memorysync-ai-sdk

MemorySync memory for the Vercel AI SDK: wrap any model once and it remembers your users — recalled context injected before generation, the exchange persisted after it, streaming-safe. Supports AI SDK v6 and v7 from one build.

npm install memorysync-ai-sdk ai

Middleware — one wrap, long-term memory

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

const model = wrapLanguageModel({
  model: openai("gpt-4o-mini"),
  middleware: createMemorySyncMiddleware({
    userId: "customer-7",   // per-end-user scoping — required
    sessionId: "thread-42", // groups the stored transcript
  }),
});

await generateText({ model, prompt: "I'm vegetarian and I fly aisle." });

// Any later call — same user, any process, any provider:
const { text } = await generateText({
  model,
  prompt: "Book my trip: flight plus a dinner spot.",
});
// The model already saw: vegetarian, aisle seat — injected from memory.

Works identically with streamText: persistence happens once when the stream finishes; an aborted stream stores nothing. Memory failures never break generation — recall failures degrade to no context, persist failures are reported through onError, never thrown.

Options: recall: false (write-only), persist: false (read-only), k, template, onError, specificationVersion: "v3" for exact types on AI SDK 6.

Recalled memories ride the result: result.providerMetadata.memorysync.memories → [{ id, text, score }].

Helpers

import { getMemoryContext, searchMemories, saveTurn, createMemorySyncOnFinish } from "memorysync-ai-sdk";

const context = await getMemoryContext("what should I cook?", { userId: "customer-7" });
const hits = await searchMemories("dietary preferences", { userId: "customer-7" });
await saveTurn({ user: "I'm vegetarian", assistant: "Noted!" }, { userId: "customer-7" });

const result = streamText({
  model,
  prompt,
  onFinish: createMemorySyncOnFinish({ userId: "customer-7", user: prompt }),
});

All writes share the same idempotency seeds — retries and mixed styles cannot double-store a turn.

Agent tools

import { generateText, stepCountIs } from "ai";
import { memorySyncTools } from "memorysync-ai-sdk";

const { text } = await generateText({
  model: openai("gpt-4o-mini"),
  tools: { ...memorySyncTools({ userId: "customer-7" }) },
  stopWhen: stepCountIs(5),
  prompt: "Remember that I prefer aisle seats.",
});

Five tools — addMemory, searchMemory, listMemories, updateMemory, deleteMemory — that never throw (failures come back as readable strings) with idempotent writes. readOnly: true returns search + list only.

Version support

| Package | Requires | Runtime | |---|---|---| | memorysync-ai-sdk | ai >=6 <8 (peer) | Node 18+ (ai@7 itself requires Node 22+) |

Edge-safe: no Node-only imports — runs on Vercel Edge and Cloudflare Workers.

Documentation

Full guide: https://docs.memorysync.io/guides/vercel-ai-sdk

Set MEMORYSYNC_API_KEY in the environment (create a key at memorysync.io). MIT licensed.