@syncmem/sdk

The universal memory layer for AI agents — TypeScript/JavaScript SDK

Give your AI agents persistent, cross-platform memory in under 2 minutes. SyncMem stores facts extracted from conversations, retrieves them in sub-4ms, and probabilistically links the same user across multiple accounts and devices — without changing your LLM or your stack.

Installation

npm install @syncmem/sdk
# or
yarn add @syncmem/sdk
# or
pnpm add @syncmem/sdk

Quickstart

import { SyncMem } from '@syncmem/sdk';

const client = new SyncMem({
  llmProvider: 'openai',
  llmApiKey: process.env.OPENAI_API_KEY!, // your existing key — no new credentials
  baseUrl: 'https://api.syncmem.com',
});

// First message — facts are extracted and stored automatically
await client.chat({
  accountId: '[email protected]',
  message: "I'm a vegetarian living in Tokyo. I'm a backend engineer.",
});

// Later — memory is injected automatically
const response = await client.chat({
  accountId: '[email protected]',
  message: 'What restaurants would you recommend near my office?',
});

console.log(response.reply);
// → "For vegetarian options in Tokyo, I'd suggest Ain Soph Ripple in Shinjuku..."
console.log(`Injected ${response.factsInjected} memories`);

Authentication

Keyless (recommended)

Your existing LLM API key is your SyncMem identity — zero sign-up, zero new credentials.

const client = new SyncMem({
  llmProvider: 'openai',   // openai | gemini | claude | groq | ollama | mistral
  llmApiKey: 'sk-...',
});

SyncMem API Key

For multiple isolated namespaces or production deployments, create a sm_live_... key from the dashboard.

const client = new SyncMem({
  apiKey: 'sm_live_...',
});

Core API

client.chat(options)

Send a message and get a memory-enriched reply. Facts are extracted and stored asynchronously.

const response = await client.chat({
  accountId: '[email protected]',  // any stable user identifier
  message: 'What did we discuss about the Q4 roadmap?',
  deviceId: 'device-uuid-123',     // optional: improves cross-device linking
});

// response.reply          — the LLM's response, enriched with memory
// response.factsInjected  — number of memories injected into context
// response.accountId      — normalized account identifier

client.getFacts(accountId)

Retrieve all stored facts for a user.

const { facts } = await client.getFacts('[email protected]');

facts.forEach(f => {
  console.log(`[${f.category}] ${f.content} (importance: ${f.importance})`);
});

client.addFact(accountId, content, options?)

Manually store a fact for a user.

await client.addFact('[email protected]', 'Prefers dark mode in all apps', {
  category: 'preferences',
  importance: 0.8,
});

client.deleteFact(factId)

Delete a specific fact by ID.

await client.deleteFact('fact-uuid-123');

client.searchFacts(accountId, query)

Full-text search across a user's facts.

const results = await client.searchFacts('[email protected]', 'dietary preferences');

client.getIdentityLinks(accountId)

Get probabilistic identity links — other accounts that likely belong to the same person.

const { links } = await client.getIdentityLinks('[email protected]');

links.forEach(link => {
  console.log(`Linked to ${link.accountId} with ${link.confidence * 100}% confidence`);
});

Supported LLM Providers

| Provider | llmProvider value | |----------|-------------------| | OpenAI | openai | | Google Gemini | gemini | | Anthropic Claude | claude | | Groq | groq | | Mistral | mistral | | Together AI | together | | DeepSeek | deepseek | | Ollama (local) | ollama | | Custom / proxy | custom |

MCP Server

SyncMem ships a built-in MCP server for Claude Desktop, Cursor, Windsurf, and any other MCP-compatible agent:

{
  "mcpServers": {
    "syncmem": {
      "command": "npx",
      "args": ["-y", "@syncmem/mcp"],
      "env": {
        "SYNCMEM_API_KEY": "sm_live_...",
        "LLM_PROVIDER": "openai",
        "LLM_API_KEY": "sk-..."
      }
    }
  }
}

Self-hosting

SyncMem is fully open source. Run your own instance:

git clone https://github.com/syncmem/syncmem
cd syncmem/production
cp .env.example .env   # fill in secrets
bash scripts/deploy.sh

Links

• Documentation
• Dashboard
• GitHub
• syncmem.com

License

MIT © SyncMem