ff-174 Custom API entry points for OpenAI and Gemini clients. (#304)

README.md

@@ -109,6 +109,17 @@ You can set the API key for collections in the processor's config.
 
 **DANGER!!!** You can set the "general API key" in the processor's config; in this case, the processor will use it to process **ALL** news. It may be convenient if you self-host the service and fully control who has access to it.
 
+#### Specify API entry points
+
+You can set custom URLs as entry points for OpenAi and Gemini API by setting nthe ext environment variables:
+
+```
+FFUN_OPENAI_API_ENTRY_POINT="<your url>"
+FFUN_GOOGLE_GEMINI_API_ENTRY_POINT="<your url>"
+```
+
+That will allow you to use any compatible API provider.
+
 ## Backend
 
 ```

changes/unreleased.md

@@ -1,2 +1,2 @@
 
-No changes.
+- ff-174 — Custom API entry points for OpenAI and Gemini clients.

ffun/ffun/openai/provider_interface.py

@@ -12,6 +12,7 @@
 from ffun.llms_framework.entities import KeyStatus, LLMConfiguration, LLMProvider
 from ffun.llms_framework.keys_statuses import Statuses
 from ffun.llms_framework.provider_interface import ChatRequest, ChatResponse, ProviderInterface
+from ffun.openai.settings import settings
 
 logger = logging.get_module_logger()
 
@@ -20,6 +21,10 @@ class OpenAIChatRequest(ChatRequest):
     messages: list[ChatCompletionMessageParam]
 
 
+def _client(api_key: str) -> openai.AsyncOpenAI:
+    return openai.AsyncOpenAI(api_key=api_key, base_url=settings.api_entry_point, timeout=settings.api_timeout)
+
+
 class OpenAIChatResponse(ChatResponse):
     content: str
     prompt_tokens: int
@@ -82,7 +87,7 @@ async def chat_request(  # type: ignore
     ) -> OpenAIChatResponse:
         try:
             with track_key_status(api_key, self.api_keys_statuses):
-                answer = await openai.AsyncOpenAI(api_key=api_key).chat.completions.create(
+                answer = await _client(api_key=api_key).chat.completions.create(
                     model=config.model,
                     temperature=float(config.temperature),
                     max_tokens=config.max_return_tokens,
@@ -127,7 +132,7 @@ def prepare_requests(self, config: LLMConfiguration, text: str) -> Sequence[Open
     async def check_api_key(self, config: LLMConfiguration, api_key: str) -> KeyStatus:
         with track_key_status(api_key, self.api_keys_statuses):
             try:
-                await openai.AsyncOpenAI(api_key=api_key).models.list()
+                await _client(api_key=api_key).models.list()
             except openai.APIError:
                 pass
 

ffun/ffun/openai/settings.py

@@ -0,0 +1,18 @@
+import pathlib
+
+import pydantic_settings
+
+from ffun.core.settings import BaseSettings
+
+_root = pathlib.Path(__file__).parent
+
+
+class Settings(BaseSettings):
+
+    api_entry_point: str | None = None
+    api_timeout: float = 20.0
+
+    model_config = pydantic_settings.SettingsConfigDict(env_prefix="FFUN_OPENAI_")
+
+
+settings = Settings()