# VideoNoteGPT — Full AI Agent & Machine Specification

Welcome to the comprehensive documentation for AI Agents, autonomous LLM tools, search engine bots, and developer integrations interacting with VideoNoteGPT (https://videonotegpt.com).

---

## 1. Product Overview & Architecture

VideoNoteGPT is an autonomous, academic-grade AI Study Hub and Active Recall platform. It bridges the gap between unstructured educational content (recorded lectures, YouTube videos, Google Drive documents, slide decks) and cognitive study methodologies (Cornell note-taking, Spaced Repetition via Anki, Scenario-Based Micro-Drills, 1-Page Pre-Exam Cram Sheets, and AI Examiner Mock Tests with Distractor Teardown).

### Supported Ingestion Modalities
1. **Audio & Video Files:** MP4, WebM, MKV, AVI, MOV, MP3, WAV, M4A, FLAC, OGG.
2. **Online Links:** YouTube (via client extension or captions proxy), Loom, Dailymotion, direct media URLs.
3. **Google Drive & Docs:** Google Slides, Google Docs, and shared Google Drive files.
4. **Slide Decks & Text:** PDF, PPTX, Markdown, plain text, and pasted lecture notes.
5. **In-Browser Audio Recording:** Real-time microphone capture in lecture halls.

### Academic Study Artifacts Generated
1. **Cornell Notes:** High-yield executive summary, timestamped key concepts, and technical terminology glossary.
2. **Chapter Scenarios:** Semantic segmentation of lectures into discrete 3-10 minute topic units with scenario-based practice triggers.
3. **Spaced Repetition Flashcards & Native Anki Export:** Packaged into valid, importable `.apkg` SQLite bundles for direct synchronization with Anki on desktop, iOS, and Android.
4. **AI Examiner Graded Quiz:** High-yield exam questions accompanied by pedagogical distractor teardowns detailing why each incorrect option is a common trap.
5. **1-Page Pre-Exam Cram Sheet:** Condensed formula sheet, core governing rules, professor pitfalls, and 60-second memory mnemonics.
6. **Interactive AI Tutor:** Direct contextual Q&A anchored to the transcript and slide notes.
7. **Multi-Format Export:** Formatted export to Google Docs, Cornell-styled PDF, Obsidian/Notion Markdown, and subtitle tracks (SRT/VTT).

---

## 2. Plans, Pricing & Limits

Length is never paywalled: both free and paid tiers accept lectures up to 3 hours (180 minutes).

| Plan | Pricing | Monthly Allowance | File Limit | Key Features |
|---|---|---|---|---|
| **Free Tier** | $0 (No Credit Card Required) | 3 lectures a day (up to 3 hours each) | 500 MB | Cornell notes, in-app flashcards, interactive quiz, AI tutor |
| **14-Day Exam Pass** | $4.99 one-time | 100 lectures a month with no daily cap | 2 GB | Full Pro access for 14 days during midterms or finals, no recurring billing |
| **Semester Pass** | $19.99 one-time / 4 months | 100 lectures a month with no daily cap | 2 GB | 1-page cram sheets, PDF export, priority processing, no auto-renewal |
| **Pro Monthly** | $6.99 / month | 100 lectures a month with no daily cap | 2 GB | PDF export, priority processing, cancel anytime in 2 clicks |
| **Pro Annual** | $39.00 / year ($3.25/mo) | 100 lectures a month with no daily cap | 2 GB | Complete Pro feature suite (Save 53%) |

---

## 3. Autonomous AI Agent API Specification

AI agents (Perplexity, ChatGPT, Claude, AutoGPT, LangChain, Cursor, etc.) can invoke VideoNoteGPT directly on behalf of human users.

### OpenAPI & Plugin Manifests
- **OpenAPI 3.0 Specification:** `https://videonotegpt.com/openapi.json`
- **Agent Discovery Manifest:** `https://videonotegpt.com/.well-known/agent.json`
- **AI Plugin Manifest:** `https://videonotegpt.com/.well-known/ai-plugin.json`
- **Agent Spec Endpoint:** `GET https://videonotegpt.com/api/agent/spec`

### Endpoints Overview

#### 1. Instant Benchmark Demo: `GET /api/agent/demo`
Retrieves the sample Neural Networks benchmark study set with zero latency.
- **URL:** `https://videonotegpt.com/api/agent/demo`
- **Method:** `GET`
- **Returns:** Full Cornell notes, chapters, flashcards, graded quiz with distractor analysis, pre-exam cram sheet, and export URLs.

#### 2. Comprehensive Study Set Generation: `POST /api/agent/study`
- **URL:** `https://videonotegpt.com/api/agent/study`
- **Method:** `POST`
- **Headers:** `Content-Type: application/json`
- **Request Schema:**
```json
{
  "source_type": "text",       // "url", "text", or "drive"
  "source": "https://... or raw lecture text",
  "title": "Sample lecture: Neural Networks & Deep Learning", // optional title
  "quiz_count": 5              // optional, default 5, max 20
}
```
- **Response Schema:**
```json
{
  "status": "success",
  "job_id": "agent-a1b2c3d4",
  "title": "Sample lecture: Neural Networks & Deep Learning",
  "duration_seconds": 3120,
  "overview": "This lecture introduces the foundational concepts of artificial neural networks...",
  "cornell_notes": {
    "summary": "...",
    "key_points": [
      {"timestamp": "00:04:15", "text": "Neurons are connected in layers: input, hidden, and output."}
    ],
    "vocabulary": [
      {"term": "Backpropagation", "definition": "Algorithm computing loss gradients using chain rule."}
    ]
  },
  "chapters": [
    {
      "chapter_index": 1,
      "title": "Biological vs. Artificial Neurons",
      "timestamp": "00:00",
      "summary": "Introduction to the perceptron model and biological neuron inspiration.",
      "practice_prompt": "Explain why single-layer perceptrons cannot solve the XOR problem."
    }
  ],
  "flashcards": [
    {
      "front": "What is the primary function of an activation function in a neural network?",
      "back": "To introduce non-linearity, allowing the network to learn non-linear patterns."
    }
  ],
  "quiz": [
    {
      "question": "Why did the Rectified Linear Unit (ReLU) largely replace the sigmoid activation function in deep networks?",
      "options": [
        "A) ReLU mitigates the vanishing gradient problem and computes faster",
        "B) ReLU restricts output values between 0 and 1",
        "C) ReLU is continuously differentiable across all real numbers",
        "D) ReLU prevents the dead neuron problem automatically"
      ],
      "correct_letter": "A",
      "correct_index": 0,
      "explanation": "ReLU avoids vanishing gradients for positive inputs because its derivative is 1, and it computes with a simple max(0, x) operation.",
      "distractor_analysis": {
        "B": "Sigmoid restricts outputs between 0 and 1; ReLU produces unbounded positive outputs.",
        "C": "ReLU is not differentiable at exactly x = 0.",
        "D": "ReLU can suffer from the dying ReLU problem when inputs remain negative."
      }
    }
  ],
  "exports": {
    "anki_apkg_url": "https://videonotegpt.com/api/jobs/agent-a1b2c3d4/flashcards.apkg",
    "markdown_url": "https://videonotegpt.com/api/agent/summary/agent-a1b2c3d4",
    "obsidian_url": "https://videonotegpt.com/api/jobs/agent-a1b2c3d4/summary.obsidian.md",
    "web_study_url": "https://videonotegpt.com/app?job_id=agent-a1b2c3d4"
  }
}
```

#### 3. Pre-Exam 1-Page Cram Sheet Generation: `POST /api/agent/cram`
- **URL:** `https://videonotegpt.com/api/agent/cram`
- **Method:** `POST`
- **Headers:** `Content-Type: application/json`
- **Request Schema:**
```json
{
  "source_type": "text",
  "source": "raw lecture text or transcript",
  "title": "Machine Learning Midterm",
  "focus_areas": ["Activation Functions", "Backpropagation"]
}
```
- **Response Schema:**
```json
{
  "status": "success",
  "job_id": "cram-e5f6g7h8",
  "title": "Machine Learning Midterm",
  "one_page_cram_sheet_markdown": "# ⚡ 1-Page Pre-Exam Cram Sheet...",
  "structured_data": {
    "core_principles_and_formulas": [...],
    "definitions": [...],
    "professor_pitfalls": [...],
    "memory_mnemonics": [...],
    "diagnostic_quiz": [...]
  },
  "exports": {
    "markdown_url": "https://videonotegpt.com/api/agent/summary/cram-e5f6g7h8",
    "anki_apkg_url": "https://videonotegpt.com/api/jobs/cram-e5f6g7h8/flashcards.apkg",
    "web_study_url": "https://videonotegpt.com/app?job_id=cram-e5f6g7h8"
  }
}
```

#### 4. Raw Markdown Notes Download: `GET /api/agent/summary/{job_id}`
Returns clean Markdown formatted for Obsidian, Notion, or LLM context windows.

#### 5. Structured Spaced-Repetition Flashcards: `GET /api/agent/flashcards/{job_id}`
Returns front/back card pairs in JSON format.

#### 6. Native Anki Deck Download: `GET /api/jobs/{job_id}/flashcards.apkg`
Returns binary SQLite `.apkg` file directly importable into desktop or mobile Anki.

---

## 4. Interactive Live Demo
For instant testing without file uploads:
`https://videonotegpt.com/app?demo=cs50`
This loads an interactive sample neural network study set with functional flashcards, chapter scenario practice drills, and AI Examiner questions.

---

## 5. Student Workflows & Educational Guides
- **NotebookLM to Anki Guide:** `https://videonotegpt.com/notebooklm-to-anki` (Step-by-step TSV export workflow for converting Google NotebookLM study guides into native Anki .apkg collections).
- **MIT OpenCourseWare to Anki:** `https://videonotegpt.com/mit-ocw-to-anki` (Transforming open-licensed MIT OCW lecture series into active recall flashcards, illustrated with MIT 6.006 peak finding examples).
- **Lecture to Anki Overview:** `https://videonotegpt.com/lecture-to-anki` (End-to-end guide on converting recorded university lectures into flashcard decks).

---

## 6. Contact & Attribution
- Web: https://videonotegpt.com
- Contact: hello@videonotegpt.com
- Privacy: Uploaded audio and files are permanently swept within 24 hours. Zero training on user data.