Transform your social media saves into an intelligent, self-organizing knowledge system.
Thredion is not just a link saver — it's a cognitive layer that understands, connects, and resurfaces your saved content. Think of it as an AI Second Brain that grows smarter with every link you send.
| Link | |
|---|---|
| Dashboard | thredion.vercel.app |
| API | thredion-api.azurewebsites.net |
| WhatsApp Bot | Chat on WhatsApp → |
How to connect the WhatsApp Bot:
- Click the WhatsApp link above (or scan the QR code in Twilio Console)
- It opens WhatsApp with the Twilio Sandbox number +1 (415) 523-8886
- Send
join deep-thirdto activate the sandbox- Once connected, just paste any link (Instagram reel, YouTube video, article, tweet) and the bot will process it through the full cognitive pipeline
- Open the Dashboard to see your memories, knowledge graph, and analytics
📹 Watch the full walkthrough: Thredion Demo Video
We all save hundreds of Instagram reels, tweets, and articles — but never look at them again. They're buried, forgotten, and effectively lost knowledge.
Existing solutions just store links. They don't understand them.
Thredion introduces 5 cognitive capabilities that turn passive saving into active knowledge building:
Every saved link is processed through AI to extract meaning — not just tags, but a hierarchical topic graph.
Fitness → Bodyweight Training → Core Strength → Home Workout
When you save new content, Thredion automatically surfaces forgotten insights that are semantically related.
"You saved a similar Python optimization trick 12 days ago"
This is real cognitive behavior — not just notifications.
Related ideas are automatically connected, forming a personal knowledge network you can visualize:
Python optimization
↓ connected to
FastAPI performance
↓ connected to
Async programming reel
Each memory gets a transparent score (0-100) based on:
- Content Richness (0-25)
- Novelty (0-25)
- Connectivity (0-25)
- Topic Relevance (0-25)
Every score comes with human-readable reasons — full explainability.
Not just cards — a full cognitive interface with:
- Recent Memories
- Resurfaced Insights
- Interactive Knowledge Graph
- Analytics & Category Distribution
- Random Inspiration button
- Embedded Video/Post Players — YouTube and Instagram content plays inline
| Feature | Description |
|---|---|
| Inline Embeds | YouTube videos and Instagram posts play directly inside the dashboard |
| Random Inspiration | Rediscover a forgotten memory at the click of a button |
| Knowledge Graph | Interactive force-directed graph connecting related memories |
| Smart Resurfacing | Automatically recalls forgotten content when you save something related |
| Explainable AI | Every importance score comes with transparent reasoning |
| 6 Platforms | Instagram, Twitter/X, YouTube, Reddit, TikTok, and any article URL |
| 3-Tier Fallback | Embedding: sentence-transformers → TF-IDF → hash; Classification: GPT → keywords |
┌────────────────────────┐
│ WhatsApp Bot │ ← User sends Instagram/Twitter/article link
│ (Twilio Webhook) │
└──────────┬─────────────┘
│
▼
┌────────────────────────┐
│ FastAPI Backend │ ← thredion-engine
│ │
│ ┌──────────────────┐ │
│ │ Cognitive Engine │ │
│ │ │ │
│ │ • Extractor │ │ Extracts content from any URL
│ │ • Embeddings │ │ sentence-transformers/all-MiniLM-L6-v2
│ │ • Classifier │ │ AI categorization + summarization
│ │ • Knowledge Graph│ │ Connects related memories
│ │ • Importance │ │ Explainable scoring algorithm
│ │ • Resurfacing │ │ Surfaces forgotten insights
│ └──────────────────┘ │
│ │
│ SQLite Database │
└──────────┬─────────────┘
│
▼
┌────────────────────────┐
│ Next.js Dashboard │ ← thredion-dashboard
│ (Tailwind CSS) │
│ │
│ • Memory Cards │
│ • Knowledge Graph │
│ • Resurfaced Panel │
│ • Analytics View │
│ • Search & Filters │
│ • Random Inspiration │
└────────────────────────┘
| Layer | Technology |
|---|---|
| Bot | WhatsApp (Twilio Sandbox) |
| Backend | Python / FastAPI |
| AI Embeddings | sentence-transformers/all-MiniLM-L6-v2 |
| AI Classification | OpenAI GPT-3.5 (with keyword fallback) |
| Similarity | Cosine Similarity |
| Database | SQLite (zero-config) |
| Frontend | Next.js 14 / React / Tailwind CSS |
| Icons | Lucide React |
When a URL is received (via WhatsApp or dashboard):
1. EXTRACT → Pull title, caption, content, thumbnail from the URL
2. EMBED → Generate 384-dim vector embedding (MiniLM-L6-v2)
3. CLASSIFY → AI categorization + summary + tags + topic graph
4. STORE → Save enriched memory to SQLite
5. CONNECT → Build knowledge graph edges (cosine similarity > 0.55)
6. SCORE → Compute explainable importance (0-100)
7. RESURFACE → Find & surface forgotten related memories
- Python 3.10+
- Node.js 18+
- (Optional) OpenAI API key
- (Optional) Twilio account for WhatsApp
cd thredion-engine
cp .env.example .env # Edit with your API keys (optional)
pip install -r requirements.txt
python main.pyBackend runs at http://localhost:8000
API docs at http://localhost:8000/docs
cd thredion-dashboard
npm install
npm run devDashboard runs at http://localhost:3000
- Create a free Twilio account
- Go to Messaging → Try it Out → WhatsApp Sandbox
- Set webhook URL to:
https://thredion-api.azurewebsites.net/api/whatsapp/webhook(POST) - Add Twilio credentials to
.env - Join the sandbox: Open WhatsApp → (send
join deep-third) - Send any link to the bot and it processes through the full cognitive pipeline
| Method | Endpoint | Description |
|---|---|---|
GET |
/api/memories |
List all memories (search, filter, sort) |
GET |
/api/memories/{id} |
Get memory with connections |
POST |
/api/process?url=... |
Process URL through cognitive pipeline |
DELETE |
/api/memories/{id} |
Delete a memory |
GET |
/api/graph |
Full knowledge graph (nodes + edges) |
GET |
/api/resurfaced |
Recently resurfaced insights |
GET |
/api/stats |
Dashboard statistics |
GET |
/api/categories |
Category distribution |
GET |
/api/random |
Random memory for inspiration |
POST |
/api/whatsapp/webhook |
Twilio WhatsApp webhook |
- Instagram — Reels, Posts (via oEmbed + meta tags) — inline embed player
- Twitter / X — Tweets, Threads (via oEmbed)
- YouTube — Videos, Shorts (via oEmbed) — inline embed player
- Reddit — Posts (via JSON API)
- TikTok — Videos (via oEmbed)
- Blog / Articles — Any URL (via content extraction + BeautifulSoup)
- URL with no extractable content → falls back to meta tags → then to URL itself
- No OpenAI key → keyword-based classification fallback
- sentence-transformers not installed → TF-IDF fallback → hash-based fallback
- Empty database → graceful empty states in dashboard
- Duplicate URL → detected with URL normalization, user notified instead of re-saving
- Concurrent duplicate submissions → thread-safe locking prevents race conditions
- Duplicate connections → prevented at database level
- Resurfacing cooldown → same memory won't resurface within 7 days
- WhatsApp message with no URL → help reply sent
- Multiple URLs in single message → processes up to 3
- Invalid URL (no http/https) → 400 error with clear message
- Image load failure → gracefully hidden in dashboard
- API timeout → retry-safe, idempotent operations
- Cascade delete → deleting memory removes connections + resurfaced entries
Thredion is designed to work completely offline with zero API keys:
| Component | Primary | Fallback 1 | Fallback 2 |
|---|---|---|---|
| Embeddings | sentence-transformers (MiniLM-L6-v2) | TF-IDF (sklearn) | Hash-based (MD5) |
| Classification | OpenAI GPT-3.5 | Keyword matching (20 categories) | — |
| Extraction | Platform oEmbed API | HTML content scraping | Meta tag fallback |
This 3-tier architecture ensures the system never crashes — even without internet access or API keys.
thredion/
├── assets/ # README images
│ ├── dashboard-preview.png # Dashboard screenshot
│ └── architecture-diagram.png # System architecture diagram
│
├── thredion-engine/ # Python backend
│ ├── main.py # FastAPI app entry point
│ ├── requirements.txt
│ ├── .env.example
│ ├── core/
│ │ └── config.py # Central configuration
│ ├── db/
│ │ ├── database.py # SQLAlchemy setup
│ │ └── models.py # ORM models (with unique URL constraint)
│ ├── models/
│ │ └── schemas.py # Pydantic schemas
│ ├── api/
│ │ ├── routes.py # REST API endpoints
│ │ └── whatsapp.py # Twilio webhook
│ ├── services/
│ │ ├── pipeline.py # Full cognitive pipeline orchestrator
│ │ ├── extractor.py # URL content extraction
│ │ ├── embeddings.py # Vector embedding generation
│ │ ├── classifier.py # AI classification & summarization
│ │ ├── knowledge_graph.py # Graph builder
│ │ ├── importance.py # Explainable importance scoring
│ │ └── resurfacing.py # Smart resurfacing engine
│ └── tests/ # 93 automated tests
│ ├── conftest.py # Shared fixtures (in-memory SQLite)
│ ├── test_api.py # API endpoint tests
│ ├── test_database.py # Database model tests
│ ├── test_embeddings.py # Embedding generation tests
│ ├── test_pipeline.py # Cognitive pipeline tests
│ ├── test_services.py # Service module tests
│ └── test_demo_reliability.py # Demo failure scenario tests
│
├── thredion-dashboard/ # Next.js frontend
│ ├── src/
│ │ ├── app/
│ │ │ ├── layout.tsx
│ │ │ ├── page.tsx # Main dashboard page
│ │ │ └── globals.css
│ │ ├── components/
│ │ │ ├── Header.tsx
│ │ │ ├── MemoryCard.tsx
│ │ │ ├── StatsBar.tsx
│ │ │ ├── CategoryFilter.tsx
│ │ │ ├── ResurfacedPanel.tsx
│ │ │ ├── KnowledgeGraphView.tsx
│ │ │ ├── StatsView.tsx
│ │ │ └── InspireModal.tsx
│ │ └── lib/
│ │ ├── api.ts # API client (with timeout & retry)
│ │ ├── types.ts # TypeScript types
│ │ ├── utils.ts # Helpers
│ │ └── __tests__/
│ │ └── test-plan.ts # Frontend integration test plan
│ ├── package.json
│ ├── tailwind.config.js
│ └── next.config.js
│
└── README.md
93 automated tests covering all critical paths:
cd thredion-engine
python -m pytest tests/ -v| Test File | Tests | Covers |
|---|---|---|
test_api.py |
23 | All REST endpoints, CRUD, search, filters, error handling |
test_database.py |
15 | ORM models, relationships, cascade delete, constraints |
test_embeddings.py |
12 | 3-tier embedding fallback, cosine similarity, edge cases |
test_pipeline.py |
12 | Full pipeline, duplicate detection, thread safety |
test_services.py |
19 | Extractor, classifier, knowledge graph, importance, resurfacing |
test_demo_reliability.py |
12 | Startup resilience, timeout handling, concurrent requests |
| Component | Platform | URL |
|---|---|---|
| Backend API | Azure App Service (Free F1) | thredion-api.azurewebsites.net |
| Dashboard | Vercel | thredion.vercel.app |
| WhatsApp Bot | Twilio Sandbox | Chat on WhatsApp → |
| API Docs | Swagger UI | thredion-api.azurewebsites.net/docs |
Thredion evolves beyond a saver into a cognitive operating system for human memory augmentation:
- Scheduled "Memory Digest" emails
- Voice-note memory capture
- Browser extension for instant saves
- Collaborative knowledge graphs
- Advanced RAG for Q&A over your saved knowledge
Hack The Thread — Turning Instagram Saves into a Knowledge Base
MIT