zai provider row still stores the retired Chat endpoint after the Responses move #4320

Client or integration

OpenCodex dashboard

Area

Provider adapter

Summary

After #4307 moved the zai registry row to the Responses protocol, an existing install still shows the retired Chat endpoint in the dashboard and in config.json:

• Adapter: openai-chat
• Base URL: https://api.z.ai/api/coding/paas/v4

The actual requests are already fine. routedProviderConfig() canonicalizes the row on every request because the zai registry entry owns a fixed destination, so the live wire is openai-responses against https://api.z.ai with /api/v1/responses. Only the stored row was never updated.

That mismatch has three visible costs: the provider page reports a protocol the proxy does not use, ocx doctor and any direct config reader inherit the same wrong value, and each boot logs a warning about a base URL the user never chose.

Expected: the saved row is rewritten once to the destination the router already applies, so stored state and live routing agree.

Reproduction

1. Start from a config where the zai provider was added before the Responses move, so it holds adapter: "openai-chat" and baseUrl: "https://api.z.ai/api/coding/paas/v4".
2. Update to a build that contains feat(zai): default to the Responses protocol and keep Chat as a per-model opt-in #4307.
3. ocx start --port 10100, open the dashboard at /#providers, select Zai and open 설정 / Settings.
4. The adapter field still reads openai-chat and the base URL still reads the /api/coding/paas/v4 endpoint.
5. Meanwhile the proxy logs configured baseUrl https://api.z.ai/… is ignored because this provider's endpoint is fixed at https://api.z.ai/, and requests leave on /api/v1/responses.

Version

2.52.0

Operating system

macOS 15.5

Provider and model

zai / glm-5.3

Logs or error output

⚠️  config.json provider "zai": configured baseUrl https://api.z.ai/… is ignored because this provider's endpoint is fixed at https://api.z.ai/. A URL saved for a different account or region is a common cause of 401s here — drop it, or use the provider whose endpoint matches.

Redacted configuration

{
  "providers": {
    "zai": {
      "adapter": "openai-chat",
      "baseUrl": "https://api.z.ai/api/coding/paas/v4",
      "authMode": "key",
      "defaultModel": "glm-5.3"
    }
  }
}

Checks

• I searched existing issues and documentation.
• I removed secrets, tokens, account details, request credentials, and personal data.