Architecture Overview

Understanding the file structure and components of Mega Brain

Mega Brain follows a modular architecture designed to separate concerns between processing engine, knowledge storage, agent systems, and user workspace.

Directory Structure

core/

Processing Engine

Tasks, workflows, protocols, schemas, and intelligence scripts that power the knowledge pipeline.

agents/

AI Agents

Mind clones, cargo roles, conclave deliberation agents, and agent templates.

.claude/

Claude Integration

Hooks, skills, slash commands, and rules for Claude Code integration.

knowledge/

Knowledge Base

Playbooks, dossiers, DNA schemas, and sources with full traceability.

artifacts/

Processing Artifacts

Intermediate pipeline outputs: chunks, insights, narratives, and extractions.

inbox/

Input Directory

Raw materials for processing: videos, PDFs, transcriptions, courses.

Core Components

Processing Engine (core/)

The core engine contains everything needed to transform raw materials into structured knowledge:

core/tasks/

Task definitions for atomic operations:

• extract-dna.md - Extract 5-layer DNA from sources
• analyze-themes.md - Identify themes and patterns
• normalize-entities.md - Resolve entity names
• process-batch.md - Batch processing orchestration
• validate-cascade.md - Validation and integrity checks

core/workflows/

YAML workflow definitions:

• wf-pipeline-full.yaml - Complete 5-phase pipeline
• wf-ingest.yaml - Material ingestion
• wf-extract-dna.yaml - DNA extraction workflow
• wf-conclave.yaml - Multi-agent deliberation

core/schemas/

JSON Schema definitions for all state files:

• chunks-state.schema.json
• insights-state.schema.json
• narratives-state.schema.json
• canonical-map.schema.json
• decisions-registry.schema.json

core/intelligence/

Python intelligence scripts:

• audit_layers.py - Layer system validation
• RAG and semantic processing utilities
• Quality control and validation scripts

Agent System (agents/)

Mega Brain includes a hierarchical agent system with four main types:

agents/
├── minds/          # L3 - Expert mind clones (Hormozi, Cole Gordon, etc.)
├── cargo/          # L4 - Functional roles (Sales, Marketing, Ops, Finance)
├── conclave/       # L1 - Deliberative agents (Critic, Advocate, Synthesizer)
├── boardroom/      # L2 - C-Level strategic debates
├── _templates/     # Official agent templates (V3)
└── AGENT-INDEX.yaml # Master catalog (auto-updated)

See the Agents page for detailed information.

Claude Code Integration (.claude/)

The .claude/ directory contains all Claude Code integration files:

Hooks

20+ lifecycle hooks for automated operations:

• agent_index_updater.py - Auto-update agent catalog
• continuous_save.py - Session auto-save
• enforce_dual_location.py - Dual-location logging
• memory_persister.py - Agent memory updates
• inbox_age_alert.py - Unprocessed material alerts
• creation_validator.py - File creation validation

Commands

Slash commands for common operations:

• /jarvis-briefing - System status
• /ingest - Ingest new material
• /process-jarvis - Run full pipeline
• /conclave - Multi-agent session
• /save - Save session state
• /resume - Resume previous session

Skills

40+ skills with keyword-based routing:

Skills automatically activate based on context and keywords, providing specialized instructions for specific tasks.

Knowledge Base (knowledge/)

All extracted and structured knowledge lives here:

knowledge/
├── dossiers/
│   ├── persons/    # Expert dossiers (ALEX-HORMOZI.md, etc.)
│   └── themes/     # Theme-based compilations
├── playbooks/      # Operational playbooks
├── dna/            # 5-layer DNA schemas
└── sources/        # Source material metadata

Every piece of knowledge traces back to source material with chunk_id, file path, and original context.

Pipeline Artifacts (artifacts/)

Intermediate processing stages store data here:

artifacts/
├── chunks/         # Segmented content with metadata
├── insights/       # Extracted actionable insights
├── narratives/     # Synthesized narratives per person/theme
├── extractions/    # DNA layer extractions
└── canonical/      # Entity resolution maps

State Management

Mega Brain maintains several critical state files:

State File	Purpose	Location
CHUNKS-STATE.json	All content chunks	artifacts/chunks/
CANONICAL-MAP.json	Entity name resolution	artifacts/canonical/
INSIGHTS-STATE.json	Extracted insights	artifacts/insights/
NARRATIVES-STATE.json	Synthesized narratives	artifacts/narratives/
AGENT-INDEX.yaml	Agent registry	agents/
file-registry.json	Processed files tracking	system/REGISTRY/

Binaries and CLI (bin/)

Command-line tools for setup and operations:

• setup.js - Interactive setup wizard
• cli.js - Main CLI entry point
• pre-publish-gate.js - Layer validation before publishing
• validate-layers.js - Layer system integrity check

Development Workflow

1. Add Material

   Place raw materials (PDFs, videos, transcripts) in inbox/

2. Run Pipeline

   Execute /process-jarvis to run the 8-phase processing pipeline

3. Review Artifacts

   Check artifacts/ for chunks, insights, and narratives

4. Access Knowledge

   Query structured knowledge in knowledge/dossiers/ and knowledge/playbooks/

5. Consult Agents

   Use mind clones and cargo agents for expert reasoning

Next Steps

Knowledge Pipeline

Learn about the 8-phase processing pipeline

DNA Schema

Understand the 5-layer knowledge extraction

Layer System

Explore the L1/L2/L3 distribution layers

Agents

Deep dive into the agent system