Hi, thank you for publishing this project.
I am testing an OpenAI-compatible gateway setup for Codex using a custom provider in ~/.codex/config.toml, similar to:
model = "openai/gpt-5.x"
model_provider = "ai_api_gateway"
[model_providers.ai_api_gateway]
name = "AI API Gateway"
base_url = "https://example.com/v1"
wire_api = "responses"
env_key = "AIAPI_TOKEN"
The gateway requests work for new Codex conversations. However, when a user switches from their normal OpenAI/ChatGPT Codex provider to the gateway provider, their existing Codex local project conversations/history appear to disappear or become unavailable in Codex Desktop / resume flows. New conversations still work.
I found a related upstream Codex issue where switching model_provider hides existing local sessions from resume/history surfaces, even though the session files are still present: openai/codex#15494
Do you have a recommended ai-api/Codex setup pattern to avoid this problem? For example:
- avoid setting top-level
model_provider globally and only register [model_providers.*]
- launch Codex with a per-session CLI override such as
codex -m ... -c 'model_provider="ai_api_gateway"'
- use a separate
CODEX_HOME for short-lived classroom/workshop tokens
- any other profile/project setup that preserves long-term project history
The use case is important because a global provider switch is acceptable for short-term classroom usage, but not for long-term development if existing project conversations become hidden.
Thanks for any guidance or prior experience.
Hi, thank you for publishing this project.
I am testing an OpenAI-compatible gateway setup for Codex using a custom provider in
~/.codex/config.toml, similar to:The gateway requests work for new Codex conversations. However, when a user switches from their normal OpenAI/ChatGPT Codex provider to the gateway provider, their existing Codex local project conversations/history appear to disappear or become unavailable in Codex Desktop / resume flows. New conversations still work.
I found a related upstream Codex issue where switching
model_providerhides existing local sessions from resume/history surfaces, even though the session files are still present: openai/codex#15494Do you have a recommended ai-api/Codex setup pattern to avoid this problem? For example:
model_providerglobally and only register[model_providers.*]codex -m ... -c 'model_provider="ai_api_gateway"'CODEX_HOMEfor short-lived classroom/workshop tokensThe use case is important because a global provider switch is acceptable for short-term classroom usage, but not for long-term development if existing project conversations become hidden.
Thanks for any guidance or prior experience.