Aiden Hardware is a physical AI agent device built on the Luckfox Pico Zero (RV1106) that plugs into any smartphone or computer via USB, captures the screen through HDMI, listens and speaks through full-duplex audio, and controls the connected device autonomously through keyboard, mouse, and touch inputs — driven by an on-device Go-based AI agent runtime powered by a large language model.

We built it because we kept running into the same wall.

Every AI agent we built in software needed something from the host device — an API, an app install, admin permissions, a browser extension. Something that required the host to cooperate. And in the real world, host devices don’t always cooperate. Enterprise IT locks things down. Phones don’t expose the APIs you need. Old systems have no API at all.

So we asked a different question. What if the agent didn’t need permission from the host?

The idea

A keyboard and a mouse don’t ask for permission. They plug in and the device just accepts them. The host has no idea what’s on the other end — a human or a machine.

That’s the foundation Aiden Hardware is built on. Connect as a USB HID peripheral. Watch the screen through HDMI. Listen through audio. Then let the AI agent do what a human would do — look at the screen, decide what to do next, and act.

No SDK. No API. No install. The host just sees a keyboard and a mouse.

The full technical architecture is at github.com/AidenAI-IO/aiden-hardware-demo and deepwiki.com/AidenAI-IO/aiden-hardware-demo.

What we built

Aiden Hardware runs on the Luckfox Pico Zero (RV1106). Small enough to be unobtrusive. Powerful enough to run a real agent loop.

The stack has four layers:

Perception. HDMI capture gives the agent a continuous view of the connected device’s screen. Full-duplex audio handles listening and speaking simultaneously — not the turn-based interaction of most voice interfaces. Voice activity detection runs on-device via a Silero VAD model, which means the device can filter silence locally without sending audio to the cloud.

Reasoning. A Go-based agent runtime processes what the device sees and hears, calls an LLM, and decides what to do next. The agent has access to a defined tool set: keyboard input, mouse input, touch input, shell commands, web search. It looks at the current screen state and picks the right tool for the next step.

Action. USB HID output sends keystrokes, mouse movements, and touch events back to the connected device. From the host’s perspective, this is indistinguishable from a human at a keyboard. No drivers. No permissions. It just works.

Memory and skills. The agent remembers context across sessions. It can acquire skills — structured task routines — that expand what it knows how to do. Configuration lives in a simple agent.toml file and a web management portal. Firmware updates over the air.

The architecture is deliberately layered. C++ services handle the hardware interfaces — frame capture, audio, HID. The Go agent sits on top and only sees clean abstractions. This separation made the system significantly easier to test and extend.

Why this matters

Most AI agent deployment today fails at the integration layer. The agent is smart enough. The tools aren’t connected. The permissions aren’t granted. The software isn’t installed.

We’ve written about why most AI agents fail in production — the pattern is almost always the same. Great reasoning, broken tooling. Aiden Hardware sidesteps the tooling problem entirely by making the interface universal. If a human can use the device, Aiden can use it too.

The autonomous agents market is forecast to grow from USD 6.18 billion in 2026 to USD 127.86 billion by 2035. Most of that infrastructure will be software. But the software has to connect to the physical world somewhere. That connection point is what we’re building.

What it can do

• Navigate any app on any connected device without software installation
• Listen to voice commands and execute them through the connected device
• Automate repetitive screen-based workflows — forms, searches, data entry, navigation
• Remember what it’s done across sessions and improve how it handles recurring tasks
• Acquire new skills for specific workflows without reprogramming
• Update itself over the air as the agent software improves

For teams thinking about how to build on top of foundation models for real-world automation, see how to compare AI agent frameworks in 2026.

Where we are

Aiden Hardware is in the demo phase. The architecture works. The core capabilities are functional. We’re building toward production for US and Europe markets.

We’re sharing this now because we think the problem we’re solving is real and the approach is worth discussing — with developers, with potential users, and with anyone building in the AI agent space who’s run into the same integration wall we did.

If you’re building AI agents and hitting the host integration problem, we’d like to talk. Visit aidenai.io or explore the technical documentation.

Explore Aiden →

FAQ

What is Aiden Hardware?
Aiden Hardware is a physical AI agent device built on the Luckfox Pico Zero that connects to any smartphone or computer via USB. It captures the screen through HDMI, processes full-duplex audio, and controls the connected device autonomously through keyboard, mouse, and touch inputs using an on-device LLM agent runtime.

How is Aiden Hardware different from software AI agents?
Software agents need to be installed on the host device or require API access. Aiden connects as a standard USB HID device — the same protocol as a keyboard or mouse — and can observe and control any connected device without installation, admin rights, or API access on the host side.

What devices can Aiden Hardware connect to?
Any device with a USB port and HDMI output — smartphones, PCs, laptops, kiosks, and embedded systems. Because Aiden presents as a standard HID peripheral, compatibility is broad.

Does it require internet connectivity?
Core capabilities run on-device. LLM inference may use cloud API connectivity depending on configuration. The device supports local management through USB networking and a web portal.

Where can I read the technical documentation?
Full architecture documentation is at github.com/AidenAI-IO/aiden-hardware-demo and deepwiki.com/AidenAI-IO/aiden-hardware-demo.