How it Works

Overview

Memex gives your AI coding assistants long-term memory by:

1. Collecting raw conversations from Claude Code, Codex, and other tools
2. Storing everything in a local SQLite database (never losing original data)
3. Optionally generating multi-level summaries for better search
4. Searching via full-text, semantic vectors, or both

┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│  JSONL/JSON │ ──▶ │   SQLite    │ ──▶ │   Search    │
│   (source)  │     │  (messages) │     │  FTS + MCP  │
└─────────────┘     └──────┬──────┘     └─────────────┘
                          │
                          ▼ (optional)
                   ┌─────────────┐     ┌─────────────┐
                   │   Compact   │ ──▶ │  LanceDB    │
                   │  L1/L2/L3   │     │  (vectors)  │
                   └─────────────┘     └─────────────┘

"Never Compacted" means: your original conversations are always preserved. The Compact layer is optional and additive — it generates summaries to enhance search, but never replaces the source data.

Contents

• Collection - Where data comes from and how it's collected
• Storage - SQLite and LanceDB architecture
• Compact - Multi-level summary system (L0-L3)
• Search - Full-text, semantic, and hybrid search