## What happened?

I found this while using the OpenAI Responses API background mode through the LiteLLM proxy.

When resuming a background response via:

`GET /v1/responses/{response_id}?stream=true&starting_after=<sequence_number>`

LiteLLM did not correctly handle the retrieve streaming path.

I expected the proxy to support OpenAI-style cursor-based stream resume on the retrieve endpoint and return a valid SSE stream for background responses.

## Steps to Reproduce

1. Send a Responses API request through the LiteLLM proxy using OpenAI background mode. 2. Wait until the response can be retrieved by `response_id`. 3. Call `GET /v1/responses/{response_id}?stream=true&starting_after=<sequence_number>`. 4. Observe that the retrieve streaming path is not handled correctly.

## Relevant log output

No additional production logs attached.

I was able to reproduce this consistently through the proxy retrieve path and then narrowed it down to the retrieve streaming flow.

## What part of LiteLLM is this about?

Proxy

## What LiteLLM version are you on?

v1.83.14

## Additional context

This was discovered while trying to use background mode together with streaming resume, which is a va…