MyMemex

Your AI Document Memory — For 80 years it was science fiction. Now, it's yours.

mymemex.app • GitHub • Documentation

---

What is MyMemex?

MyMemex transforms your personal document archive (PDFs, scans, images) into an AI-powered memory system. Search with natural language, auto-extract structured data, and chat with your documents through MCP.

Privacy-first: Run fully air-gapped with local LLMs via Ollama. Your documents never leave your machine.

Named after Vannevar Bush's 1945 vision of a personal knowledge system — finally made real with AI.

Core Features

• Intelligent Ingestion — Watch directories, auto-detect files, deduplicate by hash
• 100% Private — Local OCR (Tesseract) + local LLMs (Ollama). No cloud required.
• Semantic Search — Ask questions in natural language, get instant answers with citations
• Auto-Extraction — AI extracts amounts, dates, entities, and categories automatically
• Tag-Based Filing — Automatic routing rules move files to tag-matched subdirectories
• Multi-User Auth — JWT-based login with per-user access control
• Admin Panel — Watch folders, routing rules, queue management, logs, backup
• MCP Integration — Works with Claude Desktop, OpenClaw, and other AI assistants via stdio or HTTP transport

Quick Start

Docker (Recommended)

# Pull and run
docker run -d -p 8000:8000 ghcr.io/joywareapps/mymemex:latest

# Or with docker-compose
curl -O https://raw.githubusercontent.com/joywareapps/mymemex/main/docker-compose.yml
docker-compose up -d

Visit http://localhost:8000/ui/ to access the web UI.

From Source

# Clone
git clone https://github.com/joywareapps/mymemex.git
cd mymemex

# Install
pip install -e ".[all]"

# Configure
cp config/config.example.yaml config/config.yaml
# Edit config.yaml with your document paths

# Run
mymemex serve

Personal Instance (Private Deploy)

Use scripts/deploy-private.sh with a .env file:

LIBRARY_PATH=/path/to/your/library   # must contain inbox/ and archive/
PRIVATE_HTTP_PORT=8002
MYMEMEX_LLM__PROVIDER=ollama
MYMEMEX_LLM__API_BASE=http://your-ollama-host:11434
MYMEMEX_LLM__MODEL=gemma3:12b

bash scripts/deploy-private.sh
# Access at http://localhost:8002/ui/

See DEPLOYMENT.md for all deployment options.

MCP for Claude Desktop

stdio transport (local install):

{
  "mcpServers": {
    "mymemex": {
      "command": "mymemex",
      "args": ["mcp", "serve"]
    }
  }
}

HTTP transport (Docker instance):

Enable in Settings → MCP → Transport → http, then:

{
  "mcpServers": {
    "mymemex": {
      "type": "http",
      "url": "http://your-server:8002/mcp"
    }
  }
}

Now Claude can search your documents, extract data, and answer questions about your archive.

Architecture

┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│   Watcher   │────▶│  Ingestion  │────▶│ OCR / Embed │
│ (filesystem)│     │   Pipeline  │     │  (Tesseract)│
└─────────────┘     └─────────────┘     └─────────────┘
                                               │
                                               ▼
                    ┌─────────────────────────────────────┐
                    │           STORAGE LAYER             │
                    │  SQLite (metadata) + ChromaDB       │
                    └─────────────────────────────────────┘
                                               │
                      ┌────────────────────────┼────────────────────────┐
                      ▼                        ▼                        ▼
              ┌─────────────┐          ┌─────────────┐          ┌─────────────┐
              │  Web UI     │          │  REST API   │          │  MCP Server │
              │  (search)   │          │  (CRUD)     │          │  (Claude)   │
              └─────────────┘          └─────────────┘          └─────────────┘

Project Structure

mymemex/
├── src/mymemex/
│   ├── api/            # REST API endpoints
│   ├── core/           # Queue, events, watcher, scheduler
│   ├── processing/     # Ingestion pipeline
│   ├── intelligence/   # Embeddings + OCR
│   ├── services/       # Business logic
│   ├── storage/        # Database + vector store
│   ├── web/            # Web UI routes + templates
│   ├── mcp/            # MCP server (stdio + HTTP)
│   └── middleware/     # Auth, demo mode, MCP auth
├── skills/             # OpenClaw skill definitions
├── website/            # Astro marketing website
├── workers/            # Cloudflare Workers
├── config/             # Configuration
├── tests/              # Test suite (202 tests)
└── docs/               # Documentation

Requirements

Component	Minimum	Recommended
Python	3.10+	3.12+
RAM	4GB	8GB+
Storage	10GB	50GB+ (depends on docs)
Ollama	Optional	For local AI

With Ollama (Local AI)

# Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh

# Pull models
ollama pull nomic-embed-text  # Embeddings (274MB)
ollama pull gemma3:12b        # Chat/reasoning (larger)

Configure via .env or config.yaml:

MYMEMEX_LLM__PROVIDER=ollama
MYMEMEX_LLM__API_BASE=http://localhost:11434
MYMEMEX_LLM__MODEL=gemma3:12b
MYMEMEX_AI__SEMANTIC_SEARCH_ENABLED=true
MYMEMEX_AI__EMBEDDING_MODEL=nomic-embed-text

Without Ollama (Cloud AI)

MYMEMEX_LLM__PROVIDER=openai   # or anthropic
MYMEMEX_LLM__API_KEY=sk-...
MYMEMEX_LLM__MODEL=gpt-4o

Documentation

• DEPLOYMENT.md — All deployment options (local, private, demo, production)
• docs/MILESTONES.md — Project roadmap and history
• docs/ARCHITECTURE.md — Technical deep dive
• docs/SPECIFICATION.md — Full feature specification
• skills/mymemex/SKILL.md — MCP tools reference for OpenClaw

Status

Milestone	Status	Description
M1–M6	✅ Complete	Core features (ingest, OCR, search, classify, extract)
M7	✅ Complete	MCP Server for Claude (stdio)
M8	✅ Complete	Web UI
M9	✅ Complete	Structured extraction + amounts
M10	✅ Complete	Cloud LLM support (OpenAI, Anthropic)
M11	✅ Complete	Admin panel
M12	✅ Complete	Multi-user auth + login
M12.5–M12.7	✅ Complete	AI pause/resume, tag-based routing, MCP HTTP, SQLite retry
M13	🔜 Planned	Chat interface, document Q&A

202 tests passing • Live demo — read-only, no sign-up

License

AGPL v3 + Commercial Dual License

• Free for personal/open-source use (AGPL v3)
• Commercial license available for businesses (contact)

Support the Project

If you find MyMemex useful, consider:

• Starring the repo on GitHub
• Sponsoring development (GitHub Sponsors)

See SUPPORTERS.md for supporter benefits.

---

"Thus science may implement the ways in which man produces, stores, and consults the record of the race." — Vannevar Bush, 1945

---

Live Demo • View on GitHub • Report Issue