Fragmented device semantics
Firmware, cloud, apps, voice, and AI each redefine the same capabilities, state, and events.
Device Agent · Open Beta
Quick start
curl -fsSL https://emqx.sh/device-agent | shirm https://emqx.sh/device-agent.ps1 | iexDevice Agent in action
The agent understands device capabilities, reads live state, executes commands, and returns verified results — through chat, voice, IM, or your own app.

Define properties, commands, and events once. Device Agent turns them into a reusable device model for simulation, control, code generation, and A2A collaboration.

The challenge
Agentic AI can reason and decide, but the path from intent to physical device stays broken. Device semantics get rebuilt across firmware, cloud, apps, voice, and AI logic.
Firmware, cloud, apps, voice, and AI each redefine the same capabilities, state, and events.
Reasoning models can plan, but don't natively grasp target devices, commands, parameter limits, or execution results.
Issues only surface after firmware, cloud, apps, and AI are wired together.
Web, mobile, IM, voice, and business systems each rebuild the same device controls.
Different stacks, different teams, the same device semantics rebuilt over and over.
The solution
Define a device once, and the same definition powers the runtime, A2A collaboration, generated firmware, voice and vision, and every application surface.
Capabilities
Modeling, runtime, A2A collaboration, program generation, voice and vision, and application integration — one toolkit for AI-native devices.
Describe a device in plain language — or hand over a structured file — and get a single DeviceSpec covering its commands, state, and events.
What's the current temperature?
The current temperature is 24.5°C with 62% humidity. The target is set to 23°C in cooling mode.
It feels warm. Can you lower it?
Done! I've lowered the target to 22°C and switched to turbo cooling. You should feel the difference in about 3 minutes.
Each device class gets one AI entry point. Chat to check state, send commands, compare devices side by side, and switch between real hardware and simulation in the same view.
Expose device capabilities as skills other agents can discover and call, so devices take part in cross-device and cross-system workflows.
Once the DeviceSpec is confirmed, generate the device-side program. It handles connection, presence, telemetry, and command response — leaving sensors, actuators, drivers, and business logic to your team.
Turn on the living room lights and set brightness to 80%
Done! Living room lights are on at 80% brightness.
Bring voice and vision into the same device session. Talk to a device, or let the device read the image, screen, or scene in front of it.
/thermostat set 22°C
Reach the same Device Agent from Slack, Telegram, Discord, and any IM tool your team already uses — one agent, every chat.
Open architecture
Connect any hardware to any model. No rewrites, no vendor lock-in.
AI modelsIntelligence
Hardware ecosystemCompute
Use cases
From consumer electronics to industrial equipment, Device Agent powers smart devices across every industry.
Voice conversation, emotional interaction, and personalized learning experiences for the next generation of smart toys.

Home devices with voice commands, scene automation, energy optimization, and multi-device sync.

Autonomous navigation, task planning, remote operation, and fleet management for robotic devices.

Predictive maintenance, anomaly detection, process optimization, and compliance reporting.

Open the console, create your first agent, and connect a device — all in a few minutes.