许多小伙伴可能不知道，我两个星期前就做了一次比较完善的更新，修复了compact报错的问题，修复了无法使用google系列endpoint接入问题。现在整个体验是非常流畅的了，感兴趣的可以看这次commit：Add Google API proxy support and improve proxy service · YougLin-dev/amp-server@24224b6 · GitHub
提供一个参考配置文件。
你可以为每个端点都指定不同的上游中转，可以通过 custom_headers 提供 自定义的authorization 分开各自鉴权。
.env配置文件
HOST=127.0.0.1
PORT=8888
MAX_REQUEST_BODY_BYTES=104857600
# 写大点就行了
CONNECT_TIMEOUT_SECS=500
REQUEST_TIMEOUT_SECS=60000
# 健全，google的可以单独定义在这里，也可以像我下面提到的，直接用custom header去实现。
AMP_API_KEY=test-api-key-please-change-me
GOOGLE_API_KEY=test-api-key-please-change-me
# 日志，喜欢多点就debug，少点就info
RUST_LOG=debug
proxy_config.yaml配置文件
endpoints:
- path: "/api/provider/openai/v1/responses"
target_url: "https://你的某个支持response接口的中专/v1/responses"
method: "POST"
response_type: "stream"
custom_headers: {}
forward_request_headers:
- "authorization"
- "content-type"
- "user-agent"
- "accept"
- "accept-encoding"
forward_response_headers:
- "content-type"
- "cache-control"
enabled: true
- path: "/api/provider/anthropic/v1/messages"
target_url: "https://你的某个支持v1/messages接口的中专/v1/messages"
method: "POST"
response_type: "stream"
custom_headers: {}
forward_request_headers:
- "authorization"
- "content-type"
- "user-agent"
- "accept"
- "accept-encoding"
- "anthropic-version"
- "x-api-key"
forward_response_headers:
- "content-type"
- "cache-control"
enabled: true
- path: "/api/tab/llm-proxy"
target_url: "https://ampcode.com/api/tab/llm-proxy"
method: "POST"
response_type: "sse"
custom_headers: {}
forward_request_headers:
- "authorization"
- "user-agent"
- "x-amp-feature"
- "accept-language"
- "sec-fetch-mode"
forward_response_headers:
- "alt-svc"
- "content-security-policy"
- "fireworks-backend-host"
- "fireworks-cached-prompt-tokens"
- "fireworks-deployment"
- "fireworks-generation-queue-duration"
- "fireworks-num-concurrent-requests"
- "fireworks-prefill-duration"
- "fireworks-prefill-queue-duration"
- "fireworks-prompt-tokens"
- "fireworks-sampling-options"
- "fireworks-server-time-to-first-token"
- "fireworks-speculation-matched-tokens"
- "fireworks-speculation-prompt-tokens"
- "fireworks-tokenizer-duration"
- "fireworks-tokenizer-queue-duration"
enabled: true
# Google Generative Language (gemini)
# Matches: /api/provider/google/v1beta1/publishers/google/models/<model>:generateContent
# For example: /api/provider/google/v1beta1/publishers/google/models/gemini-2.5-flash-lite-preview-09-2025:generateContent
# Captures the remainder after /models/ as "rest"; the server derives {model} = rest.split(":").next()
- path: "/api/provider/google/v1beta1/publishers/google/models/{*rest}"
target_url: "https://某支持google格式的接口/v1/v1beta/models/gemini-2.5-pro:generateContent"
method: "POST"
response_type: "stream"
custom_headers: {}
forward_request_headers:
- "content-type"
- "user-agent"
- "accept"
- "accept-encoding"
forward_response_headers:
- "content-type"
- "cache-control"
enabled: true
如果有任何使用上的问题，大家可以在下面互相探讨，配置其实很简单。
放几个主要平台的可执行文件，晚点我写个action去自动发版吧，虽然本来比较懒没做 
window64：https://c.wss.cc/f/ia0grsp3qjp
