.

backends_base.py

@@ -0,0 +1,12 @@
+# backends_base.py
+from typing import Any, AsyncIterable, Dict
+
+class ChatBackend:
+    """Abstract interface for chat backends (OpenAI-compatible)."""
+    async def stream(self, request: Dict[str, Any]) -> AsyncIterable[Dict[str, Any]]:
+        raise NotImplementedError
+
+class ImagesBackend:
+    """Abstract interface for image backends (OpenAI-compatible)."""
+    async def generate_b64(self, request: Dict[str, Any]) -> str:
+        raise NotImplementedError

oa_server.py

@@ -2,6 +2,7 @@
 from __future__ import annotations
 import json, time, uuid, logging
 from typing import Any, Dict, List, AsyncIterable, Optional
+from backends_base import ChatBackend, ImagesBackend
 
 from rabbit_repo import RabbitRepo
 

vllm_backend.py

@@ -3,7 +3,7 @@ import time, logging
 from typing import Any, Dict, AsyncIterable
 
 from vllm.sampling_params import SamplingParams
-from oa_server import ChatBackend, ImagesBackend
+from backends_base import ChatBackend, ImagesBackend
 from state import vllm_engine   # ✅ the single source of truth
 
 logger = logging.getLogger(__name__)