AI Knowledge Management System
Transform expert materials into structured playbooks, DNA schemas, and mind-clone agents.
Mega Brain is a Claude Code-powered system that ingests expert materials — videos, PDFs, transcriptions, podcasts, training courses — and transforms them into structured knowledge. It produces playbooks, DNA schemas, and AI agents that reason with traced evidence.
Built for solo entrepreneurs and small teams who want to operationalize the expertise they have accumulated across dozens of courses, mentors, and resources.
# 1. Install dependencies (only in the first run)
npm install
# 2. Install and configure
npx mega-brain-ai setup
# 3. Fill in API keys when prompted (only OPENAI_API_KEY is required)
# 4. Open Claude Code and check system status
/jarvis-briefingSetup auto-triggers on first use if .env is missing.
| Requirement | Version | Notes |
|---|---|---|
| Claude Code | Max or Pro plan | Core runtime |
| Node.js | >= 18.0.0 | CLI and tooling |
| Python | >= 3.10 | Intelligence scripts |
| Key | Purpose | Required? |
|---|---|---|
OPENAI_API_KEY |
Whisper transcription | Yes |
VOYAGE_API_KEY |
Semantic embeddings (RAG) | Recommended |
GOOGLE_CLIENT_ID |
Google Drive import | Optional |
Run /setup in Claude Code to configure keys interactively. Keys are stored in .env (gitignored).
- Ingest any format — videos, PDFs, transcriptions, podcasts, training courses
- Extract structured DNA across 5 knowledge layers (philosophies, mental models, heuristics, frameworks, methodologies)
- Build playbooks, dossiers, and theme-based knowledge bases with full source traceability
- Mind Clones — AI agents that reason like specific experts, grounded in their actual materials
- Cargo Agents — Functional role agents (Sales, Marketing, Operations, Finance) that synthesize knowledge from multiple sources
- Conclave — Multi-agent deliberation sessions with evidence-based debate and structured output
- 20+ hooks for automated validation, session management, and quality control
- Slash commands for common operations (
/ingest,/save,/resume,/conclave) - Skill system with keyword-based auto-routing
- Session persistence with auto-save and resume
mega-brain/
├── core/ -> Processing engine (tasks, workflows, protocols, schemas)
├── agents/ -> AI agent definitions (conclave, cargo, minds, templates)
├── bin/ -> CLI tools and entry points
├── .claude/ -> Claude Code integration (hooks, skills, commands, rules)
├── knowledge/ -> Knowledge base (playbooks, dossiers, DNA schemas)
├── artifacts/ -> Pipeline processing stages (chunks, insights, narratives)
├── inbox/ -> Raw materials input directory
├── docs/ -> Documentation, PRDs, plans
└── logs/ -> Session and processing logs
Content is organized into three distribution layers:
| Layer | Content | Distribution |
|---|---|---|
| L1 (Community) | Core engine, templates, hooks, skills, CLI | npm package (public) |
| L2 (Pro) | Populated knowledge base, mind clones, pipeline | Premium (tracked) |
| L3 (Personal) | Your materials, sessions, environment config | Local only (gitignored) |
| Feature | Community (L1) | Pro (L2) |
|---|---|---|
| CLI and setup wizard | Yes | Yes |
| Core engine and templates | Yes | Yes |
| Skills and hooks | Yes | Yes |
| Agent templates and examples | Yes | Yes |
| Populated knowledge base | -- | Yes |
| Mind clone agents | -- | Yes |
| Pipeline processing | -- | Yes |
| Council / Conclave | -- | Yes |
Use these slash commands inside Claude Code:
| Command | Description |
|---|---|
/jarvis-briefing |
System status and health score |
/ingest |
Ingest new material into the pipeline |
/process-jarvis |
Run the 5-phase processing pipeline |
/conclave |
Multi-agent deliberation session |
/save |
Save current session state |
/resume |
Resume previous session |
/setup |
Environment setup wizard |
Knowledge is extracted and structured in 5 layers:
| Layer | Name | Description |
|---|---|---|
| L1 | Philosophies | Core beliefs and worldview |
| L2 | Mental Models | Thinking and decision frameworks |
| L3 | Heuristics | Practical rules and decision shortcuts |
| L4 | Frameworks | Structured methodologies and processes |
| L5 | Methodologies | Step-by-step implementations |
Every piece of extracted knowledge traces back to its source material with file path, line number, and original context.
Verify the package before publishing:
# Check that only L1 content is in the package
npm run validate:layers
# Full pre-publish gate (secrets scan + layer validation)
node bin/pre-publish-gate.jsSee CONTRIBUTING.md for guidelines.
UNLICENSED — See package.json for details.