@chaoszh/mnemo-core

Core service logic for Mnemo — persistent memory for LLM agents. Provides tool handlers, embeddings, and database client.

Install

npm install @chaoszh/mnemo-core

Usage

import { createHandlers } from '@chaoszh/mnemo-core';

const handlers = createHandlers({
  databaseUrl: process.env.DATABASE_URL,
  agentId: 'my-agent',
  embedding: {
    type: 'gemini',
    apiKey: process.env.GEMINI_API_KEY,
  },
});

// Save a memory
await handlers.save_memory({
  content: 'User prefers dark mode',
  memory_type: 'semantic',
});

// Search memories
await handlers.search_memories({
  query: 'user preferences',
  limit: 5,
});

API

createHandlers(config)

Returns a map of pre-wired handlers bound to a single database + embedder instance:

| Handler | Description | |---|---| | save_memory | Persist a memory | | search_memories | Find memories by semantic similarity | | delete_memory | Delete a memory by ID | | update_memory | Correct or amend an existing memory |

Config

| Field | Required | Description | |---|---|---| | databaseUrl | ✅ | Postgres connection string | | agentId | ✅ | Unique identifier for this agent | | embedding.type | ✅ | Embedding provider — "gemini" | | embedding.apiKey | ✅ | API key for the embedding provider | | defaultSearchLimit | — | Max results from search_memories (default 5) |

Database

Requires PostgreSQL with pgvector. Apply migrations from db/ before use:

for f in db/*.sql; do psql "$DATABASE_URL" -f "$f"; done