orion O6 Armv9 mini-ITX motherboard
Subscribe
Notify of
guest
The comment form collects your name, email and content to allow us keep track of the comments placed on the website. Please read and accept our website Terms and Privacy Policy to post a comment.
0 Comments
oldest
newest
Boardcon MINI1126B-P AI vision system-on-module wit Rockchip RV1126B-P SoC