Glass AI Chat

Purpose

Demonstrate booting AI chat sessions, submitting user prompts, and collecting incremental model output on the glasses.

Entry points

Glass home:

• HomeActivity → SdkMediaActivity

Sample Activity:

• com.rokid.glass.SdkMediaActivity

Platform

• Glass

Main file

• SdkMediaActivity.kt

Hot spots

• binding.btnAiChat.setOnClickListener
• binding.btnStopAiChat.setOnClickListener
• mAIChaLister stub implementation

Bundle with:

• SendMessageActivity.kt — ASR/TTS flows
• HomeActivity.kt — offline wake-word registry

Related docs

• Glass3 Glass SDK API reference
• Quick start

Flow

1. Tap the AI shortcut.
2. GlassSdk.getGlassAiChatService()?.startAiChat(false) boots the session channel.
3. toAiChat(...) submits the canned prompt bundled with upstream (same wording as upstream repo).
4. AiChatListener streams partial + final payloads.
5. endAiChat() tears down jobs when user aborts.

Snippets

binding.btnAiChat.setOnClickListener {
    GlassSdk.getGlassAiChatService()?.startAiChat(false)
    GlassSdk.getGlassAiChatService()?.toAiChat(
        "Will it rain in Hangzhou tomorrow?",
        mAIChaLister
    )
}

binding.btnStopAiChat.setOnClickListener {
    GlassSdk.getGlassAiChatService()?.endAiChat()
}

private val mAIChaLister = object : AiChatListener.Stub() {
    override fun onAiChatAnswer(
        answer: String?,
        isFinish: Boolean,
        contentType: String?,
        sessionId: String?
    ) {
        L.i(TAG, "onAiChatAnswer: answer = $answer")
    }

    override fun onError(code: Int, message: String) {}
}

Callback matrix handled

• onContinuousModeUpdate(...)
• onAiChatAnswer(answer, isFinish, contentType, sessionId)
• onError(code, message)
• onAiTakePhoto(filePath)

Sequence

sequenceDiagram
    participant UI as Glass UI
    participant AI as AI Chat Service

    UI->>AI: startAiChat(false)
    UI->>AI: toAiChat(prompt, listener)
    AI-->>UI: onAiChatAnswer(chunk, isFinish=false)
    AI-->>UI: onAiChatAnswer(chunk, isFinish=false)
    AI-->>UI: onAiChatAnswer(chunk, isFinish=true)
    UI->>AI: endAiChat()

FAQs

Why isFinish exists

Answers stream chunk-by-chunk — answer deltas arrive until isFinish asserts completion for that turn.

onAiTakePhoto

Signals device-action intents (e.g. capture) beyond pure texting—helps design assistant scenarios.

sessionId

Log correlation + future multi-turn context—persist for production analytics.

Voice stack integration

Typical storyboard:

• ASR → user text
• AI Chat → reasoning / answer
• TTS → audible reply

Samples currently split these across pages but you can unify them in one state machine.

Suggested composition

Stage	Capability	Sample reference
Input	Cloud ASR	SendMessageActivity
Reasoning	AI Chat	SdkMediaActivity
Output	TTS	SendMessageActivity

Caveats

• Demo focuses on wiring, not final chat UI.
• Production assistants should orchestrate ASR + AI + TTS lifecycle in one cohesive controller layer.