从 Cline、Roo Code、Kilo Code 这类IDE插件，到 Codex、Claude Code、Open Code 这类CLI工具，Vibe Coding 正在加速进化中，如今 Copilot Chat 已成为 VS Code 的内置扩展，AI的迭代让很多用户跟不上步伐。
本着少折腾的原则，Copilot Chat + Codex 算是当前更适合快速转型的方案，后续可以根据需求，将 Codex 替换为 Claude Code、Open Code 或与 Codex 共存使用。
Copilot Chat
虽然官方给出了使用 chat.disableAIFeatures 禁用 Copilot Chat 的方法，但实际使用中，为 Copilot Chat 配置自定义模型，可以更好的把控项目及处理一些零碎场景。
使用开源插件 OAI Compatible Provider for Copilot 能够实现自定义模型的效果，安装插件后，打开设置，配置 oaicopilot.models 中的 setting.json 使插件生效。
这里给出一份添加 DeepSeek V4 的示例，配置中没有密钥，首次运行模型时会弹窗要求输入密钥。
"oaicopilot.baseUrl": "https://api.deepseek.com",
"oaicopilot.models": [
{
"id": "deepseek-v4-flash",
"displayName": "DeepSeek V4 Flash",
"configId": "deepseek-v4-flash",
"owned_by": "deepseek",
"context_length": 192000,
"max_tokens": 38400,
"temperature": 0
},
{
"id": "deepseek-v4-flash",
"displayName": "DeepSeek V4 Flash (Think)",
"configId": "deepseek-v4-flash-think-max",
"owned_by": "deepseek",
"context_length": 192000,
"max_tokens": 38400,
"temperature": 0,
"include_reasoning_in_request": true,
"thinking": {
"type": "enabled"
},
"extra": {
"reasoning_effort": "max"
}
},
{
"id": "deepseek-v4-pro",
"displayName": "DeepSeek V4 Pro",
"configId": "deepseek-v4-pro",
"owned_by": "deepseek",
"context_length": 192000,
"max_tokens": 38400,
"temperature": 0
},
{
"id": "deepseek-v4-pro",
"displayName": "DeepSeek V4 Pro (Think)",
"configId": "deepseek-v4-pro-think-max",
"owned_by": "deepseek",
"context_length": 192000,
"max_tokens": 38400,
"temperature": 0,
"include_reasoning_in_request": true,
"thinking": {
"type": "enabled"
},
"extra": {
"reasoning_effort": "max"
}
}
]
在部分情况下，仅使用一个厂商的模型是不够用的，从中转站接入 Claude 是很有性价比的，在接入 DeepSeek 的基础上，给出一份接入第三方中转站的示例。
"oaicopilot.baseUrl": "https://api.deepseek.com",
"oaicopilot.models": [
{
"id": "deepseek-v4-flash",
"displayName": "DeepSeek V4 Flash",
"configId": "deepseek-v4-flash",
"owned_by": "deepseek",
"context_length": 192000,
"max_tokens": 38400,
"temperature": 0
},
{
"id": "deepseek-v4-flash",
"displayName": "DeepSeek V4 Flash (Think)",
"configId": "deepseek-v4-flash-think-max",
"owned_by": "deepseek",
"context_length": 192000,
"max_tokens": 38400,
"temperature": 0,
"include_reasoning_in_request": true,
"thinking": {
"type": "enabled"
},
"extra": {
"reasoning_effort": "max"
}
},
{
"id": "deepseek-v4-pro",
"displayName": "DeepSeek V4 Pro",
"configId": "deepseek-v4-pro",
"owned_by": "deepseek",
"context_length": 192000,
"max_tokens": 38400,
"temperature": 0
},
{
"id": "deepseek-v4-pro",
"displayName": "DeepSeek V4 Pro (Think)",
"configId": "deepseek-v4-pro-think-max",
"owned_by": "deepseek",
"context_length": 192000,
"max_tokens": 38400,
"temperature": 0,
"include_reasoning_in_request": true,
"thinking": {
"type": "enabled"
},
"extra": {
"reasoning_effort": "max"
}
},
{
"id": "claude-sonnet-4-6",
"displayName": "Claude Sonnet 4.6",
"configId": "claude-sonnet-4-6",
"owned_by": "anthropic",
"baseUrl": "https://your-api-domain",
"apiMode": "anthropic",
"context_length": 192000,
"max_tokens": 38400,
"temperature": 0,
"headers": {
"User-Agent": "CherryStudio/1.7.25"
}
},
{
"id": "claude-sonnet-4-6",
"displayName": "Claude Sonnet 4.6 (Think)",
"configId": "claude-sonnet-4-6-think-high",
"owned_by": "anthropic",
"baseUrl": "https://your-api-domain",
"apiMode": "anthropic",
"context_length": 192000,
"max_tokens": 38400,
"temperature": 0,
"headers": {
"User-Agent": "CherryStudio/1.7.25"
},
"include_reasoning_in_request": true,
"extra": {
"stream": true,
"thinking": {
"type": "adaptive"
}
}
},
{
"id": "claude-opus-4-7",
"displayName": "Claude Opus 4.7",
"configId": "claude-opus-4-7",
"owned_by": "anthropic",
"baseUrl": "https://your-api-domain",
"apiMode": "anthropic",
"context_length": 192000,
"max_tokens": 38400,
"temperature": 0,
"headers": {
"User-Agent": "CherryStudio/1.7.25"
}
},
{
"id": "claude-opus-4-7",
"displayName": "Claude Opus 4.7 (Think)",
"configId": "claude-opus-4-7-think-high",
"owned_by": "anthropic",
"baseUrl": "https://your-api-domain",
"apiMode": "anthropic",
"context_length": 192000,
"max_tokens": 38400,
"temperature": 0,
"headers": {
"User-Agent": "CherryStudio/1.7.25"
},
"extra": {
"stream": true,
"thinking": {
"type": "adaptive"
}
}
}
]
这里需要说明两个参数：
“baseUrl”: “https://your-api-domain” 需要修改成中转站的地址
“User-Agent”: “CherryStudio/1.7.25” 可以让一些限制客户端的中转站放行
其它的参数可以去原项目中查看文档，这里不做过多解释。
Codex
全自动干活的神器，这里主要讲解如何配置中转站的API使用。
安装好插件后，选择通过API模式登录，API密钥使用在中转站配置的。
然后在根目录找到 .codex 文件夹中的 config.toml 进行配置：
model_provider = "OpenAI" model = "gpt-5.5" review_model = "gpt-5.4" model_reasoning_effort = "xhigh" disable_response_storage = true network_access = "enabled" windows_wsl_setup_acknowledged = true model_context_window = 1000000 model_auto_compact_token_limit = 900000 personality = "pragmatic" [model_providers.OpenAI] name = "OpenAI" base_url = "https://your-api-domain" wire_api = "responses" requires_openai_auth = true
将 base_url = “https://your-api-domain” 修改中中转站API的地址即可，一般主流的中转站也都会在帮助文档里给出 Codex 的配置。
最后建议在 Codex 的设置中，添加一个叫 superpowers 的插件，注意不是在 VSCode 的插件中心添加，会强化 Codex 的工程能力。
中转站
如何选择可靠的中转站，这里我不推荐具体的站点，只提供一个思路，去一些相关开源项目的 Readme 中即可找到答案，例如：sub2api、new-api、cc-switch等…