[Question] codex

[shijunti19]

Your Question

9:50:23
INFO
No active account found
09:41:47
INFO
>>> /responses | model=gpt-5.4 | 46658B | encoding=none | tools=13
model=gpt-5.4
tools=13
09:41:43
INFO
>>> /responses | model=gpt-5.4 | 46658B | encoding=none | tools=13
model=gpt-5.4
tools=13
09:41:41
INFO
>>> /responses | model=gpt-5.4 | 46658B | encoding=none | tools=13
model=gpt-5.4
tools=13
09:41:40
INFO
>>> /responses | model=gpt-5.4 | 46658B | encoding=none | tools=13
model=gpt-5.4
tools=13
09:41:39
INFO
>>> /responses | model=gpt-5.4 | 46658B | encoding=none | tools=13
model=gpt-5.4
tools=13
09:41:39
INFO
>>> /responses | model=gpt-5.4 | 46658B | encoding=none | tools=13
model=gpt-5.4
tools=13
09:39:15
INFO
No active account found
09:39:03
INFO
No active account found
09:30:45
INFO
Account strategy: sequential
09:30:29
INFO
Discovery complete: moonshot:1, zhipu:7, minimax:7
09:30:29
WARN
Local runtime discovery failed: fetch failed
09:30:29
INFO
API key minimax: found 7 models
09:30:29
INFO
API key zhipu: found 7 models
09:30:29
INFO
API key moonshot: found 1 models

What have you tried?

No response

CliGate Version

No response

Topic

None

Operating System

None

[codeking-ai]

@shijunti19 感谢反馈，我看了一下你贴的日志。

这里看起来不是单纯的模型发现问题：moonshot / zhipu / minimax 的模型已经被发现了，但 Codex 这次请求走的是 /responses，请求模型是 gpt-5.4，同时日志里出现了 No active account found。这说明当前 Codex 路由可能没有命中可用的 ChatGPT account，或者 API Key 路由没有正确接管这次 Codex /responses 请求。

我会优先检查这里的路由逻辑，尤其是 Codex /responses 在没有 active account 时，是否应该继续走已配置的 API Key provider。

方便的话请再补充一下：

1. CliGate 版本号
2. 你当前的 Routing Priority 是 account-first 还是 apikey-first
3. 是否开启了 Strict Codex Compatibility
4. Codex CLI 里配置的 openai_base_url

另外你已经连续反馈了几个真实使用问题，这些信息对我很有价值。我也希望能更直接了解你的使用场景，方便我把 Codex / 模型映射 / API Key 路由这块改得更稳定。

可以直接邮件联系我：steven.geng2026@gmail.com
也可以加 README 里的微信二维码，备注 CliGate 就行。

感谢！

[shijunti19]

@codeking-ai
v1.0.9 It still shows as 1.0.9 after I updated.
apikey-first
Okay, sorry!

model_provider = "custom"
model = "gpt-5.4"
model_reasoning_effort = "medium"
disable_response_storage = true

chatgpt_base_url = "http://localhost:8081/backend-api/"
openai_base_url = "http://localhost:8081"
allow_login_shell = false
sandbox_mode = "workspace-write"
[model_providers.custom]
name = "bigmodel"
base_url = "http://localhost:8081"
wire_api = "responses"
requires_openai_auth = true

[mcp_servers]

[mcp_servers.db-mcp]
type = "sse"
url = "http://192.168.1.45:5001/sse"

[mcp_servers.bb-browser]
type = "http"
url = "http://192.168.1.18:19824/mcp"

[projects.'z:\program files\rime\weasel-0.17.4']
trust_level = "trusted"

[tui.model_availability_nux]
"gpt-5.5" = 2

[notice.model_migrations]
"gpt-5.3-codex" = "gpt-5.4"

[windows]
sandbox = "unelevated"

[features]
powershell_utf8 = false

Model mapping is often lost.

I always thought it was the latest version; I just realized you didn't send me an update notification.

[shijunti19]

I'm using yours because I want this feature: MCP supports switching AI models and service providers. I'd like to add this feature!

[codeking-ai]

@shijunti19 Thanks, this is very helpful.

The first important point is that you are still on v1.0.9. The latest release is now v1.2.2, and several Codex / routing / model-mapping fixes landed after v1.0.9. Please upgrade first and then retry the same Codex config.

If you installed with npm, try:

npm install -g cligate@latest
cligate --version

If you are using the desktop app, please download the latest release from the GitHub Releases page. The fact that you did not receive an update notification is also useful feedback; I will check the update notification path separately.

About your config: since you are using apikey-first, Codex /responses should be able to route through the configured API key provider. If it still falls back to No active account found after upgrading, that is a bug and I will treat it as a priority. Your note that "Model mapping is often lost" is also important; I will check whether mappings are not being persisted, overwritten after model discovery, or reset when provider settings refresh.

For the MCP idea, I want to make sure I understand the feature you want:

Do you mean an MCP tool/API that can switch the active AI model and service provider at runtime, for example switching between OpenAI / Zhipu / MiniMax / Moonshot without manually changing the dashboard settings? If yes, I think this is a very good direction.

A few details would help me design it correctly:

1. Should the MCP switch be global, or only for the current conversation/session?
2. Should it switch by provider name, model name, or both?
3. Should the selection persist after restart, or only last for the current run?
4. Do you expect this to work mainly for Codex, or also for Claude Code / Gemini / Web Chat?

Thanks again for the detailed config and screenshots. Your usage is exactly the kind of real workflow I want CliGate to support better.

[shijunti19]

@codeking-ai
@zerone0x @Ayush-Kotlin-Dev
Oh, and this is very important: you can't force me to enable various client configurations! I use cc-switch because it allows me to configure MCP skills, etc., and you can't replace it! Your job is specifically to handle the API interface.

Your updates are really weird!

The version number should be 1.2.2.

Logically, there should be 3 corresponding configurations here.

I suggest you check the box labeled "claude" under Z:\Program Files\CliGate. I still prefer using AI for configuration.