@violoop/memoir

Tree-structured memory system with full-text search for AI assistants.

Features

• Tree Hierarchy: Organize memories with dot-separated keys (e.g., life.work.project)
• BM25 Search: Full-text search using SQLite FTS5
• Markdown Storage: Memories stored as .md files with frontmatter

Installation

npm install @violoop/memoir

Quick Start

const { Memoir } = require('@violoop/memoir')

const memory = new Memoir({
  memoryDir: './memory',
  dataDir: './memoir-data'
})

await memory.initialize()

// Add memory
await memory.set('life.work.project_a', 'Project A notes', { type: 'core' })

// Get memory tree for LLM prompt
const tree = await memory.getTreeForPrompt()
// ## 记忆目录
// ### life
//   - life.work: work [core]
//     - life.work.project_a: project_a [core]

// Search
const results = await memory.search('project')

await memory.close()

Key Format

life.work.project_a     → memory/life/work/project_a.md
skills.programming.js    → memory/skills/programming/js.md

Important: Don't use . in the last part:

✅ core.test-language  → core/test-language.md
❌ core.test.language  → path error

API

Method	Description
initialize()	Initialize
set(key, content, metadata)	Add/update
get(key)	Get by key
delete(key)	Delete
list()	List tree
getTreeForPrompt()	For LLM
search(query)	Search

Storage

memory/
├── core/
│   └── preferences.md
├── life/
│   └── work/
└── skills/

memoir-data/
└── index.sqlite

License

MIT