memrig

SQLite memory brain for AI coding assistants. Per-user + shared team memory, hybrid search, zero config.

What It Does

memrig gives AI coding assistants (Claude Code, Cursor, etc.) persistent memory that survives across sessions. Each team member gets their own private memory plus a shared team brain — all searchable via full-text + semantic search.

your-project/
├── .mcp.json              ← committed to git
├── .memory/
│   ├── shared.db          ← team knowledge (committed)
│   └── users/
│       ├── alice.db       ← Alice's memories (gitignored)
│       └── bob.db         ← Bob's memories (gitignored)

Install

npm install memrig

Quick Start

# In your project directory:
npx memrig init

# That's it. Restart Claude Code and memrig is connected.

memrig init does the following:

1. Creates .memory/ directory with shared.db and users/{you}.db
2. Adds .memory/users/ to .gitignore (personal memories stay private)
3. Creates .mcp.json with the memrig server config

Manual Setup

If you prefer to configure manually, add this to your .mcp.json:

{
  "mcpServers": {
    "memrig": {
      "command": "npx",
      "args": ["-y", "memrig"],
      "env": {
        "MEMORY_DIR": ".memory",
        "MEMRIG_USER": "${USER}"
      }
    }
  }
}

MCP Tools

Tool	What It Does
remember	Save a memory (personal by default, or shared)
recall	Search memories — hybrid FTS5 + vector search
forget	Delete a memory by ID
share	Promote a personal memory to shared team memory
import_memory	Copy a shared memory to your personal space
list_memories	Browse memories with filters

Memory Types

• decision — Architectural or product decisions
• preference — Personal or team preferences
• context — Background context about the project
• bug — Known bugs and workarounds
• pattern — Code patterns and conventions
• architecture — System architecture notes

How Search Works

recall uses hybrid retrieval:

1. FTS5 BM25 — keyword matching across content and tags
2. sqlite-vec — semantic similarity (MiniLM-L6-v2 embeddings, runs locally)
3. Reciprocal Rank Fusion — merges both ranked lists with score = Σ 1/(k + rank)
4. Ebbinghaus decay — older, less-accessed memories rank lower (decay rate varies by type)
5. Cross-DB merge — personal results weighted 1.1x higher than shared

Memories below 5% strength are auto-pruned during recall.

Configuration

Environment variables (set in .mcp.json):

Variable	Default	Description
MEMORY_DIR	.memory	Path to memory directory
MEMRIG_USER	$USER	Username for personal memory isolation

For Teams

1. One person runs npx memrig init and commits .mcp.json + .memory/shared.db
2. Everyone else pulls and runs npx memrig init (idempotent — won't overwrite existing config)
3. Shared memories sync via git. Personal memories stay local.
4. Use the share tool to promote personal memories to the team, import_memory to pull shared memories into your personal space.

Development

git clone https://github.com/RiggdAI/memrig.git
cd memrig
npm install
npm test        # run tests (vitest)
npm run build   # compile with tsup

License

MIT