rt-claw  by zevorn

AI assistants for low-cost embedded devices

Created 1 week ago

New!

677 stars

Top 49.9% on SourcePulse

GitHubView on GitHub
Project Summary

Making AI Assistants Cheap Again! RT-Claw is an open-source intelligent assistant framework designed for low-cost embedded devices, primarily targeting the ESP32-S3. It enables developers and hobbyists to deploy sophisticated AI agents directly onto hardware costing as little as one dollar, facilitating seamless integration of AI into daily workflows and bridging the digital and physical worlds.

How It Works

RT-Claw employs a layered architecture, abstracting hardware interactions through an OS Abstraction Layer (OSAL) supporting FreeRTOS (via ESP-IDF) and RT-Thread, allowing for cross-platform compatibility. At its core, an LLM Chat Engine facilitates interactive conversations, augmented by a robust Tool Use system enabling LLM-driven function calls to interact with hardware and services. Features like composable Skills allow AI to orchestrate multiple tools, while Conversation Memory provides short-term RAM and long-term NVS Flash persistence. Swarm Intelligence enables networked nodes to discover each other, share capabilities, and invoke tools remotely.

Quick Start & Requirements

  • Primary install/run: Use provided scripts for ESP-IDF and QEMU setup (./scripts/setup-esp-env.sh). Build and run via make build-esp32c3-qemu and make run-esp32c3-qemu.
  • Prerequisites: ESP-IDF, QEMU, and an AI API key (set via export RTCLAW_AI_API_KEY=' ').
  • Environment: CNB Cloud IDE offers a pre-configured environment for browser-based development.
  • Docs: Refer to the Getting Started Guide for real hardware setup and advanced features.

Highlighted Details

  • Multi-RTOS Support: OSAL enables code to run on FreeRTOS (ESP-IDF) and RT-Thread with zero modifications.
  • Swarm Intelligence: Facilitates node discovery, heartbeats, capability sharing, and remote tool invocation across networked devices.
  • Chat-first Shell: Provides a UART REPL with advanced editing features, where direct input is processed by the AI, and system commands are prefixed with /.
  • Web Flash & Serial: Includes browser-based firmware flashing (esptool-js) and a serial terminal with ANSI color rendering.
  • Extensive Tooling: Over 30 built-in tools for hardware interaction (GPIO, LCD, audio), system functions, networking, and memory management.

Maintenance & Community

RT-Claw is inspired by OpenClaw, Nanobot, and MimiClaw. Community channels include Discord, QQ Group, Telegram, and Bilibili.

Licensing & Compatibility

The project is licensed under the permissive MIT license, allowing for broad compatibility with commercial and closed-source applications.

Limitations & Caveats

Some integrations, such as DingTalk, QQ, and Telegram, are marked as "In Progress." Features like Multi-Model API support, a Web Config Portal, and serving as a Claw Skill Provider are listed as "Planned." While verified on ESP32-C3/S3 and QEMU vexpress-a9, other platforms may require further integration.

Health Check
Last Commit

20 hours ago

Responsiveness

Inactive

Pull Requests (30d)
98
Issues (30d)
13
Star History
698 stars in the last 13 days

Explore Similar Projects

Feedback? Help us improve.