FAQ

What is Device Agent?

Device Agent is an AI assistant that describes device capabilities through a device model and communicates with devices over MQTT.

Devices managed directly by Device Agent use DeviceSpec; adapted IoT platforms retain their native models and device paths. Built-in platform adapters include EMQX Fleets and ThingsNeural. See IoT Platform Integration for access options.

What kinds of devices is Device Agent for?

It fits devices and systems that can be described with clear state, commands, or events.

How is Device Agent different from general AI assistants?

General AI assistants center on conversations or tasks. Device Agent centers on device semantics. DeviceSpec defines commands, properties, and events; Device Agent uses that definition for device access, online state, command execution, and event handling.

What is DeviceSpec?

DeviceSpec is the device semantic definition for a device category. It describes basic device information, commands, properties, and events, so the Agent, SDKs, simulator, and A2A cards use the same capability model.

Does one Device Agent manage only one device?

No. One Device Agent maps to one device category and one device model. That device category can have many real devices connected under it. Each device has its own deviceId, online state, reported property values, and event history, while sharing the same capability definition.

Do I have to rebuild my existing device system?

Not necessarily. Existing firmware, gateways, or backend services can connect as long as they report state, respond to commands, and send events using the MQTT topics and payloads shown by Device Agent. New devices can use SDK code generated by the console.

Can Device Agent run automation?

Use conversations and scheduled tasks to run operations, workflows to process telemetry, device events, status, command responses, and validation errors, and A2A for cross-device collaboration.

Does Device Agent support voice and vision?

Yes. Voice handles input and spoken replies. Vision adds images or image frames to the device context.

What is A2A?

A2A is Device Agent's collaboration capability based on A2A over MQTT in EMQX 6.2.0 and later. When enabled, Device Agent publishes an A2A card from the device model and maps available operations into skills that other agents can discover and call.

EMQX provides card registration and discovery, online state, optional schema validation, and MQTT routing. Device Agent generates the card, handles SendMessage requests, invokes device capabilities, and returns results.

How is a device model different from a device?

A device model describes what a type of device can do and its limits. For example, a thermostat model defines temperature readings, target-temperature control, and the allowed temperature range.

“Living room thermostat” and “Bedroom thermostat” are two devices in the device list. They can share the same model, while each has its own device ID, current temperature, online state, and event history.

How do I save preferences? Does clearing remove memory?

Ask explicitly to remember, inspect, or forget preferences. Clearing archives history and resets context while retaining memory. Memory currently uses a shared deployment identity and does not automatically separate real users. See Conversations and Memory.

Should I use schedules or workflows?

Use Scheduled Tasks for future or periodic execution, and Workflows for reactions to reports. Devices and dependent services must be available at execution time.

Does A2A publication allow control of every device?

Publication makes capabilities discoverable. Collaboration in ordinary conversations also requires controller-target bindings and checks scope and online state. A2A over MQTT uses EMQX 6.2.0 or later. See A2A Multi-Agent Collaboration.

How does IoT Platform Integration relate to the commercial Platform?

IoT Platform Integration connects existing device systems so Agents can use native capabilities. The Device Agent Platform is a commercial production deployment offering. It adds operation at scale, monitoring, device AI credits, and licensing to common features. Contact us through Platform Overview for deployment needs.

Does a License include model and speech service credits?

A License authorizes Platform operation with request capacity and validity. It excludes third-party model, speech-recognition, and speech-synthesis allowances. Prepare their credentials and budgets separately. See License Management.