# Clone — Product Requirements Document (draft; Gemini 2.5 Pro, Fable pass pending credit)

Here is the definitive, fully-functioning PRD for the "Clone" presence platform.

---

# **PRD: Clone Presence Platform**
*   **Product Lead:** Byron Arnao
*   **Version:** 1.0 (Project Bedrock)
*   **Status:** Proposal for Engineering & Design Kickoff
*   **Date:** 2026-08-10
*   **Target:** V1 Launch (The Bedtime Storyteller) - Q1 2027

## 1. Product Overview & Thesis

**Clone is not a tool you use; it's a presence you meet.**

Clone is a warm, retail-friendly presence platform that allows anyone to create, interact with, and share AI-driven representations of themselves, their loved ones, or historical figures. It bridges the gap between the sterile, text-based world of AI and the deeply human need for connection, memory, and understanding.

Our core thesis is that the most profound applications of AI are not utilitarian agents, but relational presences. Clone preserves the *manner* as much as the *matter* of a person—their cadence, humor, and unique way of being—creating a convincing and emotionally resonant experience. We begin not with a blank search box, but with a face and a voice that greets you, ready to connect.

The platform is designed around personas, which can be a photorealistic **Portrait** of a real person or an expressive, non-photorealistic **Character** (e.g., a block, creature, robot, or orb), ensuring a comfortable and appropriate entry point for all users, especially children.

## 2. The Tunable Experience (HIGH PRIORITY)

The fidelity of presence is not one-size-fits-all. It's a spectrum of quality, speed, and cost that the user must control transparently. The **Fidelity Dial** is a central, persistent UI element in every interaction.

**Core Interaction:** A user can, at any time, adjust the dial to move between interaction modes. The UI will transparently update with the provider, estimated latency, and cost-per-interaction.

### 2.1. The Fidelity Dial: UI & States

```
+-----------------------------------------------------------------+
|   [ Avatar Display Area ]                                       |
|                                                                 |
|                                                                 |
+-----------------------------------------------------------------+
|   Fidelity Dial                                                 |
|   <--[ ● ]----|---------|---------|---------|---------|------>   |
|      Text     Voice    Presence   Portrait     Live             |
|                                                                 |
|   Provider: None      Latency: ~50ms      Cost: Free            |
+-----------------------------------------------------------------+
```

**States & Backends:**

1.  **Text (Free):**
    *   **Description:** The fastest, cheapest mode. A simple, real-time chat interface.
    *   **Provider:** OpenRouter (for model selection flexibility).
    *   **Latency:** <100ms
    *   **Cost:** Free tier / lowest credit usage.
2.  **Voice (Standard):**
    *   **Description:** High-quality, low-latency text-to-speech. The avatar is a static image that pulses gently when speaking.
    *   **Provider:** Gemini TTS (default), ElevenLabs (premium, selectable).
    *   **Latency:** ~300-800ms
    *   **Cost:** ~$0.0003 / interaction.
3.  **Presence (Premium Voice + Image):**
    *   **Description:** Voice plus a subtly animated 2.5D portrait. The character exhibits micro-expressions and shifts gaze. This is the "convincing" tier.
    *   **Provider:** Gemini TTS/ElevenLabs + Nano Banana image generation/animation (self-hosted GPU).
    *   **Latency:** ~1-2s
    *   **Cost:** ~$0.005 / interaction (our highest margin tier).
4.  **Portrait (Video):**
    *   **Description:** Full, photorealistic talking-head video generation.
    *   **Provider:** HeyGen API (requires user's API key initially).
    *   **Latency:** ~5-15s
    *   **Cost:** ~$0.05 / interaction (pass-through cost).
5.  **Live (Future):**
    *   **Description:** Real-time, streaming video avatar.
    *   **Provider:** Tavus / Internal R&D.
    *   **Latency:** <1s (Target)
    *   **Cost:** TBD (High).

### 2.2. Advanced Tuning (Click-through from Fidelity Dial)

For users who want to perfect the *manner* of their Clone, advanced calibration provides fine-grained control.

*   **Expression Calibration:** A modal displays the Clone's face. Sliders allow the user to tune the default expression and sensitivity.
    *   `Default State:` [Slider: Frown <---> Neutral <---> Smile]
    *   `Smile Intensity:` [Slider: Subtle <---> Genuine <---> Exuberant]
    *   `Frown Sensitivity:` [Slider: Stoic <---> Expressive <---> Concerned]
*   **Gesture Tuning (V2):** A library of idle animations and conversational gestures (e.g., hand-on-chin thinking, a slight nod, a questioning head-tilt) can be enabled or disabled to match the source person's mannerisms. The system will auto-suggest gestures based on analysis of source video.

## 3. Mirror Mode & Digital Twin (HIGH PRIORITY)

**Thesis:** The best way to understand your Clone is to meet it. The most effective way to prepare for a difficult conversation is to rehearse it.

**Mirror Mode** is a dedicated environment for interacting with your own Clone.

*   **UI:** A split-screen or reflective interface. On one side, the user's live camera feed. On the other, their Clone in **Presence** mode.
*   **Interaction:** The user speaks, and their Clone responds in real-time. The Clone has access to the user's full corpus (emails, calendar, documents) and can answer questions, brainstorm ideas, or role-play a scenario.
*   **Use Cases:**
    *   **Rehearsal:** "Let's practice my pitch for the Q3 review."
    *   **Self-Reflection:** "Based on my emails this week, what seems to be my biggest priority?"
    *   **Calibration:** "Does that sound like something I would say?" The user can provide instant feedback to refine the Clone's style.

## 4. Characters & The Kid-Friendly Concept

Not all Clones should be photorealistic. **Characters** are a first-class entity, crucial for safety, comfort, and imagination, especially for younger users.

*   **Default Avatars:** When a user signs up, they can choose from a gallery of friendly, pre-designed Characters (orbs, friendly robots, cute creatures). This removes the pressure to upload a personal photo immediately.
*   **The Storyteller:** The flagship "Bedtime Storyteller" will default to a warm, friendly Character avatar (e.g., "Barnaby the Bear," "The Glow-Orb"). The *voice* is the grandparent's, but the visual is comforting and age-appropriate, avoiding the uncanny valley for children.
*   **Creator Studio:** Users can create their own Characters using simple prompts (V2).

## 5. THE MONETIZABLE FLAGSHIP: The Bedtime Storyteller

This is our emotional and commercial anchor for V1. We are not selling an AI; we are selling the experience of a grandparent's voice telling a new, personalized story to their grandchild every night. **This is our primary revenue engine.**

### 5.1. The Narrative & User Flow

1.  **The Gifter (Adult Child):** Sarah wants her kids to know her father, who lives far away. She sees an ad for Clone: "Give the gift of your voice. A new bedtime story every night."
2.  **The Gift:** Sarah purchases a "Family Plan" subscription and a one-time "Heirloom Voice Sitting" gift package for her father, John.
3.  **The Invitation:** John receives a beautifully designed email/physical card with a unique link: "Sarah has invited you to share your voice, so you can tell stories to your grandkids anytime."
4.  **The Sitting (The Capture Process):** John clicks the link and enters a warm, guided web experience.
    *   **Consent First:** The first screen is a clear, plain-language consent form. A short video of Byron explains what the platform is, how the voice clone will be used (only for the family, only for stories), and that he is in full control. John gives explicit, recorded video consent: "I, John Smith, consent to creating a voice clone for my family's use on the Clone platform."
    *   **The Reading:** John is guided to read 3-5 minutes of text into his microphone. The interface is encouraging and simple, like a teleprompter.
    *   **The Stories:** John is then prompted with questions to seed the story generator: "Tell me about the dog you had as a kid." "What was your favorite family vacation?" "What's a lesson your parents taught you?" He records 5-10 short audio anecdotes. This captures his narrative *style*.
5.  **The Creation:** The platform processes the voice and stories. Sarah is notified: "John's Storyteller is ready."
6.  **The Experience (The Child):** That night, Sarah's son opens the Clone Kids app on a tablet. He sees "Barnaby the Bear." He says, "Barnaby, can you tell me a story about a brave knight and a dragon?"
7.  **The Magic:** In his Grandpa John's voice, the bear tells a completely new, personalized story. The story subtly weaves in themes from John's own anecdotes. It feels authentic. It feels like him.

### 5.2. Business Model

*   **Family Plan (Subscription):** **$29/month or $299/year.**
    *   Includes 1 active Storyteller voice.
    *   Up to 5 family member accounts (listeners).
    *   Unlimited story generation.
    *   Access to the Character library.
*   **Heirloom Voice Sitting (One-Time Fee):** **$199.**
    *   Covers the premium voice cloning (ElevenLabs backend) and guided "Sitting" experience for one person.
    *   Can be purchased as a gift.
    *   Includes permanent, secure storage of the source audio and consent record.
*   **Additional Voices:** **+$10/month** per additional Storyteller voice on the Family Plan.

## 6. Constellation (Marquee Vision)

While V1 focuses on a single, powerful interaction, the future of Clone is multi-presence. **Constellation** (as detailed in the attached plan) is our vision for a living, navigable universe of relationships.

*   **Summary:** Constellation transforms a flat list of Clones into an interactive 3D graph.
    *   **The Hearth:** A private, intimate space for your personal relationships (family, friends).
    *   **The Pantheon:** A 'board of directors' of historical figures, mentors, and experts.
*   **Core Mechanics:**
    *   **Depth & Opacity:** The UI uses focus, blur, and transparency to make navigating complex relationships intuitive.
    *   **Salon Mode:** A user can convene multiple Clones into a single, shared conversation, creating a "reunion of the departed" or a "writer's room."
*   **Key Innovations:**
    *   **Grief-Safety:** Built-in circuit breakers ("Pause & Reflect") to manage emotionally intense interactions.
    *   **Temporal Slider:** A timeline that allows users to see how their relationship graph has evolved over time.
*   **Roadmap:** A static, 2.5D prototype of the Constellation graph will be developed in V2 to validate the navigation metaphor.

## 7. Future Vectors & Bold Ideas

### 7.1. Co-Located Clones (NEW)

**Problem:** What happens when two people who use Clone are in the same physical space?
**Proposal: The Consented Briefing.**

This is a high-utility feature for professional networking or social gatherings.

*   **Mechanism:** Using Bluetooth LE, two phones with the Clone app detect proximity.
*   **The Flow:**
    1.  Byron and Sarah, who have never met, are at a conference. Their phones detect each other.
    2.  Byron's phone buzzes. A notification appears: `Clone: You are near Sarah Vance. Her public Clone is available for a briefing.`
    3.  Byron taps the notification. He can either read a one-paragraph summary or tap a "Listen" button.
    4.  He brings his phone to his ear (or uses an earbud). His *own* Clone's voice whispers, "Sarah Vance is the lead on Project Chimera. Based on her public profile, she's interested in generative architecture and responsible AI. A good icebreaker would be to ask about her recent paper on emergent design."
*   **RAI:** This only works with Clones that have been explicitly set to "Public" or "Discoverable." The briefing is generated from publicly available data linked to Sarah's Clone, respecting her privacy. It's an intelligent, contextual icebreaker.

### 7.2. Wearable Companion (Garmin)

Byron's **Garmin** watch becomes an always-on portal to a Clone.

*   **Use Case: "Walk & Talk."** The user can initiate a voice-only conversation with any of their Clones directly from their watch, using a paired Bluetooth headset. This is perfect for walks, commutes, or brainstorming on the move.
*   **Demoable MVP:**
    *   A simple Garmin watch face showing the active Clone's avatar.
    *   Tapping the screen initiates a call.
    *   The phone app handles the processing, streaming audio to/from the watch/headset.
    *   The interaction feels seamless and ambient, untethering the Clone from the phone screen.

### 7.3. Historical Figures (Making it 'not-dry')

The standard "Ask Washington a question" is sterile. We will make it a dynamic, interpretive experience.

*   **Dual-Voice Interpretation:** When you ask George Washington a question, you don't just get one answer. You get two, from different interpretive lenses.
    *   **The Historian:** "Based on his collected writings, Washington would likely have emphasized..." (Cites sources on-screen).
    *   **The Critic:** "However, a modern critique would highlight that his perspective was limited by..." (Presents a counter-argument).
*   **Source Transparency:** Every statement made by a historical Clone is directly tied to a source document (a letter, a speech, a biography), which the user can click to explore. This grounds the experience in evidence and turns it into a powerful educational tool.

### 7.4. Ambient & Spatial Presence (Lower Priority)

*   **Ambient Home Device:** A dedicated smart-speaker-like device that houses a family's primary Clone (e.g., the Storyteller). It could have a simple screen to show the Character's face.
*   **Spatial/AR:** The long-term vision is to have Clones appear as presences in AR glasses. This is a V3+ concept, dependent on hardware maturity. We will not commit engineering resources in V1 but will design our APIs to be compatible with future spatial endpoints.

## 8. Technical & Foundational Architecture

### 8.1. Data Model (Simplified)

```json
// /clones/{clone_id}
{
  "clone_id": "c_12345",
  "owner_user_id": "u_abcde",
  "name": "Grandpa John",
  "avatar_type": "CHARACTER", // or "PORTRAIT"
  "avatar_uri": "s3://clone-assets/barnaby_bear.png",
  "voice_id": "v_67890", // links to ElevenLabs/etc.
  "corpus_sources": ["cs_a1", "cs_b2"],
  "consent_ledger_id": "cl_xyz"
}

// /corpus_sources/{source_id}
{
  "source_id": "cs_a1",
  "type": "AUDIO_RECORDING", // or "DOCUMENT", "EMAIL_INGEST"
  "uri": "s3://clone-corpus/john_sitting_audio.mp3",
  "provenance": "Captured during 'Heirloom Voice Sitting' on 2026-08-10",
  "metadata": { "topic": "childhood_dog" }
}

// /consent_ledger/{ledger_id}
{
  "ledger_id": "cl_xyz",
  "subject_name": "John Smith",
  "consent_type": "FAMILY_STORYTELLER",
  "consent_record_uri": "s3://clone-consent/john_video_consent.mp4",
  "is_revocable": true,
  "granted_by": "John Smith",
  "date": "2026-08-10",
  "authorized_users": ["u_sarah", "u_child1"]
}
```

### 8.2. Core APIs

*   `POST /clones`: Create a new clone.
*   `GET /clones/{id}`: Retrieve a clone's profile.
*   `POST /clones/{id}/chat`: The primary interaction endpoint.
    *   `body: { "prompt": "Tell me a story", "fidelity": "VOICE" }`
    *   Returns a response with audio URI and transcript.
*   `POST /clones/{id}/corpus`: Add a new source to the clone's corpus.
*   `POST /sittings`: Initiate a new "Heirloom Voice Sitting" flow.

## 9. Responsible AI (RAI) Framework

This is non-negotiable and foundational to Byron's brand and the product's integrity.

1.  **Radical Disclosure:** Every interaction with a Clone will have a persistent, subtle visual/auditory watermark. It begins conversations with a phrase like, "This is the AI clone of John..."
2.  **Consent by Design & The Digital Estate:** Consent is not a checkbox. It's a recorded, auditable event ("The Sitting"). For deceased individuals, control of the Clone passes to a designated Digital Heir, specified during setup. The heir can modify, pause, or delete the Clone.
3.  **The Evidence Boundary:** Clones must cite their sources. When asked a factual question, the Clone will respond, "Based on the story John told about his childhood dog,..." or "I don't have any information about that." It will never invent memories.
4.  **The Kill-Switch:** The creator of a Clone, or their heir, has a simple, one-click button to immediately and permanently deactivate the Clone. No dark patterns.
5.  **Grief-Safety & Emotional Circuit Breakers:** As designed for Constellation, the system will use sentiment analysis to detect emotionally fraught conversations and proactively offer a "Pause & Reflect" state, preventing harmful feedback loops.

## 10. Monetization & Tiers

*   **Free Tier:**
    *   Create one Clone of yourself (Character only).
    *   Limited to **Text** and **Voice (Gemini)** fidelity.
    *   $1 of initial credits for ~2000 voice interactions.
*   **Personal Plan ($15/month):**
    *   Unlock **Mirror Mode** with your own full corpus.
    *   Unlock premium **Presence** fidelity mode.
    *   Higher monthly credit allowance.
*   **Family Plan ($29/month):**
    *   The flagship **Bedtime Storyteller** product.
    *   Supports one premium cloned voice.
    *   Up to 5 family member accounts.
*   **Enterprise (Contact Us):**
    *   Brand Clones, spokesperson avatars, API access.

## 11. Phased Roadmap

*   **Q4 2026 (Pre-Launch):**
    *   Build the end-to-end "Bedtime Storyteller" flow.
    *   Finalize ElevenLabs integration for premium voice.
    *   Develop the guided "Sitting" web experience.
*   **Q1 2027 (V1 Launch):**
    *   Launch with the **Bedtime Storyteller** as the hero product.
    *   Marketing push focused on families and gifting for the holidays.
    *   Implement Free and Family plans.
*   **Q2 2027 (V1.5):**
    *   Launch **Mirror Mode** and the **Personal Plan**.
    *   Roll out the full **Fidelity Dial** UI with HeyGen integration (user API key).
    *   Launch the **Garmin "Walk & Talk"** companion app.
*   **Q3 2027 (V2):**
    *   Ship static **Constellation** prototype (2.5D graph view).
    *   Develop and test the **Co-Located Clone Briefing** feature with beta users.
    *   Introduce advanced **Expression Calibration**.

## 12. What We're NOT Doing (Yet)

*   **Open-ended Agents:** Clones cannot book flights, send emails, or act autonomously in the world. They are for conversation and presence, not action. This is a hard boundary for V1/V2.
*   **Un-permissioned Cloning:** We will not allow the creation of a Clone of any living public figure without their explicit, verified consent. Historical figures are the only exception and are governed by our strict sourcing rules.
*   **A "Deepfake-for-Hire" API:** Our video/voice generation capabilities will not be exposed as a raw, pay-per-use API to prevent misuse. Access is only available through the integrated, RAI-governed product experience.
*   **Social Network Features:** There is no "news feed," "liking," or "following" of Clones. Interaction is intentional and direct, not performative.