refac

backend/open_webui/main.py

@@ -252,7 +252,7 @@ from open_webui.utils.oauth import (
 from open_webui.utils.plugin import install_tool_and_function_dependencies
 from open_webui.utils.redis import get_redis_client
 from open_webui.utils.security_headers import SecurityHeadersMiddleware
-from open_webui.utils.session_pool import get_session
+from open_webui.utils.session_pool import cleanup_response, get_session, stream_wrapper
 from open_webui.utils.tools import set_terminal_servers, set_tool_servers
 
 if SAFE_MODE:
@@ -1779,6 +1779,7 @@ app.state.CHAT_COMPLETION_HANDLER = chat_completion
 from open_webui.utils.anthropic import (
     convert_anthropic_to_openai_payload,
     convert_openai_to_anthropic_response,
+    is_anthropic_messages_passthrough,
     openai_stream_to_anthropic_stream,
 )
 
@@ -1793,6 +1794,65 @@ async def count_message_tokens(
     return {'input_tokens': await openai.count_anthropic_tokens(request, form_data, user)}
 
 
+async def passthrough_anthropic_messages(request: Request, form_data: dict, user) -> Response | dict:
+    requested_model, payload, url, key, headers, cookies = await openai.get_anthropic_token_count_target(
+        request, form_data, user
+    )
+    request_url = f'{url.rstrip("/")}/messages'
+    response = None
+    streaming = False
+
+    try:
+        session = await get_session()
+        response = await session.request(
+            method='POST',
+            url=request_url,
+            data=json.dumps(payload),
+            headers=headers,
+            cookies=cookies,
+            ssl=AIOHTTP_CLIENT_SESSION_SSL,
+            timeout=aiohttp.ClientTimeout(total=openai.AIOHTTP_CLIENT_TIMEOUT),
+        )
+
+        if 'text/event-stream' in response.headers.get('Content-Type', ''):
+            streaming = True
+            return StreamingResponse(
+                stream_wrapper(response),
+                status_code=response.status,
+                headers=openai._clean_proxy_headers(response.headers),
+            )
+
+        try:
+            response_data = await response.json()
+        except Exception:
+            response_data = await response.text()
+
+        if response.status >= 400:
+            await openai.publish_model_provider_request_failed(
+                request,
+                actor=user,
+                provider='openai-compatible',
+                base_url=url,
+                api_key=key,
+                status=response.status,
+                requested_model=requested_model,
+                upstream_error=response_data,
+            )
+            if isinstance(response_data, (dict, list)):
+                return JSONResponse(status_code=response.status, content=response_data)
+            return Response(status_code=response.status, content=response_data)
+
+        return response_data
+    except HTTPException:
+        raise
+    except Exception:
+        log.exception('Failed to passthrough Anthropic Messages request for model %s', requested_model)
+        raise HTTPException(status_code=502, detail=ERROR_MESSAGES.SERVER_CONNECTION_ERROR)
+    finally:
+        if not streaming:
+            await cleanup_response(response)
+
+
 @app.post('/api/message')
 @app.post('/api/v1/messages')  # Anthropic Messages API compatible endpoint
 async def generate_messages(
@@ -1833,7 +1893,9 @@ async def generate_messages(
         models = request.app.state.OPENAI_MODELS
     model = models.get(model_id)
     if model:
-        _, _, api_config = await openai.get_openai_connection(model['urlIdx'])
+        url, _, api_config = await openai.get_openai_connection(model['urlIdx'])
+        if is_anthropic_messages_passthrough(url, api_config):
+            return await passthrough_anthropic_messages(request, form_data, user)
         passthrough_params = api_config.get('passthrough_params') or []
 
     # Convert Anthropic payload to OpenAI format