Add an openrouter provider

OpenRouter is a "muxing provider" which itself provides access to
multiple models and providers. It speaks a dialect of the OpenAI protocol, but
for our purposes, we can say it's OpenAI.

There are some differences in handling the requests, though:
1) we need to know where to forward the request to, by default this is
   `https://openrouter.ai/api/v1`, this is done by setting the base_url
   parameter
2) we need to prefix the model with `openrouter/`. This is a
   lite-LLM-ism (see https://docs.litellm.ai/docs/providers/openrouter)
   which we'll be able to remove once we ditch litellm

Initially I was considering just exposing the OpenAI provider on an
additional route and handling the prefix based on the route, but I think
having an explicit provider class is better as it allows us to handle
any differences in OpenRouter dialect easily in the future.

Related: #878

Author: jhrozek

src/codegate/config.py

@@ -17,6 +17,7 @@
 # Default provider URLs
 DEFAULT_PROVIDER_URLS = {
     "openai": "https://api.openai.com/v1",
+    "openrouter": "https://openrouter.ai/api/v1",
     "anthropic": "https://api.anthropic.com/v1",
     "vllm": "http://localhost:8000",  # Base URL without /v1 path
     "ollama": "http://localhost:11434",  # Default Ollama server URL

src/codegate/providers/__init__.py

@@ -2,13 +2,15 @@
 from codegate.providers.base import BaseProvider
 from codegate.providers.ollama.provider import OllamaProvider
 from codegate.providers.openai.provider import OpenAIProvider
+from codegate.providers.openrouter.provider import OpenRouterProvider
 from codegate.providers.registry import ProviderRegistry
 from codegate.providers.vllm.provider import VLLMProvider
 
 __all__ = [
     "BaseProvider",
     "ProviderRegistry",
     "OpenAIProvider",
+    "OpenRouterProvider",
     "AnthropicProvider",
     "VLLMProvider",
     "OllamaProvider",

src/codegate/providers/openai/__init__.py

@@ -0,0 +1,3 @@
+from codegate.providers.openai.provider import OpenAIProvider
+
+__all__ = ["OpenAIProvider"]

src/codegate/providers/openrouter/provider.py

@@ -0,0 +1,47 @@
+import json
+
+from fastapi import Header, HTTPException, Request
+
+from codegate.clients.detector import DetectClient
+from codegate.pipeline.factory import PipelineFactory
+from codegate.providers.openai import OpenAIProvider
+
+
+class OpenRouterProvider(OpenAIProvider):
+    def __init__(self, pipeline_factory: PipelineFactory):
+        super().__init__(pipeline_factory)
+
+    @property
+    def provider_route_name(self) -> str:
+        return "openrouter"
+
+    def _setup_routes(self):
+        @self.router.post(f"/{self.provider_route_name}/api/v1/chat/completions")
+        @self.router.post(f"/{self.provider_route_name}/chat/completions")
+        @DetectClient()
+        async def create_completion(
+            request: Request,
+            authorization: str = Header(..., description="Bearer token"),
+        ):
+            if not authorization.startswith("Bearer "):
+                raise HTTPException(status_code=401, detail="Invalid authorization header")
+
+            api_key = authorization.split(" ")[1]
+            body = await request.body()
+            data = json.loads(body)
+
+            base_url = self._get_base_url()
+            data["base_url"] = base_url
+
+            # litellm workaround - add openrouter/ prefix to model name to make it openai-compatible
+            # once we get rid of litellm, this can simply be removed
+            original_model = data.get("model", "")
+            if not original_model.startswith("openrouter/"):
+                data["model"] = f"openrouter/{original_model}"
+
+            return await self.process_request(
+                data,
+                api_key,
+                request.url.path,
+                request.state.detected_client,
+            )

src/codegate/providers/vllm/provider.py

@@ -9,7 +9,6 @@
 
 from codegate.clients.clients import ClientType
 from codegate.clients.detector import DetectClient
-from codegate.config import Config
 from codegate.pipeline.factory import PipelineFactory
 from codegate.providers.base import BaseProvider, ModelFetchError
 from codegate.providers.litellmshim import LiteLLmShim, sse_stream_generator

src/codegate/server.py

@@ -18,6 +18,7 @@
 from codegate.providers.lm_studio.provider import LmStudioProvider
 from codegate.providers.ollama.provider import OllamaProvider
 from codegate.providers.openai.provider import OpenAIProvider
+from codegate.providers.openrouter.provider import OpenRouterProvider
 from codegate.providers.registry import ProviderRegistry, get_provider_registry
 from codegate.providers.vllm.provider import VLLMProvider
 
@@ -75,6 +76,10 @@ async def log_user_agent(request: Request, call_next):
         ProviderType.openai,
         OpenAIProvider(pipeline_factory),
     )
+    registry.add_provider(
+        ProviderType.openai,
+        OpenRouterProvider(pipeline_factory),
+    )
     registry.add_provider(
         ProviderType.anthropic,
         AnthropicProvider(

tests/test_server.py

@@ -108,8 +108,8 @@ def test_provider_registration(mock_registry, mock_secrets_mgr, mock_pipeline_fa
     # Verify all providers were registered
     registry_instance = mock_registry.return_value
     assert (
-        registry_instance.add_provider.call_count == 6
-    )  # openai, anthropic, llamacpp, vllm, ollama, lm_studio
+        registry_instance.add_provider.call_count == 7
+    )  # openai, anthropic, llamacpp, vllm, ollama, lm_studio, openrouter
 
     # Verify specific providers were registered
     provider_names = [call.args[0] for call in registry_instance.add_provider.call_args_list]