BUILD ME A JARVIS
The extended Moby prompt

A personalized, evidence-backed briefing in an immersive, branded experience.
This edition uses prerecorded narration started by a local voice cue, not live conversation.

If you are Moby and the user asked you to follow this document, use the full prompt below within the user's permissions and available tools. If you cannot retrieve the full document, ask the user to paste or upload it; do not pretend to have read it.

BEGIN PROMPT

Build me a personal, voice-triggered Moby briefing: a real analysis of where the model selected in this chat can help me most, presented as an immersive fullscreen experience in my brand. Do the research and build the working experience—not just a plan or mockup.

The intended recording is one continuous interaction: I enable the microphone, ask Moby which use cases this model has unlocked for me, pause, and the branded core wakes up and delivers my personalized briefing. Generate that briefing in advance; my voice only triggers playback. Do not build or imply live conversational AI for this request.

## 1. Find what matters to me, then test it
Use only my authorized account data, available conversation history, saved context, uploaded files, and previous outputs. Look for recurring goals, difficult assignments, corrections, and unfinished work—not just my latest conversation. State what history and source material you could actually inspect. Never invent access to a complete archive.

Choose 3–5 important workflows with enough evidence to evaluate. Where original inputs and prior outputs exist, rerun a bounded version of the original task, preserving its constraints, dataset, metric definitions, and dates. Set concrete success checks before judging results. Work in separate test copies; leave original artifacts and live systems unchanged.

For each case, show: why it matters to me, the historical result, what you tested now, the evidence, the verdict, and the next useful assignment. Mark results improved, unchanged, regressed, or unproven. Do not manufacture wins. A comparison with saved work is not proof that the model alone caused an improvement, and a proposed test is not a completed test.

If history or original materials are unavailable, say so. Use accessible current evidence for a clearly labeled current-state briefing, without invented before/after claims. Ask for a small batch of missing examples only if there is otherwise too little evidence to produce something useful. Do not substitute a generic sales dashboard for this personal workflow review.

Keep business discovery read-only. Do not change campaigns, budgets, CRM records, settings, or automations; do not publish or send anything externally.

## 2. Turn the findings into a spoken briefing
Write a concise, candid 4–6 minute narrative that sounds like a direct response to my spoken question: “Moby, what has this model unlocked for me? Which use cases has it made the biggest difference in?” Identify the actual selected model only when verified; do not assume every user is running Astra.

Start with the answer, not “Here is your briefing” or a generic introduction. A useful structure is “For you, the clearest gains are in…” followed by the strongest evidence-backed use cases, then the supporting story. If no gains were demonstrated, say so rather than forcing that opening. Explain uncertainty naturally. Use a calm, precise, lightly witty assistant voice, not an imitation of an actor or fictional character. Do not include my personal name in the public narration unless I request it.

Generate a downloadable narration using an available voice tool, with a matching transcript and source-backed written report. Do not invent audio or claim a voice capability that is unavailable. Keep customer identities, personal details, confidential quotes, and sensitive commercial figures out of the shareable narration and visible experience; keep detailed evidence in the private report.

## 3. Build the experience
Create a working, desktop-first HTML experience that fills the viewport and adapts to widescreen displays—not a phone frame, slideshow, analytics card grid, or obvious video player. It should feel like an assistant becoming present, not someone pressing play on a presentation.

Use my latest Brand Vault, supplied guidelines, approved logos, colors, and fonts; current supplied branding takes precedence over older references. Preserve original logo artwork. Use a single approved logo or symbol in the central core—no duplicate logo at the top left. If my identity is unavailable, use a clearly labeled Moby-style fallback: Triple Whale Blue #033855, Sky #C7FFFF, Inter, and Ocean Blue #007ABD for action buttons only. Use official Moby artwork only if accessible; never invent a logo. Do not add an OpenAI or other provider logo unless I specifically request it and its attribution is verified.

Make a restrained central core assemble on activation and respond to the actual narration’s amplitude and frequency—not a generic looping animation unrelated to the voice. Synchronize captions and chapter changes to the audio with one clock. No fake live business telemetry. Keep pause/replay, captions, reduced motion, and fullscreen controls accessible but tucked behind a discreet menu. Keep the main canvas free of playback bars, elapsed-time counters, instructional footer text, and persistent privacy paragraphs. Put source, snapshot date, playback disclosure, and privacy details in the settings/source panel and handoff, while preserving clear microphone consent and visible actionable error states. The briefing is a dated snapshot, not a live dashboard.

Add optional voice activation: one click to enable the microphone, a clear listening state in the core, local speech detection, then a short pause and wake-up transition before the fixed narration starts. Wait for me to finish a short sentence; do not trigger on the first sound or claim to recognize specific words. This is a start cue, not a conversational assistant. Do not record, transcribe, store, or upload microphone audio. Stop all microphone tracks before playback, on cancel, on timeout, and when the page is hidden. Include a discreet start-without-microphone fallback. If the embedded viewer blocks mic access, explain a verified standalone-browser route; do not bypass browser permissions or promise hands-free initial permission.

## 4. Verify and deliver
Check evidence, source coverage, brand fidelity, actual rendering, activation, audio/caption sync, pause/replay, resizing, reduced motion, microphone cleanup, blocked-permission fallback, and runtime errors. Report any checks you could not perform.

Deliver the private evidence-backed report, share-safe narration/transcript, and working branded HTML with simple launch and recording instructions. If an audio opening changes during revision, preserve the remaining narration and retime the captions, chapters, and visual response to the final soundtrack. If video export is available, also create a short, share-safe widescreen MP4 using the final branding and narration; otherwise say the MP4 was not produced. Never present an older export as an updated one.

Include a short, accurate social-post description explaining that Moby did the analysis and built a personalized briefing whose prerecorded narration starts on a voice cue. Do not describe this edition as understanding the live question, generating a fresh spoken answer, or supporting conversational follow-ups. Keep the disclosure in the post/handoff rather than cluttering the visual experience.

Do not stop at describing what you could build. Complete every supported part and identify exact blockers for the rest. Keep a future live-conversation version separate from this recording-ready briefing.

END PROMPT