Custom OpenAI Responses relay does not append /v1 when fetching models
#3,320 opened on 2026/08/20
Repository metrics
- Stars
- (1 個のスター)
- PR merge metrics
- (30d に merged PR はありません)
説明
English
What happened
When a custom OpenAI Responses-compatible relay is configured with a root URL such as:
http://relay.example:3000
the connection test succeeds, but updating the model catalog fails.
Maka requests:
GET /models
However, the relay provides the OpenAI-compatible model list at:
GET /v1/models
Cherry Studio works with the same configuration because it automatically appends /v1 when the base URL does not contain an API version. OpenAI's standard endpoints are also versioned:
GET /v1/models
POST /v1/responses
Expected behavior
We should clarify whether the configured URL is a host root or a complete API base URL, and use the same URL normalization rule for model discovery and Responses requests.
Environment
- Maka: 0.1.11
- Surface: Desktop
- OS: Windows
中文
问题描述
配置自定义 OpenAI Responses 兼容中转站时，如果服务地址填写为根地址：
http://relay.example:3000
连接测试可以成功，但更新模型目录失败。
Maka 当前请求：
GET /models
而该中转站的 OpenAI 兼容模型接口为：
GET /v1/models
相同配置在 Cherry Studio 中可以使用，因为当地址不包含 API 版本时，它会自动补充 /v1。OpenAI 的标准接口也包含版本前缀：
GET /v1/models
POST /v1/responses
期望行为
需要明确配置地址表示主机根地址还是完整 API Base URL，并让模型发现和 Responses 请求使用一致的 URL 处理规则。
环境
- Maka：0.1.11
- 使用端：Desktop
- 系统：Windows