Vercel AI SDK + MemorySync

Overview

Install memorysync-ai-sdk and wrap any model once: recalled context is injected before generation and the user’s message is sent to fact extraction after it — once per user turn, never on tool-loop steps — streaming included. Only the durable facts in the message are stored; assistant replies are not. Standalone helpers cover hand-wired setups, five structured tools give agents explicit memory, and one build supports AI SDK v6 and v7 — the current major no other memory vendor supports.

Setup status and requirements

Supported version

memorysync-ai-sdk 1.1.0 (npm)

Last setup review

2026-10-01

Permissions

A MemorySync API key with read and write scopes.

Limits

Requires the ai package v6 or newer (Node 18+; ai v7 itself requires Node 22+). JavaScript/TypeScript only — Python agents use the LangChain, LangGraph or CrewAI integrations.

Capabilities

• wrapLanguageModel middleware: recall before generation, the user’s message sent for fact extraction after it
• Streaming-safe: sends once on finish; aborted streams send nothing
• AI SDK v6 AND v7 from one build
• Provider-agnostic — zero bundled LLM dependencies
• Recalled memories exposed on providerMetadata
• Five never-throw agent tools with a read-only mode
• Edge-safe: no Node-only imports

Quick Start

import { generateText, wrapLanguageModel } from "ai";
import { openai } from "@ai-sdk/openai";
import { createMemorySyncMiddleware } from "memorysync-ai-sdk";

const model = wrapLanguageModel({
  model: openai("your-model"),
  middleware: createMemorySyncMiddleware({ userId: "customer-7" }),
});

const { text } = await generateText({ model, prompt: "Book my usual trip." });

Use Cases

Explore More