Desktop Apps.
Two shipping. More forging.

Six numbers. Both apps combined.

These six aggregate metrics summarize the entire commercial portfolio as of the May 15, 2026 audit. Per-app forensic detail follows in the dedicated cards below. Every aggregate number here is the sum of two verified per-app citations — no marketing math, no rounding tricks. The page is built so adding a third or fourth shipping app simply increments each total.

Combined production dependencies span both apps: @google/genai (1.0 in PromptDJ, 1.29 in VibeTube), electron (33 in PromptDJ, 41 in VibeTube), electron-updater (6.3 / 6.8), electron-builder (25 / 26), keytar (7.9 shared), vite (6.2 shared), react + react-dom (19.0 VibeTube only), lit (3.3 PromptDJ only), tailwindcss + framer-motion + lucide-react (VibeTube only). The architectural overlap on Electron, keytar, electron-builder, and the Google GenAI SDK is what defines the portfolio class.

$0.00 per user. Forever.

This is the architectural decision that defines the portfolio. Both apps require the user to supply their own Google Gemini API key. The key is encrypted into the operating system's keychain via keytar. The user pays Google directly for AI compute on their personal billing account. The operator — Design Delight Studio — never sits between the user and Google's metering. The result is a software business with a true fixed-cost ceiling regardless of sales volume.

Three reasons the inversion matters

• The compute-cost trap is bypassed. A traditional SaaS hosting Gemini 3.1 Pro for 1,000 users averaging 50 generations per month would be burning roughly $1,350/month in API costs alone (at $0.027 per generation). For Lyria streaming the burn rate is materially higher — 1,000 users at 30 minutes monthly equals ~$7,200/month. BYOK reduces both numbers to $0.
• Scale economics are non-existent. A typical SaaS founder must raise venture capital to subsidize early-stage compute. A BYOK operator scales linearly: every sale is pure margin minus the fixed-cost ceiling. The Bootstrapped Architect model becomes commercially viable at any scale.
• The user owns their own data path. No middleware proxy. No intermediate logging. The Gemini API call goes directly from the user's machine to Google's endpoint. This is privacy-by-architecture, not privacy-by-policy.

What BYOK gives up

The honest tradeoff: onboarding friction (risk R-04 in both per-app risk registers). A first-time user must obtain a Google Gemini API key before the app works. For a technical audience this is trivial; for a mass-market audience it is a meaningful onboarding hurdle. Both apps mitigate this by including ApiGate components that walk the user through the process, but the architectural choice consciously trades broad consumer accessibility for unbounded margin scalability. This is the right choice for a solo operator selling premium commercial software to technical users — and it is documented as a deliberate decision, not a bug.

VibeTube AI v1.1.1

A desktop-native YouTube content orchestration engine. The user provides a URL — product page, blog post, tutorial, anything. The application calls Gemini 3.1 Pro Preview with strict responseSchema enforcement and returns an 8-panel orchestration matrix: titles, descriptions, tags, hashtags, pinned comment, teleprompter script, cross-platform social copy, and visual asset directives. Built on Electron 41 + React 19 + Tailwind 4 with Framer Motion animations. The aesthetic is intentionally Y2K + Win95 — a retro chrome that signals the application is for builders, not consumers.

PromptDJ NS-9000 v1.0.0

A desktop-native generative digital audio workstation. The user blends up to 40 archetype prompts (Synthwave, Bossa Nova, Vaporwave, Trip Hop, French House, and so on) via weighted sliders, and Google's experimental lyria-realtime-exp model streams a continuous, low-latency audio synthesis through a 48 kHz Web Audio context. A custom AudioWorkletProcessor encodes the live buffer into a lossless WAV file that exports directly to the user's disk — bypassing every browser blob-download restriction. Built on Electron 33 + Lit 3.3 for hardware-accelerated rendering without React's virtual-DOM reconciliation bottleneck.

Two apps. One architecture pattern.

The portfolio shares a common architectural spine — Electron + keytar + electron-updater + BYOK + DRM gate — and diverges at the AI-model layer (request/response vs. streaming) and the frontend framework (React VDOM vs. Lit direct-DOM). The comparison below lines up every dimension that matters when reading this portfolio as a methodology rather than two isolated products.

What is verified. What is not.

This section differs structurally from the trilogy case studies. For the internal AGI systems (Sovereign Orchestrator, Synthetic Director, NicheForge) production evidence is internal-use evidence — files on disk, prompt history, output artifacts. For commercial desktop apps, production evidence is sales-side: Shopify listings, license activation counts, customer reviews, signed artifact deployments. The May 15, 2026 audit could verify the deployment infrastructure but not the sales-side commercial metrics — those are marked NOT VERIFIED with explicit pointers to what would resolve them in a future audit.

What is verified at the May 15 audit boundary: both apps are packaged, both ship NSIS installers, both have working DRM gate code, both have keytar-based key vaults, both have working auto-update artifact deployment configuration. The source code is the strongest evidence layer available without sales dashboard access — and on the source-code dimension, the verification is complete.

What this portfolio proves — and what it does not.

Five disclosures apply across the portfolio. Each one labels a specific claim with its evidence tier and the gap that would need to close before the tier could be upgraded. This format is identical to the trilogy case studies — the Vibe Academy curriculum refuses to publish marketing claims without an explicit evidence ledger attached.

Why these disclosures lead — same as the trilogy

The DDS Vibe Academy methodology requires that evidence tiers be the first thing a reader sees. What is anchored to file:line is anchored. What is calculated is shown with its inputs. What is inferred is labeled. What is testimony is testimony. What is unverified is flagged with a pointer to what would resolve it. Reading the rest of the page with these tiers in mind tells the reader exactly how much confidence each claim deserves. The technical-architecture claims should be read with full confidence; the sales-side claims do not exist on this page until a future audit resolves them; the cost projections and replacement-value figures should be read with the qualifiers above applied.

What it would cost to rebuild both apps.

The combined valuation across both shipping apps. Per-app detail is in the individual case study cards above; this section aggregates them under a single set of builder profiles so a reader can see the total portfolio cost discipline at a glance. Same methodology as the Sovereign AGI Suite trilogy — visible inputs, transparent math, adjustable to local market rates.

Why bootstrapped costs scale better than agency costs

The bootstrapped tier doesn't scale linearly with app count the way agency or senior-team tiers do. An agency must repeat the project management, requirements gathering, design system, and QA cycles for each app. A senior in-house team must rebuild the team for each project. The Bootstrapped Architect has no per-project setup cost — the methodology, the AI coding pattern, the Electron + keytar + DRM gate spine, and the publishing pipeline carry from app one to app two to app N. App three in this portfolio will cost less than $15K to ship. Not because the work is smaller, but because the marginal cost of producing another commercial app on this spine approaches the cost of writing the new domain-specific AI integration alone.

The portfolio is built to grow.

This page is intentionally architected for additional shipping apps. The 2-of-N chrome pill in the hero, the ItemList JSON-LD schema, the comparison table, and the side-by-side card layout all extend naturally. When app three ships, it joins the existing layout with no rebuild: another case study card slots between the existing two, the comparison table grows a column, the ItemList schema increments its numberOfItems value, the at-a-glance stats update.

Reserved slots — future ships under the same spine

Five architectural decisions worth studying.

The DDS Vibe Academy commercial portfolio teaches five lessons that an apprentice shipping their first commercial desktop app should internalize. These are distinct from the lessons taught by the Sovereign AGI Suite trilogy — those covered internal autonomous systems, where the operator pays the API costs. These cover external commercial software, where the user pays.

Lesson 1: BYOK before SaaS

A first-time builder almost always reaches for the SaaS model. It looks like the default. It is not. BYOK is the default for solo-operator commercial AI software in 2026. The compute economics of modern frontier models — Gemini 3.1 Pro at $1.25/$5.00 per million tokens, Lyria Realtime at $0.004 per second — make SaaS-style margin compression a hostile environment for a single-operator builder. BYOK eliminates the compute cost entirely from the operator's books. The tradeoff is consumer onboarding friction; that tradeoff is correct for a technical audience and incorrect for a mass-market audience. Pick the audience first, pick the model second.

Lesson 2: Schema enforcement is the moat

VibeTube AI ships a stable 8-panel matrix renderer in 964 lines of App.tsx because the Gemini API contractually cannot return malformed output. Without responseSchema enforcement, the same renderer would be 1,500+ lines covering Markdown parsing, fallback states, retry-on-malformed-output, and partial-render handling. The architectural decision to enforce the schema at the AI boundary is what makes a 2,234-LOC codebase sufficient to ship a commercial product. An apprentice should reach for this pattern on every structured AI output.

Lesson 3: Framework choice is cost discipline

PromptDJ NS-9000 uses Lit instead of React not because Lit is hipper, but because the audio-rate UI cannot afford virtual DOM reconciliation overhead. React would suffer noticeable render-cycle bottlenecking under 40 simultaneously animated sliders plus a live waveform visualizer plus 3-band EQ knobs. Lit's direct DOM updates have zero VDOM cost. The framework choice was a performance budget decision, not an aesthetic decision. An apprentice should always ask: what is the dominant rendering frequency of this UI, and what framework matches that frequency?

Lesson 4: Native OS hooks justify the Electron weight

Electron carries real cost — bundle size, memory footprint, install friction. Both portfolio apps justify that cost through capabilities that pure web apps cannot match: keytar for OS-level credential storage (vs. browser localStorage which is trivially extractable), fs.writeFileSync for native filesystem archives (vs. browser blob downloads), AudioWorkletProcessor-backed lossless WAV export (vs. compressed audio output through MediaRecorder), setWindowOpenHandler for hardened external-link navigation. If the app doesn't need any of these, it shouldn't be Electron. If it needs even one, Electron is the right wrapper.

Lesson 5: Document evidence tiers honestly

Both apps' audits flag commercial sales-side evidence as NOT VERIFIED because the auditor had code access but not sales-dashboard access. This page surfaces that fact directly in the Production Evidence and Honest Disclosure sections rather than hiding it. A solo operator who buries their evidence gaps publishes case studies that collapse the moment a serious reader audits them. The DDS Vibe Academy methodology requires the opposite: label every claim with its tier, publish the exact pointer that would resolve each unverified row, and let the reader weigh the resulting confidence accordingly.

Fifteen questions. Every answer audit-anchored.

Each answer below mirrors the FAQPage JSON-LD schema in the page head exactly, so search engines and AI assistants pull the same content the page presents. Every claim cites either a file:line in one of the two audited repositories, the Antigravity Forensic Engine audit dated May 15, 2026, or a public Gemini pricing source.