Memex

Memex is a desktop knowledge archive for your AI conversations.

It lets you import, search, and revisit chat history from multiple LLM providers — locally, privately, and instantly.

Built with React 19 + Tauri 2 + SQLite (FTS5).

What it does

• Import conversation archives (Claude supported, others planned)
• Full-text search across all messages (SQLite FTS5)
• Prefix search support (salary*)
• Ranked results (BM25)
• Highlighted matches
• Per-conversation message search
• Source filtering
• Activity tracking by day
• Dark / Light / System theme
• Local-first (no cloud, no telemetry)

Core concept

Memex is not a chat app.

It is a recall engine — a searchable memory layer over your past AI conversations.

Architecture

• Frontend: React 19 + Vite 7
• Desktop runtime: Tauri 2
• Database: SQLite (memex.db)
• Search: FTS5 (BM25 ranking, snippet highlighting)
• Styling: Tailwind CSS v4
• Icons: lucide-react + custom LLM brand icons

All data is stored locally in memex.db.

Database schema (simplified)

conversations

• id
• source
• title
• created_at
• updated_at
• message_count

messages

• id
• conversation_id
• sender
• content
• created_at

messages_fts (FTS5 virtual table)

• content
• title
• conversation_id (UNINDEXED)
• message_id (UNINDEXED)

Getting started

From the repo root:

cd apps/desktop

# install dependencies
NPM_CONFIG_REGISTRY=https://registry.npmjs.org npm install

# start dev mode
npm run tauri dev

Search behavior

• Full-text search via SQLite FTS5
• Prefix search supported (term*)
• Ranked by BM25
• Results grouped by conversation
• Message-level highlighting
• Conversation viewer with in-place search

Preference is persisted in localStorage.

Current focus

Memex is evolving toward a thinking dashboard:

• Better recall
• Relevance scoring
• Activity visualization
• Related conversation surfacing

Status

Actively developed. Local-first. No telemetry. No accounts.