@gifamossThat Reconnecting 5/5 status in Codex usually isn't a model problem at all. It often comes down to how Codex handles WebSockets or the local proxy setup. If you are using native Codex, try disabling Websockets by setting approvals_reviewer to user and model_provider to openai_http. You need to configure the [model_providers.openai_http] section with wire_api as responses and supports_websockets as false. Make sure network_access is true inside [sandbox_workspace_write]. Users running Codex via a custom API or proxy should check their system proxy settings instead. You can ask Codex to write proxy information to ~/.codex/.env to determine the current proxy port used on your computer. Once you have that info, set HTTP_PROXY, HTTPS_PROXY, ALL_PROXY, and NO_PROXY variables correctly. The example configuration uses http://127.0.0.1:15236 for these values but you must replace 15236 with the real proxy port found in your environment file. After changing configurations, fully quit and reopen Codex for changes to take effect. This approach helps prevent broken long-running sessions where Agents or Sub-Agents die at Reconnecting 1/5 → 5/5. #OpenAI
원본 게시물 보기
