@anthropic/eternal-memory-hooks

Claude Code hooks for persistent AI memory across sessions via Eternal Memory.

Quick Start

# Install and setup
npx @anthropic/eternal-memory-hooks init

# Configure API credentials
export ETERNAL_MEMORY_API_KEY="your-key"
export ETERNAL_MEMORY_URL="https://your-server.com"  # Optional

Installation

Via npx (recommended)

npx @anthropic/eternal-memory-hooks init

This will:

• Copy hook files to .claude/hooks/
• Add hook configs to .claude/settings.json
• Copy the /remember command to .claude/commands/
• Copy skill documentation to .claude/skills/

Manual Installation

npm install @anthropic/eternal-memory-hooks
node node_modules/@anthropic/eternal-memory-hooks/bin/init.cjs

Configuration

Environment Variables

| Variable | Required | Default | Description | |----------|----------|---------|-------------| | ETERNAL_MEMORY_API_KEY | Yes | - | API key for authentication | | ETERNAL_MEMORY_URL | No | https://openclaw-be-production.up.railway.app | Backend URL |

Per-Project Configuration

Set different backends for different projects using .env files:

# ~/project-a/.env
ETERNAL_MEMORY_URL=https://memory-a.example.com
ETERNAL_MEMORY_API_KEY=key-for-project-a

# ~/project-b/.env
ETERNAL_MEMORY_URL=https://memory-b.example.com
ETERNAL_MEMORY_API_KEY=key-for-project-b

Hook Toggle

Disable hooks via .claude/.ck.json:

{
  "hooks": {
    "eternal-memory-search": false,
    "eternal-memory-commit-store": false
  }
}

How It Works

SessionStart Hook (eternal-memory-search.cjs)

On every Claude Code session start:

1. Builds search query from project context (package.json, git history, CLAUDE.md)
2. Searches Eternal Memory for relevant past learnings
3. Injects results into session context via stdout

PostToolUse Hook (eternal-memory-commit-store.cjs)

After every Bash command:

1. Detects git commit output patterns
2. Classifies commit type (bugfix, pattern, insight)
3. Stores meaningful commits to Eternal Memory (fire-and-forget)

/remember Command

Manual command to extract and store learnings from current session.

Self-Hosting

Requirements

• PostgreSQL 17+ with pgvector extension
• Node.js 18+ or Python 3.11+
• Embedding API (Cloudflare AI Workers or compatible)

Docker Compose

version: '3.8'
services:
  api:
    image: ghcr.io/anthropics/eternal-memory:latest
    environment:
      DATABASE_URL: postgres://user:pass@db:5432/memories
      API_KEY: your-secret-key
    ports:
      - "8080:8080"
  db:
    image: pgvector/pgvector:pg17
    environment:
      POSTGRES_USER: user
      POSTGRES_PASSWORD: pass
      POSTGRES_DB: memories

Programmatic Usage

const { searchMemories, storeMemory } = require('@anthropic/eternal-memory-hooks');

// Search memories
const results = await searchMemories('authentication patterns', {
  types: ['pattern', 'decision'],
  limit: 5,
  mode: 'hybrid'
});

// Store memory
await storeMemory({
  type: 'decision',
  title: '[Auth] - Chose JWT over sessions',
  content: 'Rationale: stateless scaling...',
  source: 'my-app'
});

Troubleshooting

Hooks not running

1. Check ETERNAL_MEMORY_API_KEY is set
2. Verify hooks in .claude/settings.json
3. Check hook file exists: .claude/hooks/eternal-memory-search.cjs

API connection errors

1. Verify ETERNAL_MEMORY_URL is accessible
2. Check API key is valid
3. Hooks fail silently (check debug logs)

Enable Debug Logging

export CK_DEBUG=1

ClaudeKit Compatibility

This package works with or without ClaudeKit:

• With ClaudeKit: Uses ClaudeKit's config system and utilities
• Without ClaudeKit: Falls back to standalone configuration

CLI Options

npx @anthropic/eternal-memory-hooks init [options]

Options:
  --force, -f       Overwrite existing files
  --skip-settings   Don't modify settings.json
  --help, -h        Show help message

License

MIT