TypeError: terminated

[bakikucukcakiroglu]

We use openai-agents-js (default Responses API) to run our Multi-Agent Systems (MAS). During MAS runs, we intermittently hit the following error:

{
  "error": "TypeError: terminated"
}

We’ve seen this issue for a long time, but the frequency has increased significantly over the last 2 days. It happens in both streamed and non-streamed MAS runs.

We use the gpt-4.1 and gpt-5 model families across our MASes.

Full error (from our logs):

"message": "terminated",
2026-01-14 10:07:06.941 CET
"stack": "TypeError: terminated
    at Fetch.onAborted (node:internal/deps/undici/undici:11322:53)
    at Fetch.emit (node:events:524:28)
    at Fetch.terminate (node:internal/deps/undici/undici:10480:14)
    at Object.onError (node:internal/deps/undici/undici:11443:38)
    at Request.onError (node:internal/deps/undici/undici:2210:31)
    at Object.errorRequest (node:internal/deps/undici/undici:1707:17)
    at TLSSocket.<anonymous> (node:internal/deps/undici/undici:6433:16)
    at TLSSocket.emit (node:events:536:35)
    at node:net:343:12
    at TCP.done (node:_tls_wrap:648:7)"

Because the SDK uses the Responses API by default, this may be related to the following thread:

https://community.openai.com/t/200-ok-from-v1-responses-but-body-times-out-und-err-body-timeout/1371083

This issue may also be related:

#744

And this is the corresponding “twin” issue in the OpenAI Developer Community:

https://community.openai.com/t/agents-sdk-typeerror-terminated/1371661

If there are any known incidents, rate limits, networking changes, or recommended client settings for these model families that could explain the recent increase, we’d appreciate any guidance. We’d also like to understand whether this error is caused by our system (e.g., timeouts/aborts/network/proxy behavior) or if it’s originating from the OpenAI backend.

[seratch]

Hi, I will also check this issue on my end to see what we can do on the SDK side, but could you report it to our customer support team as well? Since this issue tracker is for the client-side TypeScript SDK, communicating directly with support agents with your app and account details will be smoother.

[bakikucukcakiroglu]

Hi @seratch, thanks for the quick response!

As you mentioned, this likely isn’t specific to the Agents SDK and may instead be a backend issue (possibly related to the Responses API).

Do you think this thread could be the root cause? If so, it looks like it has already been reported there.

200-ok-from-v1-responses-but-body-times-out-und-err-body-timeout

[Askir]

FWIW I created #744 at the time and solved our problems with it by retrying the full agent loop on failure.
If your agent is not a background worker I assume that's not a very good solution (and also potentially expensive) for many so it would be nice if you could somehow just retry the responses apis request itself but that part is hidden quite deep in the sdk currently.

[bakikucukcakiroglu]

FWIW I created #744 at the time and solved our problems with it by retrying the full agent loop on failure.
If your agent is not a background worker I assume that's not a very good solution (and also potentially expensive) for many so it would be nice if you could somehow just retry the responses apis request itself but that part is hidden quite deep in the sdk currently.

A full MAS or agent retry won’t work for us because many of the tools are mutating. However, a mechanism to retry only the failed Responses API call would solve our problem as well.