Cockpit Tools custom API setup guide
Cockpit Tools manages Codex profiles. Follow the full sequence—add provider, bind it to an instance, then enable and start—to make Codex actually use the qionggeme API.
Quick configuration
- Menu path
Codex page → Model Providers tab → Add Custom Provider- Base URL
https://qionggeme.com/v1- Do not enter the content-site URL or /docs.
- API Key
The API key created in the qionggeme console- Use a placeholder in documentation; never expose the real key.
- wireApi
responses- Do not choose chat_completions.
- Model
gpt-5.3-codex-spark- Actual availability depends on the model list visible to the current token.
- Required activation
Bind provider → Enable and Start- Saving is not the same as switching.
Before you begin:
- Install and open Cockpit Tools, and confirm that the Codex management page is available.
- Create an API key in the qionggeme console.
- Check the models visible to the current token in the console. The pricing page shows highlighted models.
Setup steps
Open the Codex Model Providers tab
Open the Codex page in Cockpit Tools, switch to the Model Providers tab, and click Add Custom Provider. Do not put this guide’s Base URL into Cockpit’s local service address field; you are configuring an external model provider.
Enter the name, Base URL, and API key
- Name: for example
qionggeme-codex - Base URL:
https://qionggeme.com/v1 - API Key: paste your own qionggeme API key without quotes or extra spaces
The Base URL is the API endpoint, not the content-site URL. Never append
/docs.- Name: for example
Set wireApi to responses and enter the model
Choose
responsesfor wireApi. You may usegpt-5.3-codex-sparkas the example model. If it is absent from the current token’s model list, replace it with a listed Codex-compatible model; the pricing page is only a highlighted-model reference. Do not selectchat_completionsbecause this setup uses Codex’s native Responses protocol.Save the provider configuration
Save and confirm that the provider card appears in the Model Providers list. Saving only persists the provider in Cockpit; it does not switch any Codex instance to that provider.
Return to the instance and bind the provider
Go back to the Codex instance management area, choose the target instance, and bind the saved custom provider to it. Without this binding, the Codex profile will not use the provider even though it was saved successfully.
Explicitly choose “Enable and Start”
After binding, explicitly run Enable and Start and watch the instance status. This is the action that points the selected Codex profile to the provider. Again: saving is not switching.
Verify that Codex uses the new provider
After the instance shows a running state, launch or restart Codex from that instance and send a simple test request. If an older Codex terminal or profile was running before the switch, exit it first so it does not continue using stale configuration.
Configuration example
# Cockpit Tools → Codex → Model Providers → Add Custom Provider Name: qionggeme-codex Base URL: https://qionggeme.com/v1 API Key: your qionggeme API key wireApi: responses Model: gpt-5.3-codex-spark (or a Codex model visible to the current token) # Required after saving 1. Return to the instance 2. Bind this provider 3. Choose “Enable and Start” # Important Saving the provider does not switch the instance. Do not simultaneously edit Cockpit-managed auth.json or config.toml files.
Troubleshooting
The provider is saved, but Codex still uses the old route
Saving only persists the provider. Return to the instance, confirm that the provider is bound, explicitly choose Enable and Start, and restart the running Codex process. Check the instance state instead of only checking whether a provider card exists.
I selected chat_completions and Responses requests fail
Edit the provider, set wireApi to responses, and keep Base URL at https://qionggeme.com/v1. This guide does not depend on Chat Completions conversion or local route mapping.
The provider exists but is not active for the instance
Verify that the provider is actually bound to the current instance. Binding is a separate action, and switching to another instance requires checking its binding again before choosing “Enable and Start”.
The instance does not start, or its port/status is abnormal
Check the Cockpit instance status and port information. Confirm that the service is running and the port is not already used by another process. Identify the owner before taking action; do not terminate processes blindly. After changing the port, restart the instance and test again.
Cockpit still reads an old profile, or manual edits caused inconsistent state
Stop the managed instance, confirm its binding and enabled state, and let Cockpit run “Enable and Start” again. Cockpit manages the profile’s auth.json and config.toml; use Cockpit’s mechanism to restore or switch profiles instead of editing those files while it controls them.
Ready to get started?
Create a qionggeme account, generate an API key, and access every currently available model.
Create an accountFrequently asked questions
What Base URL should I use in Cockpit Tools?
Use https://qionggeme.com/v1, the qionggeme Codex API endpoint. Do not enter the content-site URL or any address containing /docs.
Why must wireApi be responses?
Codex uses the native Responses protocol. chat_completions is a different protocol and can cause request-format or routing mismatches.
Does saving the provider mean the switch is complete?
No. Saving only stores the provider. You must bind it to the target instance, explicitly choose “Enable and Start”, and then verify the running state and a real request.
Which Codex files does Cockpit Tools manage?
Cockpit manages auth.json and config.toml for Codex profiles and points the selected profile to the chosen provider when enabled. Use Cockpit’s own disable/restore flow and avoid editing managed files at the same time.
Is gpt-5.3-codex-spark guaranteed to be available?
No. It is the example in this guide. Actual availability follows the model list visible to the current token in the console. If the ID is absent, use a listed Codex-compatible model; the pricing page is only a highlighted-model reference.