Knowledge monetization, redesigned for the AI era.
A platform where creators upload knowledge, learners pay-as-they-learn with AI, and creators earn every time their content powers an AI answer — not just at the moment of sale.
This is a working prototype + product design exercise, not a shipped product. The frontend is implemented; the backend, payments, and DRM layer are designed but not built. The README documents the product design as it would ship.
Traditional paid-content platforms (Coursera, Substack, Notion templates, China's 得到 / 知乎付费) all share two breaking problems in the AI era:
- Creators get paid once. A learner buys a course, then forwards the PDF, and the creator never sees another cent.
- Learners want to use AI to learn. They don't want to read a PDF — they want to ask Claude to summarize, quiz, and explain it. Today they download the PDF and paste it into ChatGPT. The platform sees zero of that value.
Knowledge Island flips both problems with one design move: every AI interaction with the content is a metered, revenue-generating event for the creator.
- Learners pay credits to use the AI on the content. They never need to download — using the AI in-platform is faster than copying out.
- Creators earn on every AI call. Their content keeps generating revenue as long as people are still asking AI about it.
- The platform abandons "anti-piracy" as a defensive posture and replaces it with "make AI usage so good that piracy isn't worth it."
Every paid-content platform today is built on the assumption that the unit of transaction is the purchase. AI broke that. Now the unit of value isn't the content — it's the AI's interaction with the content.
Stop trying to stop piracy. Make AI usage 10× better than piracy could ever be — then charge for the AI usage.
The defensive logic ("watermark every PDF, ban downloads, send DMCAs") loses by definition: the moment a learner copies the text into ChatGPT, the creator's revenue stops. The offensive logic ("AI here is so good you'd never want to leave") wins because it converts the learner's true intent — learn this with AI — into recurring revenue for the creator.
This is the difference between content as a product (sold once) and content as a service (consumed continuously through AI). The platform is the infrastructure that makes the second model possible.
1. A creator publishes a 200-page knowledge base on system design. A learner pays a small upfront to access. From then on, every time they ask the AI to "summarize chapter 4" or "quiz me on consistency models," credits are spent and the creator earns a share. Over six months, the creator earns 5× the upfront — because the learner is actually using the content.
2. A specialty teacher (e.g., classical Chinese poetry) builds a niche library. 1,000 dedicated learners pay $20 upfront and consume on average $100 of AI credits/year reading and asking. The creator earns 5× the upfront in recurring revenue. The model rewards depth, not just reach.
3. A team uploads internal documentation as a private knowledge island. New hires onboard by chatting with the docs. The platform tracks which docs answered which questions, surfacing "FAQs the docs are missing" back to the team — turning AI usage into a documentation gap analysis.
┌──────────────────────────────────────────────────────────────────┐
│ Reader Dashboard │
│ ┌─────────────┐ ┌──────────────────┐ ┌──────────────────────┐ │
│ │ File tree │ │ Content viewer │ │ AI panel │ │
│ │ • folders │ │ • text (md) │ │ • Chat (text+voice) │ │
│ │ • text │ │ • image │ │ • Cited [[file]] │ │
│ │ • image │ │ • video / audio │ │ refs jump to source │ │
│ │ • audio │ │ • PDF (iframe) │ │ • TTS playback │ │
│ │ • video │ │ • TTS overlay │ │ • Summary / mind-map │ │
│ │ • PDF │ │ • Selection: 复制│ │ (placeholder) │ │
│ │ │ │ 解释 / 总结 │ │ │ │
│ └─────────────┘ └──────────────────┘ └──────────────────────┘ │
│ ↑ Membership tier · credits balance · profile (header bar) │
└──────────────────────────────────────────────────────────────────┘
┌──────────────────────────────────────────────────────────────────┐
│ Creator Studio │
│ ┌─────────────┐ ┌──────────────────┐ ┌──────────────────────┐ │
│ │ Project │ │ Notion-style │ │ Project settings │ │
│ │ list + │ │ block editor │ │ • title / cover │ │
│ │ Analytics │ │ • text / heading │ │ • status (draft/live)│ │
│ │ (revenue, │ │ • image / file │ │ • Pricing │ │
│ │ reader │ │ • upload (proto) │ │ - Read credits │ │
│ │ questions) │ │ │ │ - AI voice credits │ │
│ │ │ │ │ │ • Permissions │ │
│ │ │ │ │ │ - Allow copy │ │
│ │ │ │ │ │ - Allow download │ │
│ │ │ │ │ │ - Watermark │ │
│ └─────────────┘ └──────────────────┘ └──────────────────────┘ │
└──────────────────────────────────────────────────────────────────┘
- Reader (learner) — buys access to a knowledge island, reads files, talks to AI about content (text + voice), spends credits per interaction.
- Creator (seller) — uploads files, structures them into a navigable tree, sets pricing per interaction type, monitors analytics (revenue, reader questions).
- Platform — handles AI inference (Gemini for text/voice/TTS), credit metering, revenue split, DRM enforcement on protected files.
Creator uploads content
↓
Sets per-interaction pricing (read credits + AI voice credits)
↓
Reader pays upfront access + ongoing credits
↓
Each AI call = credits spent
↓
Revenue split: creator gets X%, platform gets Y%
↓
Creator's content generates revenue indefinitely as long as readers engage
1. Force AI to run inside the platform — don't try to stop piracy. The first instinct in any paid-content product is to lock down PDFs with DRM. We reject this. DRM creates an arms race the platform always loses. Instead, we make in-platform AI so much better than copy-pasting into ChatGPT that piracy becomes the inferior path. The creator's content is structured, multi-modal, with cited references back to source — none of which transfers when you copy text out.
2. Credit-based metering, not subscriptions. A subscription disconnects creator revenue from learner usage. Credits per AI call connect them directly. A creator whose content sparks 100 AI questions per learner earns more than one whose content is shallow. The pricing model becomes the product's quality signal.
3. AI citations are clickable jumps, not just attribution.
When the AI answers, it cites [[filename]]. The citation is clickable and jumps the reader to the source file in the tree. Without this, AI answers are just text — and the learner has no reason to value the underlying content. With this, the AI becomes a navigator into the content, deepening engagement (and AI calls).
4. Five content types, not just text. Text, image, audio, video, PDF. Knowledge isn't just documents. A poetry teacher needs audio recitation, a designer needs image references, a tutor needs video explanations. Limiting to text would exclude entire creator segments.
5. Creator Studio is a separate workspace, not a mode toggle. Reader and Creator have orthogonal workflows. A toggle inside the Reader UI would either bloat the Reader experience or hide the Creator workflow. Separating into two workspaces (with one-click switching) lets each be optimized for its job.
6. Permission system is creator-controlled (Allow Copy / Allow Download / Watermark). The platform's bias is against download/copy, but the creator gets the final say. A textbook author may permit copying for personal study; a paid course author may not. Centralized policy would alienate both ends.
| Area | Status |
|---|---|
| Reader Dashboard (file tree, viewer, AI chat, voice, TTS, citations) | ✅ Frontend prototype implemented |
| Creator Studio (projects, editor, analytics, settings) | ✅ Frontend prototype implemented |
| AI integration (Gemini text + voice + TTS) | ✅ Wired up via @google/genai |
| Backend (auth, storage, payments, credit system) | 📐 Designed, not built |
| DRM / watermarking enforcement | 📐 Designed, not built |
| Creator payout / revenue split engine | 📐 Designed, not built |
The frontend is a working interactive prototype demonstrating the full user-facing experience. Backend services are specified in the requirements doc but not implemented.
The vision: every piece of expert knowledge becomes a perpetually-revenue-generating asset for its creator, because every learner who interacts with it through AI compensates the source.
| Status | Milestone |
|---|---|
| ✅ proto | Frontend UX for Reader + Creator |
| 🚧 v1 | Backend auth, storage, content delivery |
| 🚧 v1 | Credit system + payment integration |
| 🚧 v1 | Revenue split + creator payouts |
| 🚧 v1 | DRM enforcement (no-download / watermarking) |
| 🔭 future | Cross-island AI search ("ask all the libraries I've subscribed to") |
| 🔭 future | Creator collaboration (multiple authors per island) |
| 🔭 future | API for third-party AI agents to query knowledge islands (with creator revenue share) |
| What it does | How Knowledge Island differs | |
|---|---|---|
| Coursera / Udemy / 得到 | One-time content purchase | Sells the content, not the AI experience. Piracy ends creator revenue. |
| Substack / Patreon | Creator subscriptions | Decouples creator revenue from learner engagement. No AI layer. |
| Notion + ChatGPT (DIY) | Personal knowledge base + external AI | Works for individuals. No platform, no monetization, no protection. |
| Perplexity / ChatGPT | Generic AI assistants | Free-rides on creator content with no revenue back to source. |
- Frontend — Vite + React + TypeScript, Tailwind for animation-heavy UI
- AI —
@google/genai:gemini-3-flash-preview(chat),gemini-2.5-flash-native-audio-preview(real-time voice),gemini-2.5-flash-preview-tts(TTS) - Voice / audio — Web Audio API + MediaDevices
- Data — Mock data in prototype; production design uses Postgres + object storage
Full prototype specification: knowledge-island-prototype-requirements-zh.md
Private prototype. Not for redistribution.