@elizaos/plugin-inmemorydb

A pure in-memory, ephemeral database adapter for elizaOS. All data is completely lost on process restart or when close() is called.

Features

• Zero Configuration: No database setup required - just works out of the box
• Zero Persistence: Data is never written to disk or localStorage
• Completely Ephemeral: All data is lost when the process ends
• Cross-Platform: Works identically in Node.js and browsers
• Vector Search: Built-in HNSW index for semantic similarity search (also ephemeral)
• Same Interface: Implements the standard IDatabaseAdapter interface

When to Use

This plugin is ideal for:

• ✅ Testing and CI/CD: Perfect for tests that need a fresh database each run
• ✅ Stateless deployments: When you want agents to start fresh each time
• ✅ Development: Quick prototyping without persistence overhead
• ✅ Privacy-focused applications: When data should never be stored
• ✅ Demos and examples: Clean slate for each demonstration

Not recommended for:

• ❌ Production systems that need data persistence
• ❌ Long-running agents that need to remember past interactions
• ❌ Multi-process deployments (data is not shared between processes)

Installation

bun add @elizaos/plugin-inmemorydb

Quick Start

Node.js

import { plugin } from "@elizaos/plugin-inmemorydb";

// Add to your agent configuration
const agent = {
  plugins: [plugin],
  // ...
};

Browser

import { plugin } from "@elizaos/plugin-inmemorydb";

// Add to your agent configuration
const agent = {
  plugins: [plugin],
  // ...
};

API

Creating an Adapter Manually

import {
  InMemoryDatabaseAdapter,
  MemoryStorage,
} from "@elizaos/plugin-inmemorydb";

const storage = new MemoryStorage();
const adapter = new InMemoryDatabaseAdapter(storage, agentId);
await adapter.init();

// Use the adapter...

// When done, close to clear all data
await adapter.close();

Clearing Data

// Clear all data (adapter still usable)
await storage.clear();

// Or close the adapter entirely (also clears data)
await adapter.close();

Comparison with Other Database Plugins

| Feature | plugin-inmemorydb | plugin-localdb | plugin-sql | | --------------- | ------------------ | ------------------ | ------------------- | | Persistence | None (ephemeral) | JSON files | PostgreSQL | | Setup | Zero configuration | Zero configuration | Requires PostgreSQL | | Data on restart | Lost | Preserved | Preserved | | Best for | Testing/dev | Local dev | Production |

How It Works

The plugin uses simple JavaScript Map data structures to store all data:

Memory
├── Collections (Map<string, Map<string, unknown>>)
│   ├── agents: Map<id, agent>
│   ├── memories: Map<id, memory>
│   ├── rooms: Map<id, room>
│   └── ... (other collections)
└── HNSW Vector Index (in-memory)

When the process ends or close() is called, all Maps are cleared and data is gone forever.

Vector Search

The plugin includes an ephemeral HNSW (Hierarchical Navigable Small World) implementation for vector similarity search:

• Semantic memory search
• Similar content retrieval
• Embedding-based queries

The vector index is also purely in-memory and is cleared with all other data.

License

MIT