One command. Zero infrastructure. Your AI agent builds and maintains a compounding knowledge base — automatically.

npx greatmind init

Then in your AI assistant (OpenCode, Claude Code, or Codex):

/braindump "Ghostship raised $5M — rethink positioning"
/consolidate-knowledge
/daily-brief

📋 Table of Contents

• Quickstart
• Why
• Features
• Commands
• Architecture
• File Layout
• Comparison
• Requirements
• Contributing
• License

🚀 Quickstart

Install

npx greatmind init

Or install globally:

npm install -g greatmind
greatmind init

Choose your setup

The CLI walks you through two interactive choices:

| Choice | Options | Effect | |--------|---------|--------| | Scope | workspace / user | Where config files go: .greatmind/ or ~/.config/greatmind/ | | Assistant | opencode / claude-code / codex / none | Which AI assistant to wire |

No detection. No guessing. Just tell it what you use.

To remove GreatMind later: greatmind uninstall — interactively de-registers assistants and optionally deletes data.

Start using

/braindump "Ghostship just raised $5M in seed funding for no-code testing"

Your assistant will:

1. Read the schema to understand conventions
2. Classify the thought (domain, entities, projects)
3. File it as an immutable raw source
4. Create or update entity and concept pages
5. Update the page catalog and activity log

💡 Why

Andrej Karpathy's LLM Wiki pattern flips knowledge management:

Instead of the LLM re-deriving knowledge from raw files on every query (RAG), it compiles knowledge into a structured, interlinked wiki that compounds over time.

GreatMind scaffolds this pattern into your project and wires it into your AI assistant. The LLM does the organizing, cross-referencing, and synthesizing — you just dump thoughts.

"The AI should be writing and maintaining the wiki, not the human. The human's job is to dump raw thoughts." — Andrej Karpathy

✨ Features

| Feature | What it does | |---------|-------------| | 🧠 Braindump | Dump raw thoughts — LLM classifies, files, and cross-references automatically | | 🔗 Auto-linking | Every wiki page gets 2+ [[wikilinks]] — no orphan pages | | 📐 Synthesis | Consolidates scattered braindumps into structured entity, concept, and comparison pages | | ⚡ Daily Brief | Personalized briefings grounded in your recent notes | | 📊 Weekly Check-in | Cross-domain theme analysis and trend detection | | 🔍 Contradiction Resolution | Flags and resolves conflicting claims across sources | | 🧹 Self-healing Lint | Automatic detection of orphan pages, broken links, and stale claims | | 🔌 Multi-assistant | Works with OpenCode, Claude Code, and Codex CLI out of the box | | 📁 Git-native | Everything is plain Markdown — commit it, diff it, open it in Obsidian |

📟 Commands

Terminal commands

| Command | What it does | |---------|-------------| | greatmind init | Scaffold GreatMind in the current directory (workspace or user level) | | greatmind update | Refresh AGENTS.md, SCHEMA.md, and prompt files; preserves wiki data | | greatmind uninstall | Interactively de-register assistants and optionally delete GreatMind data |

Slash commands

Once installed, use these in your AI assistant:

/braindump

Capture a raw thought. No structure needed.

/braindump Been thinking about our testing bottleneck. We spend 40% of sprint time on
manual regression. What if we auto-generate integration tests from OpenAPI specs?

The LLM classifies, files, extracts entities, and cross-references against the existing wiki.

/daily-brief

Get a personalized briefing grounded in your recent notes and interests.

/daily-brief

Output: executive summary, recent signals with citations, strategic implications, action items.

/weekly-checkin

Analyze the past week — themes, patterns, cross-domain insights.

/weekly-checkin

/consolidate-knowledge

The flagship command. Synthesize all unconsolidated braindumps into structured wiki pages.

/consolidate-knowledge

The LLM reads every new braindump since the last consolidation, identifies themes, creates entity/concept/comparison/synthesis pages, cross-links them, resolves contradictions, and updates the index and log.

Tip: After updating to a new version, run greatmind update in your terminal to refresh instructions and prompts while preserving your wiki data.

🏗 Architecture

GreatMind implements Karpathy's three-layer architecture:

┌─────────────────────────────────────────────────┐
│                   YOU                             │
│  Write raw thoughts, ask questions                │
└─────────────────────┬───────────────────────────┘
                      │ /braindump
                      ▼
┌─────────────────────────────────────────────────┐
│              RAW SOURCES (wiki/raw/)              │
│  Immutable braindumps — you write, LLM reads     │
│  but never modifies                               │
└─────────────────────┬───────────────────────────┘
                      │ /consolidate-knowledge
                      ▼
┌─────────────────────────────────────────────────┐
│              THE WIKI (wiki/)                     │
│  Structured, interlinked Markdown pages           │
│  LLM creates, updates, cross-references           │
│   ├── entities/  ─ People, companies, tools       │
│   ├── concepts/  ─ Ideas, methodologies           │
│   ├── comparisons/ ─ A vs B analyses              │
│   ├── synthesis/ ─ Consolidated frameworks        │
│   └── sessions/  ─ Session digests                │
└─────────────────────┬───────────────────────────┘
                      │ Reads during execution
                      ▼
┌─────────────────────────────────────────────────┐
│              THE SCHEMA (AGENTS.md)               │
│  Page types, workflows, conventions               │
│  Updated via `greatmind update`                   │
└─────────────────────────────────────────────────┘

| Layer | Location | Who owns it | Purpose | |-------|----------|-------------|---------| | Raw sources | wiki/raw/ | You | Immutable braindumps. You write; the LLM reads but never modifies. | | The wiki | wiki/ | LLM | Structured, interlinked Markdown pages. The LLM creates, updates, and cross-references them. | | The schema | AGENTS.md (in <prefix>/) | LLM | Page types, workflows, conventions. Updated via greatmind update. |

📁 File Layout

GreatMind supports two installation scopes. You choose during greatmind init:

Workspace (per-project — .greatmind/ in your project root):

your-project/
├── .greatmind/
│   ├── AGENTS.md                  ← Schema: conventions, rules, workflows
│   ├── SCHEMA.md                  ← Detailed wiki schema (read on-demand)
│   ├── prompts/                   ← Slash command definitions
│   │   ├── braindump.md
│   │   ├── daily-brief.md
│   │   ├── weekly-checkin.md
│   │   └── consolidate-knowledge.md
│   └── wiki/
│       ├── raw/                   ← Braindumps (immutable, LLM read-only)
│       ├── entities/              ← People, companies, tools
│       ├── concepts/              ← Ideas, methodologies
│       ├── comparisons/           ← A vs B analyses
│       ├── synthesis/             ← Consolidated frameworks
│       ├── sessions/              ← Session-level digests (crystallization)
│       ├── index.md               ← Page catalog (auto-updated)
│       └── log.md                 ← Activity log (append-only)
├── opencode.json                  ← Updated with instruction reference
└── .opencode/commands/            ← Registered slash commands

User (all projects — ~/.config/greatmind/):

~/.config/greatmind/
├── AGENTS.md                      ← Same structure, shared across projects
├── SCHEMA.md
├── prompts/
└── wiki/

📊 Comparison

| | Traditional wiki | GreatMind | |---|---|---| | Filing | You manually organize | LLM classifies and files automatically | | Cross-references | You maintain links | LLM links every page with 2+ [[wikilinks]] | | Synthesis | You compile insights | LLM consolidates across braindumps | | Maintenance | You clean up | LLM lints orphans, broken links, contradictions | | Format | Plain .md files | Plain .md files — check into git, open in Obsidian |

📋 Requirements

• Node.js 18+
• An AI coding assistant: OpenCode, Claude Code, or Codex CLI
• (Optional) Obsidian — browse the wiki visually

🤝 Contributing

We welcome contributions! GreatMind is in early alpha — every issue, PR, and suggestion helps shape the direction.

Getting started

git clone https://github.com/wengpaul/greatmind.git
cd greatmind
npm install
npm run build

Project structure

src/
├── index.ts       ← CLI entry point
├── init.ts        ← Scaffold & update logic
├── install.ts     ← Assistant wiring (opencode, claude-code, codex)
├── uninstall.ts   ← Interactive uninstall
└── templates.ts   ← Inline template strings

Guidelines

• Keep the three-layer architecture intact — raw sources are immutable
• All behavioral changes belong in AGENTS.md and src/templates.ts (they must stay in sync)
• No runtime dependencies on files — templates are inlined in src/templates.ts
• PRs welcome for: new assistant integrations, improved prompts, lint rules, and documentation

📄 License

MIT © Paul Weng