OpenAI Memory API

The OpenAI memory gap

OpenAI added memory to ChatGPT for consumer users. The standard chat completions API has no equivalent. Each API call is stateless unless you send full conversation history.

Sending entire chat logs on every request burns tokens, increases latency, and still misses facts from older sessions that fell out of context.

Developers building GPT-4o, GPT-4, or GPT-3.5 apps need an external memory layer. Databaset fills that role with a simple store and recall API.

How to add memory to OpenAI apps

• After each user message, call memory.recall with the userId and query
• Inject recalled memories into the system message before calling OpenAI
• After the assistant responds, store durable new facts with memory.store
• Use metadata tags like source: openai-gpt4 for filtering and debugging

import OpenAI from "openai";
import { Databaset } from "@databaset/sdk";

const openai = new OpenAI();
const memory = new Databaset({ apiKey: process.env.DATABASET_API_KEY });

async function chat(userId: string, message: string) {
  const { memories } = await memory.recall({ userId, query: message, limit: 5 });

  const system = `You are a helpful assistant.
Known facts about this user:
${memories.map(m => "- " + m.text).join("\n")}`;

  const completion = await openai.chat.completions.create({
    model: "gpt-4o",
    messages: [
      { role: "system", content: system },
      { role: "user", content: message },
    ],
  });

  const reply = completion.choices[0].message.content ?? "";
  // Store new facts extracted from the exchange
  await memory.store({ userId, text: message, metadata: { source: "openai" } });
  return reply;
}

Databaset vs OpenAI Assistants API threads

The Assistants API stores thread history, but it is tied to OpenAI's thread model, not portable semantic memory. Threads grow unbounded and lack per-fact recall.

Databaset gives you user-scoped semantic memory that works with any OpenAI model, any framework (Vercel AI SDK, LangChain, raw fetch), and any other LLM provider simultaneously.

Works with Vercel AI SDK and LangChain

Call memory.recall in your route handler or LangChain tool chain before streaming a response. Inject memories into the system prompt or as a tool result.

See docs.databaset.com for Node.js and Python SDK quickstarts with OpenAI examples.

Cost and limits

Databaset free tier: 3,000 API calls per month. Starter plan: 100K calls at $29/mo. You still pay OpenAI for model tokens separately.

Memory reduces OpenAI costs by sending fewer tokens per request compared to full history replay.