Skip to main content

@AiAbortSignal

@AiAbortSignal() is a parameter decorator that injects an AbortSignal which fires when the client disconnects mid-stream. Forward it into your AI SDK call so a disconnect cancels the upstream model request — and the provider stops billing — instead of streaming tokens into a dead socket.

import { AiAbortSignal, AiStream } from '@nest-native/ai-sdk';
import { Body, Controller, Post } from '@nestjs/common';
import { streamText } from 'ai';

@Controller('chat')
export class ChatController {
  @Post()
  @AiStream()
  chat(@Body() body: ChatDto, @AiAbortSignal() signal: AbortSignal) {
    // When the client aborts the fetch, `signal` aborts and the AI SDK tears
    // down the model request.
    return streamText({ model, prompt: body.prompt, abortSignal: signal });
  }
}

How The Signal Is Derived​

The signal is derived from the active response's lifecycle, not the request:

  • The AI SDK writes the stream to the underlying Node ServerResponse.
  • When the client goes away mid-stream, that response emits close before it has finished.
  • @AiAbortSignal() listens for that early close and aborts the controller.

Using the response (not the request) is deliberate. The request stream can close as soon as its body is read — long before the client disconnects — so it is not a reliable disconnect signal during streaming. The response only closes early when the socket is actually torn down.

If the response has already ended or been destroyed when the signal is resolved, the returned signal is already aborted.

Adapter Behavior​

The signal works identically on Express and Fastify. Express exposes the Node ServerResponse directly; Fastify exposes it via reply.raw. The package normalizes both, so the same handler behaves the same on either adapter.

Memoization​

The signal is memoized on the response. Declaring @AiAbortSignal() more than once on the same handler — or resolving it repeatedly — returns the same signal backed by a single AbortController. There is no listener leak: the internal finish/close listeners detach themselves once either fires.

Cost Control​

Forwarding the abort signal is the primary cost-control lever for streaming LLM endpoints. Without it, a client that closes the tab still leaves the model generating tokens you pay for. Combine it with rate limiting and a maxTokens/ maxOutputTokens guard in your AI SDK call. See Production Patterns and Security.

  • sample/02-abort-signal — a focused sample whose smoke test disconnects mid-stream and asserts the AI SDK call is cancelled on both adapters.
  • @AiStream — the streaming decorator the signal pairs with.