getmnemo-vercel-ai

Mnemo adapter for the Vercel AI SDK. Drop-in tool() definitions that let any model search and write persistent memory, plus a small React hook for client-side memory views.

Install

npm install getmnemo-vercel-ai ai

getmnemo (the core SDK) is bundled as a dependency. Set GETMNEMO_API_KEY and GETMNEMO_WORKSPACE_ID in your environment, or pass them explicitly.

Quickstart (30 seconds)

import { streamText } from "ai";
import { openai } from "@ai-sdk/openai";
import { getmnemoTools } from "getmnemo-vercel-ai";

const result = await streamText({
  model: openai("gpt-4o"),
  tools: getmnemoTools, // memorySearch + memoryAdd
  maxSteps: 5,
  messages: [{ role: "user", content: "What did I tell you about my coffee?" }],
});

for await (const chunk of result.textStream) process.stdout.write(chunk);

getmnemoTools is lazy — it reads GETMNEMO_API_KEY and GETMNEMO_WORKSPACE_ID from process.env the first time a tool runs.

Per-user memory (route handler)

Use createMnemoTools when you need per-request scoping. The metadata you pass is merged into every memoryAdd call (and cannot be overwritten by the model), so it's the right place for a userId:

import { streamText } from "ai";
import { openai } from "@ai-sdk/openai";
import { createMnemoTools } from "getmnemo-vercel-ai";

export async function POST(req: Request) {
  const { messages, userId } = await req.json();
  const tools = createMnemoTools({ metadata: { userId } });

  const result = streamText({
    model: openai("gpt-4o"),
    tools,
    maxSteps: 5,
    messages,
  });

  return result.toDataStreamResponse();
}

createMnemoTools also accepts apiKey / workspaceId (instead of env vars), a pre-built client, and a defaultLimit for searches the model doesn't size:

const tools = createMnemoTools({
  apiKey: process.env.GETMNEMO_API_KEY,
  workspaceId: process.env.GETMNEMO_WORKSPACE_ID,
  defaultLimit: 8,
});

One-shot generation

The same tools work with generateText:

import { generateText } from "ai";
import { openai } from "@ai-sdk/openai";
import { getmnemoTools } from "getmnemo-vercel-ai";

const { text } = await generateText({
  model: openai("gpt-4o"),
  tools: getmnemoTools,
  maxSteps: 5,
  prompt: "Remember that I'm vegetarian, then suggest a dinner.",
});

Direct SDK access

Need to read or write memory outside a model loop? Use the core client directly:

import { Mnemo } from "getmnemo";

const memory = new Mnemo({
  apiKey: process.env.GETMNEMO_API_KEY!,
  workspaceId: process.env.GETMNEMO_WORKSPACE_ID!,
});

await memory.add({ content: "User prefers oat milk.", containerTag: "user:jane" });
const { hits } = await memory.search({
  q: "what milk does the user like?",
  containerTag: "user:jane",
});

React hook

For client-side memory views (sidebars, inspectors), use useMnemo.

⚠️ Security: a default Mnemo apiKey is full-access. A default key grants read, write, and delete over your entire workspace. NEVER ship a full-access key in a browser bundle or any NEXT_PUBLIC_ variable — anything with that prefix is inlined into client-side JavaScript and exposed to every visitor, handing them a delete-capable credential.

Scoped keys do exist: the dashboard mint dialog offers read/write/delete/billing scopes. For any client-exposed context, mint a scoped read-only key rather than exposing a full-access one. The safest path is still to proxy memory reads/writes through a server route (a Server Action or Route Handler) that holds the key server-side; useMnemo is for trusted internal/admin dashboards or a scoped read-only key.

The hook reads NEXT_PUBLIC_GETMNEMO_API_KEY / NEXT_PUBLIC_GETMNEMO_WORKSPACE_ID by default and returns SearchHit objects, keyed by memoryId. The example below is internal/admin only — not for public-facing apps:

"use client";
import { useMnemo } from "getmnemo-vercel-ai/react";
import type { SearchHit } from "getmnemo";

export function MemorySidebar() {
  const { results, search, loading } = useMnemo<SearchHit>({
    initialQuery: "preferences",
  });

  if (loading) return <p>Loading…</p>;

  return (
    <ul>
      {results.map((m) => (
        <li key={m.memoryId}>{m.content}</li>
      ))}
    </ul>
  );
}

useMnemo also exposes add(content, metadata?), remove(id), and error.

Docs

Full documentation at mnemohq.com.

License

MIT