Use a Device Agent

After a Device Agent is created, you work inside its workspace. This is where you manage devices that share the same device model: connect a device record to the browser simulator or real hardware, inspect status and events, and use conversation to control or query devices.

Workspace Areas

Select a Device Agent from the sidebar. The workspace contains a device list, a state panel, and a conversation area.

Area	Use
Device list	Inspect accessible devices, select targets, and create, synchronize, or manage them as supported
Device state	Inspect online status, reported and desired state, available actions, and recent execution records
Conversation	Interact with the Device Agent through text, voice, or media input

Devices accessed through an IoT platform keep platform identities and native models. Direct devices report through SDK or MQTT. Entries depend on the adapter. Start with IoT Platform Integration to confirm the connection and scope.

Target a Device

Select a device in the list to include it in subsequent conversations. Alternatively, type @ in the input and choose devices for this request.

@thermostat-01 Show current temperature and humidity.

@thermostat-02 Set the target temperature to 23 degrees.

Mentioned devices join the currently selected devices as targets. Check the target list before querying or controlling, especially for comparisons or operations on multiple devices.

Without a selected device, start by asking about the Agent's devices:

How many devices are currently managed by this Device Agent? Group them by online and offline status.

Device Status

After selecting a device, the state panel shows recent reports and available capabilities. With an IoT platform, data may come from platform queries or synchronization. Direct devices provide device reports. Sections depend on the current model, access option, and available data:

Section	Meaning
Reported state	Data most recently confirmed by the device or platform, used to inspect its operation
Desired state	Target values the device should reach, when supported by the access option
Delta	Fields whose reported and desired values still differ, used to check whether targets have been applied
Available actions	A summary of supported actions and parameters; request execution through conversation
Latest reported events	Device alerts or business events with their time and data

For example, if a target-temperature field has a desired value of 24 degrees but the same field reports 22, inspect the operation result and later reports. Desired state expresses the target; actual state comes from device or platform confirmation.

Enum fields display labels from the model, such as Running or Standby, and retain the stored value when a label is available. Use the information icon beside the value to inspect permitted values; the current value is highlighted. Labels help reading, while reports and action parameters use the actual values declared in the model.

Show the current device status and tell me whether there have been any recent reported events.

History and service logs depend on the access option. Data describes the state at the time of the report. If reports stop, check update time as well as online status.

Inspect command and job records

When the platform provides records, the state panel displays Command history and Jobs. Inspect the action name, time, status, and result to distinguish pending, running, succeeded, and failed operations. Device job results reflect information provided by the platform. Inspect scheduled plans and their run history on the Scheduled Tasks page.

Continue checking results while an operation is pending or running. After a successful record, verify the target using reported state or the physical device. Query and control results are available in the conversation; the available-actions list summarizes capabilities.

Control Devices Through Conversation

Describe the target and action in natural language:

Set the target temperature to 24 degrees and switch to automatic mode.

The Agent uses the model and permissions to choose a command, service, or writable-state update, then executes it through the platform or direct connection. Missing parameters may require clarification. Prompts cannot add undeclared capabilities.

Afterward, check the conversation result and Reported state. A successful submission does not always mean hardware execution has completed. Verify the response, latest state, or platform execution record. For failures, check the target, parameters, online state, and platform permissions, then inspect Logs.

Connect and simulate

If the device is not connected, use the appropriate path:

• Existing IoT platform: confirm device connectivity on that platform and complete configuration and synchronization.
• Direct device: download an access package for the selected device and follow SDK Access or MQTT Access.
• Browser simulation: use a device record approved for testing and follow Device Simulator.

Disconnect the previous client before switching clients that share an MQTT identity. Simulation uses the same identity and reports data, so confirm the device scope before testing.

Configuration and conversation management

Open Configuration Overview from the Agent card to inspect conversation rules, voice tools, feedback, and device capabilities. Configure A2A controller/target bindings when collaboration is needed. See Device Agent Configuration.

See Conversations and Memory for history, long-term memory, and clearing.

Reply language and thinking content

The Agent first follows an explicit request for a reply language. Otherwise, it uses the language of your latest message. When that language is ambiguous, it uses the current console language. You can ask directly for a Chinese explanation or an English reply.

When the selected model supports reasoning and returns thinking content, the conversation shows Thought process. Click its heading to expand or collapse it and inspect the model's analysis before the answer. Verify device execution through tool results, execution records, and device state. See Model Configuration for reasoning settings and levels.

Use a phone

The mobile workspace keeps the conversation primary and opens the device list and state panel in bottom sheets:

1. Select the device name or Choose device at the top of the workspace to open the device list.
2. Select a target to return to the conversation, check its name at the top, and enter the request.
3. Open Device status to inspect state, events, and execution records. On narrow screens, this entry appears as a status icon.
4. Close the state sheet to continue chatting. Open Configuration Overview to change settings; phones use the list view by default.

You can still use @ to specify devices for a request. Check all targets before acting. See Conversations and Memory for mobile history and memory usage.

Automation and collaboration

Need	Feature	Check before use
Future checks, reminders, or control	Scheduled Tasks	Complete instruction, explicit target and time
Automatic reactions to state or events	Workflows	Matching trigger data and actions available through the adapter
Cooperation between Device Agents	A2A	Published capabilities, correct bindings and scope
Business-system or knowledge-service calls	Skills and Tools	Enabled tools, authentication, and permissions

Automation and collaboration reuse connected device capabilities. Targets and dependent services must be available when execution starts.

Voice, Camera, and Simulated Display

After a device is selected, the conversation area can show more interaction entry points when the corresponding capabilities are enabled. These capabilities can be used in the console and wired into real devices through the device-side SDK.

• Voice: speak to the Device Agent through the microphone, and connect device-side voice interaction through the SDK.
• Camera: provide image input for the current device, such as photo recognition, scene understanding, or vision-based operations. Real devices can upload image frames through the SDK.
• Simulated display: view or simulate visual output for a device. You can ask the Device Agent to generate a control panel, status panel, extended content, or a simulated device screen.

These capabilities apply to the selected device. For configuration and enablement, see Configuration. For usage guides, see Voice Interaction, Camera and Vision, and Simulated Display. For device-side code generation, see SDK Access.

Suggested Flow

1. Confirm the platform connection, model, and accessible devices.
2. Select a target and inspect online state and current data.
3. Query one value, execute one declared operation, and verify the actual result.
4. Validate event reports before configuring workflows and schedules.
5. Add A2A bindings and tools when collaboration or external systems are needed.