Unable to disable WebSocket transport

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• #11698

Powered by Codex Action

Can you describe the behavior that you're seeing that's causing you to want to disable websocket transport? The feature flag was in place for development, but we typically remove feature flags once features are rolled out and stable.

Can you describe the behavior that you're seeing that's causing you to want to disable websocket transport? The feature flag was in place for development, but we typically remove feature flags once features are rolled out and stable.

Hi, Currently, I am running codex in an internal network environment (at a research institute's computing node), so I used a proxy. It worked normally before version v0.106.0, but after upgrading to that version, it started to encounter

{"type":"thread.started","thread_id":"xxx"}
{"type":"item.completed","item":{"id":"item_0","type":"error","message":"Under-development features enabled: realtime_conversation, responses_websockets_v2, voice_transcription. Under-development features are incomplete and may behave unpredictably. To suppress this warning, set `suppress_unstable_features_warning = true` in /home/xxx/.codex/config.toml."}}
{"type":"turn.started"}
{"type":"error","message":"Reconnecting... 2/5 (stream disconnected before completion: failed to send websocket request: Connection closed normally)"}
{"type":"error","message":"Reconnecting... 3/5 (stream disconnected before completion: failed to send websocket request: Connection closed normally)"}
{"type":"error","message":"Reconnecting... 4/5 (stream disconnected before completion: failed to send websocket request: Connection closed normally)"}
{"type":"error","message":"Reconnecting... 5/5 (stream disconnected before completion: failed to send websocket request: Connection closed normally)"}
{"type":"item.completed","item":{"id":"item_1","type":"error","message":"Falling back from WebSockets to HTTPS transport. stream disconnected before completion: failed to send websocket request: Connection closed normally"}}

This is similar to #13041

I'm not sure if the issue is with the proxy I'm using (I tried both socks and http), or a bug in this version itself, so I want to disable it first.

Hi, I just tried again without using a proxy, on a completely new virtual machine, and the same error still occurs.

Thread ID 019ca306-15e2-7dc0-8951-198b7e6d1616

@hanx-hep if you want, you can close this one and talk in https://github.com/openai/codex/issues/13041

@hanx-hep if you want, you can close this one and talk in #13041

@kali113 My issue is a bit different; I hope to provide an option to modify whether to use WS. Therefore, this might still be kept?

@hanx-hep can you do codex features list in cli

I have the same problem and also would like to disable websocket

I am also running into this. Afaik this was the timeline of events for me
I tried to turn on the responses_websockets_v2 feature via ~/.codex/config.toml. This did not work, both on the Codex app + CLI (both at the latest versions as of today) I saw

Reconnecting 1..2..3..4..5
stream disconnected before completion: failed to send websocket request: Connection closed normally

I then went back to the config file and commented out the responses_websockets_v2 feature.
However, this was never turned off, even today after a hard close + refersh of the apps, terminal and system it still seems to be using the websocket setting.

Can you describe the behavior that you're seeing that's causing you to want to disable websocket transport? The feature flag was in place for development, but we typically remove feature flags once features are rolled out and stable.

In 100% cases I met websocket error and fallback to https (with about 1min wasted). I would like to disable websocket for convenience.

@Weaverzhu go into #13041

Hi, I found a solution in #13041: https://github.com/openai/codex/issues/13041#issuecomment-3981110494.

Add the following configuration in ~/.codex/config.toml:

  model_provider = "openai_http"

  [model_providers.openai_http]
  name = "OpenAI HTTP only"
  wire_api = "responses"
  requires_openai_auth = true
  supports_websockets = false
  base_url = "https://chatgpt.com/backend-api/codex"

Hi, I found a solution in #13041: #13041 (comment). Add the following configuration in ~/.codex/config.toml: model_provider = "openai_http" [model_providers.openai_http] name = "OpenAI HTTP only" wire_api = "responses" requires_openai_auth = true supports_websockets = false base_url = "https://chatgpt.com/backend-api/codex"

Hello, indeed this method can be used, but this provider may not fully inherit all the features of the official codex provider.

However, it seems that the wss bug in version v0.106.0 has been fixed in subsequent updates, so at least I don't need this requirement anymore.