@ai-sdk-x/memo

Memoize AI SDK tool execution results. If your tool is a pure function and is called frequently in a short period of time, Memo helps you cache the results to improve performance.

Installation

npm add @ai-sdk-x/memo unstorage

Usage

import { createMemo } from "@ai-sdk-x/memo";
import { createStorage } from "unstorage";
import lruCacheDriver from "unstorage/drivers/lru-cache";

const memo = createMemo({
  storage: createStorage({ driver: lruCacheDriver({}) }),
  ttl: 60_000, // 60 seconds
});

const memoizedTool = memo(myTool, "myTool");

Example

See the In-memory example on GitHub.

See the Null-memory example on GitHub.

Manual Cache Management

The memoized tool exposes get, update, and delete helpers:

// Inspect a cache entry
const entry = await memoizedSearch.get('memo:search:{"query":"TypeScript"}');

// Overwrite a cache entry
await memoizedSearch.update('memo:search:{"query":"TypeScript"}', {
  query: "TypeScript",
  results: ["Updated result"],
});

// Remove a cache entry
await memoizedSearch.delete('memo:search:{"query":"TypeScript"}');

Storage Backends

Any unstorage driver works — Redis, Cloudflare KV, Deno KV, Vercel KV, filesystem, etc.