A personal tool that captures the caption transcript of Skool classroom lessons, stores it per lesson in Postgres, and generates AI summaries with key points and chapters. Text only: no video files are downloaded or stored.
Status: Phases 1 to 4 are done (backend, Chrome extension for Vimeo-hosted lessons, summary pipeline, viewer and web panel), plus audio file uploads with transcription. Non-Vimeo lesson videos are not yet supported.
- Chrome extension (Manifest V3) with a side panel: capture the current lesson in one click, browse a library grouped by course, search, and jump the video to any timestamp.
- Idempotent ingest: capturing the same lesson again is a no-op unless the captions changed.
- AI summaries with key points and chapters. Long transcripts are chunked on caption boundaries, summarized per chunk, and merged. Summaries are tied to a transcript hash and marked stale when captions change.
- Full-text search over transcripts and summaries, with highlighted snippets and caption start times.
- Audio uploads: any ffmpeg-readable file is split into ten-minute mono chunks, transcribed with OpenAI, re-timestamped across the whole recording, and summarized. Duplicate files are detected by content.
- Web panel to browse, search, regenerate summaries, download Markdown, or export everything.
- Bearer-token auth on every route except
/health.
flowchart LR
subgraph Browser
Top[Content script on skool.com] --> SW[Service worker]
Frame[Script in Vimeo player frame] --> SW
SP[Side panel] --> SW
Web[Web panel]
end
SW -- "Bearer API_KEY" --> API[Fastify API]
Web -- "Bearer API_KEY" --> API
API --> DB[(Postgres via Drizzle)]
API --> Jobs[Background jobs: summaries, audio transcription]
Jobs --> OAI[OpenAI API]
Jobs --> FF[ffmpeg]
How capture works:
- The page script reads lesson, course, and video data from the page's
_next/dataJSON (no DOM selectors). - A script inside the embedded Vimeo player frame reads the loaded caption cues from the video text track.
- If the player frame has not loaded, the service worker reads the player config and downloads the caption file directly.
- The service worker sends the transcript to the backend. The OpenAI key is used only by the backend, never by the extension.
| Area | Tools |
|---|---|
| Language | TypeScript, Node.js 20+ |
| API | Fastify, Drizzle ORM, PostgreSQL, zod |
| Extension | Chrome Manifest V3, Vite |
| Web panel | Vite, plain TypeScript |
| AI | OpenAI API for summaries and audio transcription |
| Tooling | npm workspaces, Vitest, Biome, Docker Compose |
Requirements: Node.js 20+, Docker (for Postgres), and optionally ffmpeg for audio uploads.
npm install
cp .env.example .env # then set API_KEY to a long random string
docker compose up -d --wait # Postgres on localhost:5433
npm run migrate
npm run dev:api # http://127.0.0.1:3100| Name | Purpose |
|---|---|
DATABASE_URL |
Postgres connection string |
API_KEY |
Bearer token for all routes except /health (at least 16 characters) |
PORT, HOST |
API port and bind address (defaults 3100 and 127.0.0.1) |
OPENAI_API_KEY, OPENAI_MODEL |
Enable summaries. Without them, capture still works and summaries return SUMMARY_NOT_CONFIGURED |
OPENAI_TRANSCRIBE_MODEL |
Speech-to-text model for uploads (default whisper-1, which returns timestamps) |
MAX_UPLOAD_MB |
Largest accepted audio upload (default 300) |
FFMPEG_PATH |
ffmpeg binary used to split audio (default ffmpeg) |
WEB_DIST |
Folder with the built web panel (default apps/web/dist) |
EXTENSION_API_URL |
Default backend URL baked into the extension build |
npm run build -w @skool-notes/extension # writes apps/extension/dist- Open
chrome://extensionsand enable Developer mode. - Click Load unpacked and choose
apps/extension/dist. - In the extension options, set the backend URL and your
API_KEY, then click Test connection. - Open a lesson, open the side panel, and click Capture this lesson.
npm run build:web # builds apps/web/dist
npm run dev:api # serves the panel at http://127.0.0.1:3100/For hot reload, run npm run dev:api and npm run dev:web, then open http://localhost:5173.
npm test # API tests use a separate skoolnotes_test database, created automatically
npm run typecheck
npm run lintSchema changes: edit apps/api/src/db/schema.ts, then run npm run db:generate -w @skool-notes/api and npm run migrate.
All routes except /health require Authorization: Bearer <API_KEY>. Errors use { "error": { "code", "message" } }.
| Method | Path | Purpose |
|---|---|---|
GET |
/health |
Liveness |
POST |
/v1/lessons/ingest |
Upsert course, lesson, and transcript (?summarize=true also queues a summary) |
GET |
/v1/lookup |
Check whether a lesson is already captured |
GET |
/v1/lessons, /v1/lessons/:id |
List lessons, or get one with transcript and current summary |
DELETE |
/v1/lessons/:id |
Delete a lesson |
POST |
/v1/lessons/:id/summary |
Queue a summary ({ "force": true } regenerates) |
GET |
/v1/lessons/:id/summary, /summary/status |
Current summary, or its state |
POST |
/v1/uploads/audio |
Upload audio for transcription and summary |
GET |
/v1/lessons/:id/transcription |
Transcription job state and chunk progress |
GET |
/v1/search?q= |
Full-text search over transcripts and summaries |
apps/
api/ Fastify API, Drizzle schema and migrations, background jobs, tests
extension/ Chrome extension: content scripts, service worker, side panel, options
web/ Static web panel served by the API
packages/
shared/ zod schemas, API types, caption and page parsing utilities
docker-compose.yml Local Postgres
Screenshots coming soon.
This is a personal-use tool. Use it only for content you have legitimate access to. You are responsible for complying with the terms of service of any platform you use it with. This project is not affiliated with Skool or Vimeo.
MIT. See LICENSE.