Vercel AI SDK compression integration

How compression fits into AI SDK routes

Vercel AI SDK routes receive a messages array and pass it to a streaming model call. The ideal place to compress is right in the route handler, before the messages reach the SDK's streaming function.

The compressor scores each message against the latest user query and removes low-value context. System messages and instructions are preserved intact.

TypeScript integration example

import { StreamingTextResponse } from "ai";
import OpenAI from "openai";
import { SuperCompress } from "supercompress";

export async function POST(req: Request) {
  const { messages } = await req.json();
  const compressor = new SuperCompress();

  // Compress conversation history against the latest query
  const latestQuery = messages[messages.length - 1].content;
  const history = messages.slice(0, -1)
    .map(m => `${m.role}: ${m.content}`)
    .join("
");

  const compressed = await compressor.compress(history, latestQuery);

  // Rebuild messages with compressed history
  const compressedMessages = [
    { role: "system", content: "You are a helpful assistant." },
    { role: "user", content: compressed.text },
    messages[messages.length - 1]
  ];

  const response = await openai.chat.completions.create({
    model: "gpt-4o",
    messages: compressedMessages,
    stream: true,
  });

  return new StreamingTextResponse(response);
}

Frequently asked questions

Does compression work with AI SDK streaming?

Yes. Compress before calling the streaming function. The streamed response is unaffected.

What about tool calls in AI SDK?

The compressor preserves tool call content. Only low-value conversation history is compressed.