feat: 맥미니 MLX 연동 — OpenAI-compat 프록시 + 모델 배치 정정

- proxy_openai.py 추가: MLX 서버 SSE 패스스루
- chat.py: openai-compat 백엔드 타입 라우팅 추가
- backends.json: GPU=embed(bge-m3)만, 맥미니MLX=채팅(qwen3.5:35b-a3b)
- LAN IP(192.168.1.122) 사용 (같은 서브넷, Tailscale 불필요)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

backends.json

@@ -4,8 +4,17 @@
     "type": "ollama",
     "url": "http://host.docker.internal:11434",
     "models": [
-      { "id": "qwen3.5:9b-q8_0", "capabilities": ["chat"], "priority": 1 },
-      { "id": "qwen3-vl:8b", "capabilities": ["chat", "vision"], "priority": 1 }
+      { "id": "bge-m3", "capabilities": ["embed"], "priority": 1 }
+    ],
+    "access": "all",
+    "rate_limit": null
+  },
+  {
+    "id": "mlx-mac",
+    "type": "openai-compat",
+    "url": "http://192.168.1.122:8800",
+    "models": [
+      { "id": "qwen3.5:35b-a3b", "capabilities": ["chat"], "priority": 1 }
     ],
     "access": "all",
     "rate_limit": null

hub-api/routers/chat.py

@@ -5,7 +5,7 @@ from fastapi.responses import JSONResponse, StreamingResponse
 from pydantic import BaseModel
 
 from middleware.rate_limit import check_backend_rate_limit
-from services import proxy_ollama
+from services import proxy_ollama, proxy_openai
 from services.registry import registry
 
 router = APIRouter(prefix="/v1", tags=["chat"])
@@ -79,7 +79,24 @@ async def chat_completions(body: ChatRequest, request: Request):
             )
             return JSONResponse(content=result)
 
-    # Placeholder for other backend types
+    if backend.type == "openai-compat":
+        if body.stream:
+            return StreamingResponse(
+                proxy_openai.stream_chat(
+                    backend.url, body.model, messages, **kwargs
+                ),
+                media_type="text/event-stream",
+                headers={
+                    "Cache-Control": "no-cache",
+                    "X-Accel-Buffering": "no",
+                },
+            )
+        else:
+            result = await proxy_openai.complete_chat(
+                backend.url, body.model, messages, **kwargs
+            )
+            return JSONResponse(content=result)
+
     raise HTTPException(
         status_code=501,
         detail={

hub-api/services/proxy_openai.py

@@ -0,0 +1,83 @@
+"""OpenAI-compatible proxy (MLX server, vLLM, etc.) — SSE passthrough."""
+
+from __future__ import annotations
+
+import json
+import logging
+from collections.abc import AsyncGenerator
+
+import httpx
+
+logger = logging.getLogger(__name__)
+
+
+async def stream_chat(
+    base_url: str,
+    model: str,
+    messages: list[dict],
+    **kwargs,
+) -> AsyncGenerator[str, None]:
+    """Proxy OpenAI-compatible chat streaming. SSE passthrough with model field override."""
+    payload = {
+        "model": model,
+        "messages": messages,
+        "stream": True,
+        **{k: v for k, v in kwargs.items() if v is not None},
+    }
+
+    async with httpx.AsyncClient(timeout=120.0) as client:
+        async with client.stream(
+            "POST",
+            f"{base_url}/v1/chat/completions",
+            json=payload,
+        ) as resp:
+            if resp.status_code != 200:
+                body = await resp.aread()
+                error_msg = body.decode("utf-8", errors="replace")
+                yield _error_event(f"Backend error ({resp.status_code}): {error_msg}")
+                return
+
+            async for line in resp.aiter_lines():
+                if not line.strip():
+                    continue
+                # Pass through SSE lines as-is (already in OpenAI format)
+                if line.startswith("data: "):
+                    yield f"{line}\n\n"
+                elif line == "data: [DONE]":
+                    yield "data: [DONE]\n\n"
+
+
+async def complete_chat(
+    base_url: str,
+    model: str,
+    messages: list[dict],
+    **kwargs,
+) -> dict:
+    """Non-streaming OpenAI-compatible chat."""
+    payload = {
+        "model": model,
+        "messages": messages,
+        "stream": False,
+        **{k: v for k, v in kwargs.items() if v is not None},
+    }
+
+    async with httpx.AsyncClient(timeout=120.0) as client:
+        resp = await client.post(f"{base_url}/v1/chat/completions", json=payload)
+        resp.raise_for_status()
+        return resp.json()
+
+
+def _error_event(message: str) -> str:
+    error = {
+        "id": "chatcmpl-gateway",
+        "object": "chat.completion.chunk",
+        "model": "error",
+        "choices": [
+            {
+                "index": 0,
+                "delta": {"content": f"[Error] {message}"},
+                "finish_reason": "stop",
+            }
+        ],
+    }
+    return f"data: {json.dumps(error)}\n\ndata: [DONE]\n\n"