Proposal: Use api keys with openai server api: A simple example with gitlab tokens

So, I would like to have ANY way of making sure that my users are authenticated before they use the openai api server.

My rationale is: If they have our server gitlab account, and they have an api key there, they can use the openai_server_api of fastchat.

This patch below adds support for gitlab keys.

With that, I can use continue.dev, for example, with FastChat being the inference server. So the same system provides me the chat web interface and the code completion/explain for VSCode.

My coding skills don't take me much further than that - but we could do some auth scheme for different providers, what do you think?

diff --git a/fastchat/serve/openai_api_server.py b/fastchat/serve/openai_api_server.py
index 9743bde..88e890c 100644
--- a/fastchat/serve/openai_api_server.py
+++ b/fastchat/serve/openai_api_server.py
@@ -88,6 +88,7 @@ async def fetch_remote(url, pload=None, name=None):
 class AppSettings(BaseSettings):
     # The address of the model controller.
     controller_address: str = "http://localhost:21001"
+    authenticator_url = "https://gitlab.com/api/v4/user"
     api_keys: Optional[List[str]] = None
 
 
@@ -115,6 +116,25 @@ async def check_api_key(
             )
         return token
     else:
+        if auth is not None and auth.credentials is not None:
+            # gitlab expect the token bearer with the string PRIVATE-TOKEN
+            headers = {"PRIVATE-TOKEN": auth.credentials}
+            response = httpx.get(app_settings.authenticator_url, headers=headers)
+            print(response.status_code)
+            if response.status_code == 200:
+                    return auth.credentials
+            raise HTTPException(
+                status_code=401,
+                detail={
+                    "error": {
+                        "message": "",
+                        "type": "invalid_request_error",
+                        "param": None,
+                        "code": "invalid_api_key",
+                    }
+                },
+            )

Start in fastchat/serve/openai_api_server.py, reading AppSettings and check_api_key, then trace how authentication is applied to the OpenAI-compatible endpoints. Compare the proposed GitLab token validation with the existing api_keys behavior. Done means the supported authentication flow is defined and users with valid GitLab credentials can access the server while invalid credentials are rejected.