@getplumb/core

Cross-session AI memory — storage abstraction, types, and local SQLite driver

The core library for Plumb — a local-first, MCP-native memory layer for AI agents.

What it does

• Two-layer memory: raw conversation log (Layer 1) + extracted fact graph (Layer 2)
• Hybrid search: BM25 + semantic vectors + RRF fusion + cross-encoder reranking
• Local storage: SQLite + sqlite-vec, zero external dependencies
• Pluggable LLM: bring your own OpenAI or Anthropic client for fact extraction

Install

npm install @getplumb/core

Quick start

import { LocalStore } from '@getplumb/core';

const store = new LocalStore({ dbPath: '~/.plumb/memory.db' });
await store.init();

// Ingest a conversation turn
await store.ingest({
  sessionId: 'my-session',
  userId: 'user-123',
  userMessage: 'I prefer TypeScript over Python',
  assistantMessage: 'Noted! TypeScript it is.'
});

// Retrieve relevant memory
const results = await store.searchRawLog('TypeScript preferences', 5);

Links

• Docs
• GitHub
• getplumb.dev

License

MIT