Mega Brain accepts any format of expert content and transforms it into structured knowledge. This guide walks you through ingesting different material types into the system.
Overview The ingestion process automatically:
Downloads and transcribes content
Extracts metadata (author, title, date)
Saves files with source traceability
Detects duplicates to prevent reprocessing
Prepares materials for the processing pipeline
All ingested materials are saved to the inbox/ directory organized by person and content type.
Supported Material Types YouTube Videos
PDFs & Documents
Podcasts & Audio
Courses & Training
Ingesting YouTube Videos The most common source. JARVIS automatically downloads transcriptions using the YouTube Transcript API. /ingest https://www.youtube.com/watch?v=EXAMPLE_ID
What happens:
Video metadata is fetched (title, author, duration)
Transcript is downloaded automatically
File is saved as inbox/[PERSON]/[TYPE]/video-title.txt
A unique SOURCE_ID is generated (e.g., CG003 for Cole Gordon video 3)
Example output: JARVIS: Material received. Source: YouTube Title: "How to Create Irresistible Offers" Author: Alex Hormozi Duration: 42:15 Words: 6,230 Saved to: inbox/alex-hormozi/MASTERCLASS/how-to-create-irresistible-offers.txt Next step: execute /process-jarvis to process.
Ingesting PDFs and Documents Upload local files including PDFs, Word documents, and text files. /ingest /path/to/playbook-vendas.pdf
Supported formats:
PDF documents
Microsoft Word (.doc, .docx)
Plain text (.txt, .md)
Google Docs (via URL)
For best results, ensure PDFs have text layers (not just scanned images).
Ingesting Podcasts and Audio Process audio content through transcription. /ingest https://podcast-url.com/episode-123
Requirements:
OPENAI_API_KEY configured (uses Whisper for transcription) Content detection: The system automatically detects podcast episodes based on keywords (“podcast”, “episode”, “ep”, “interview”), file metadata, and URL patterns.Ingesting Course Materials Process complete training courses with multiple modules. /ingest /path/to/course-module-1.pdf --type COURSE
Scope detection:
Course materials are automatically tagged with scope:
scope: "course" - Complete training programscorpus: "[course-name]" - Groups related modules Organization: inbox/ └── setterlun/ └── COURSES/ ├── module-01-foundation.txt ├── module-02-strategy.txt └── module-03-execution.txt
Advanced Ingestion Options Override auto-detected metadata:
# Specify person explicitly /ingest https://youtube.com/watch?v=abc123 --person "Cole Gordon" # Set content type /ingest /path/to/file.txt --type MASTERCLASS # Ingest and immediately process /ingest https://youtube.com/watch?v=abc123 --process
Content Type Classification
Type Auto-Detection Keywords Use Case PODCAST ”podcast”, “episode”, “interview” Conversational content MASTERCLASS ”masterclass”, “mastermind”, “training” Expert workshops COURSE ”course”, “module”, “lesson” Structured programs BLUEPRINT ”blueprint”, “pdf”, “playbook” Strategic documents VSL ”vsl”, “webinar”, “sales letter” Sales presentations SCRIPT ”script”, “template”, “copy” Templates and frameworks
Known Source Detection The system recognizes expert sources automatically:
Path Analysis
Analyzes the file path or URL to detect known experts: inbox/COLE GORDON/MASTERMINDS/video.txt ↓ SOURCE_PERSON: Cole Gordon SOURCE_COMPANY: Cole Gordon CORPUS: closers_io
Known Sources Library
Detected Pattern Person Company Corpus ”hormozi”, “acquisition” Alex Hormozi Alex Hormozi acquisition_com ”cole gordon”, “closers” Cole Gordon Cole Gordon closers_io ”leila” Leila Hormozi Alex Hormozi acquisition_com ”setterlun”, “sam ovens” Sam Ovens Setterlun University setterlun
Automatic Corpus Assignment
Related materials from the same expert are grouped into a corpus for cross-referencing.
Duplicate Detection The pipeline includes 6 levels of duplicate detection to prevent reprocessing:
Detection Levels
Exact MD5 Match - Same file already processedContent Hash - Different file, same contentPartial Match - Content is substring of processed materialYouTube ID - Same video from different sourceFile Registry - Cross-reference with processing historyChunk Fingerprint - First 1000 characters comparison
When duplicates are found: ┌─────────────────────────────────────────────────────────┐ │ ⛔ DUPLICATE DETECTED - PROCESSING INTERRUPTED │ ├─────────────────────────────────────────────────────────┤ │ │ │ Current file: inbox/new-video.txt │ │ MD5: a1b2c3d4e5f6... │ │ │ │ Duplicate of: inbox/old-video.txt │ │ Registered: 2026-02-15 │ │ SOURCE_ID: CG005 │ │ │ │ ACTION: File will NOT be processed. │ │ │ └─────────────────────────────────────────────────────────┘
Batch Ingestion Process multiple materials at once:
Ingest Multiple URLs
Ingest Directory
# List of YouTube videos /ingest https://youtube.com/watch?v=video1 /ingest https://youtube.com/watch?v=video2 /ingest https://youtube.com/watch?v=video3
Verification Check Inbox Status View pending materials:
Output: ═══════════════════════════════════════════════ INBOX STATUS ═══════════════════════════════════════════════ 📥 PENDING: 3 files Cole Gordon (2 files) ├─ masterclass-closing.txt (6,234 words) └─ podcast-interview.txt (8,901 words) Alex Hormozi (1 file) └─ offer-creation-framework.txt (4,567 words) ⭐ NEXT: /process-jarvis to process pending files
Check System Status Provides:
Health score (0-100)
Processed materials count
Pending inbox files
Active agents status
Next Steps
Process Ingested Materials Learn how to run the 5-phase processing pipeline on ingested materials
View Processing Results Check processing logs and generated artifacts
