DuckyClaw 🐣🦞
Your autonomous AI companion. Deploy agent brains to real hardware—MCUs, SoCs, PCs. One codebase.
What is DuckyClaw?
A hardware-oriented AI Agent framework. Not just a dev board—a full deployment stack for the physical world.
When your AI Agent needs to perceive, decide, and control real devices, DuckyClaw is the stack. Built on TuyaOpen C SDK—no Node.js on MCU. One codebase from ARM Cortex-M to x64.
- Device–cloud fusion: low cost, one TuyaOpen key for unified access to services.
- Unified messaging: Telegram, Discord, Feishu.
- Native IoT control: real devices in the Tuya ecosystem.
- Proactive agent: agentic loop and monitoring, not just chat.
Core features
Everything you need to run an AI Agent on the edge and in the cloud.
Device–Cloud Multi-Agent
On-device and cloud agents in one system. Tuya Cloud for LLM, RAG, MCP.
IoT device control
Control Tuya smart devices directly. IoT memory learns your preferences.
Hardware voice (ASR)
Voice input with ASR on supported boards. No extra app needed.
Device skills
Extensible hardware skills: cameras, sensors, displays. Build and plug in your own; deep integration with the agent loop.
Unified messaging
Telegram, Discord, Feishu via one gateway. No platform lock-in.
MCP tools
CRON, FILE, IoT control, EXEC (e.g. RPi). Extensible skill plugins.
Persistent memory
Agent.txt, memory.txt, IoT memory. Context that evolves over time.
Agent loop heartbeat
Heartbeat mechanism keeps the agent loop alive: scheduled pings, health checks, and continuous background operation.
Quick start
One TuyaOpen key. Pick your board. Configure and run.
Hardware support
One codebase. From microcontrollers to single-board computers to PC.
Linux Ubuntu
🧩 More boards are supported. The framework is highly portable and compatible — contributions are welcome.
Build your own hardware skills
Create and extend hardware skills. Connect cameras, sensors, and other peripherals with deep integration into the agentic AI loop.
Developers can build custom hardware skills on top of DuckyClaw: add cameras for vision, sensors for environment data, displays, audio, and more. These skills plug into the same agentic framework—the agent can reason over sensor data, trigger actions on hardware, and combine device inputs with cloud AI. Deep integration means your hardware is a first-class part of the agent loop, not a side script.
- Connect cameras, sensors, displays, and other peripherals as skills.
- Deep integration: hardware feeds the agent; agent controls hardware in the same loop.
- Build on TuyaOpen drivers and APIs for fast prototyping.
Local agent loop + Tuya Cloud
Start on device, scale with the cloud. One agent loop, unlimited possibilities.
Don't limit your agent to local—get more with Tuya Cloud.
Your agent doesn't have to choose. Run the core loop on your board for speed and privacy; when you need more power—multi-agent workflows, RAG, web search, IoT, music—Tuya Cloud extends it with one unified key. Low cost, no lock-in.
Device local agent loop
Sense, reason, act—all on your board. Core loop runs on the edge with full control; no cloud required.
Tuya Cloud: multi-agent & more
Unlock multi-agent workflows, RAG, ASR/TTS, third-party MCPs, web search, IoT control, music services—all with one unified key. Low cost.
Customizable workflow & more
Define triggers, variables, and roles. Plug in hardware skills, MCP tools, and Tuya IoT—your rules, your agent.
Architecture
Local execution layer + agentic loop + Tuya Cloud. Digital and physical in one loop.
Gateway aggregates messaging and voice; Agent Brain runs locally or in the cloud; modular skills handle both digital (FILE, CRON, MCP) and hardware (IoT, GPIO, sensors) actions.
Platform Overview
Where DuckyClaw runs, its architecture layers, and quick comparison with similar solutions.
Three-layer architecture
- Layer 1 — Local hardware execution Gateway (Telegram, Discord, Feishu, ASR); Agent Brain / LLM / Vision; persistent memory + IoT Memory; modular skills (FILE, CRON, MCP) and hardware skills.
- Layer 2 — Agentic loop & proactive monitoring Agent runs in the background, senses state changes, and triggers actions without manual wake-up.
- Layer 3 — Tuya AI Cloud Unified model access (GPT, Claude, Qwen, DeepSeek); low-cost, unified key access to services; Multi-Agent workflows, RAG, ASR/TTS/STT, cloud MCPs, Tuya Agentic IoT Control.
DuckyClaw vs others (at a glance)
| Aspect | DuckyClaw | OpenClaw / MimiClaw / others |
|---|---|---|
| Runtime | TuyaOpen C; ARM Cortex-M / Cortex-A / x64; no Node on MCU | Node.js (OpenClaw); ESP32-S3 only (MimiClaw); server/desktop |
| Deployment | MCU, SoC, PC — one codebase | Mac/Pi/VPS or single board; no edge MCU (others) |
| Cost | Low cost; unified key access to services | Bring your own API; Claude Pro/Max $20–200/mo or API-heavy |
| Device IoT control | Native Tuya ecosystem control | Typically no built-in device control |
| Voice (ASR) input | Hardware ASR on select boards | Not natively supported |
Tuya Cloud capabilities (zero-coding extendability)
- Multi-Agent workflow orchestration
- RAG knowledge retrieval
- ASR / TTS / STT full voice pipeline
- Cloud third-party MCPs
- Tuya Agentic IoT Control
Community & links
Star, contribute, and stay in touch.
See it in action
Experience DuckyClaw in action — watch highlight videos, see live demos, and explore real-world deployments below.