memo-vec

Embedded memory, indexed and searchable.

memo-vec is a local-first, vector-backed memory store for AI agents. It embeds text at write time and lets you search by meaning instead of keywords.

Install

npm install -g memo-vec

Quickstart

memo-vec config --init
memo-vec put "Customer prefers email over Slack"
memo-vec get "how does the customer like to be contacted?"

CLI

| Command | Description | |---------|-------------| | put <text> | Store a memory. Prints the memory ID. | | get <query> [-k n] | Search memories by meaning. | | forget <id> | Soft-delete a memory. | | list [--limit=20] | List recent memories. | | stats | Show store statistics. | | compact | Reclaim space and rebuild the index. |

Library

import { MemoVec } from 'memo-vec';

const mem = new MemoVec({ store: './memory' });
await mem.put('Customer prefers email over Slack');
const results = await mem.get('how does the customer like to be contacted?', { topK: 5 });
console.log(results);

Configuration

Set OPENAI_API_KEY or configure another OpenAI-compatible endpoint. See memo-vec config --init.