Include context

app.py

@@ -284,6 +284,29 @@ def _format_game_context_for_prompt(game_context: Optional[Union[str, Dict[str,
                         parts.append(f"Visited stations so far: {', '.join(uniq)}.")
                 except Exception:
                     pass
+            # Memory transcript (last messages)
+            mem = game_context.get('__memory') if isinstance(game_context, dict) else None
+            if isinstance(mem, dict):
+                try:
+                    transcript = mem.get('transcript') or []
+                    if transcript:
+                        # Take last few lines and embed compactly
+                        lines = []
+                        for m in transcript[-20:]:
+                            role = (m.get('role') or '').strip()
+                            src = (m.get('source') or '').strip()
+                            sta = (m.get('stationName') or '').strip()
+                            txt = (m.get('content') or '').replace('\n',' ').strip()
+                            if len(txt) > 2000:
+                                txt = txt[:2000] + '…'
+                            label = role if role else 'msg'
+                            if src or sta:
+                                label += f"[{src}{'/' + sta if sta else ''}]"
+                            lines.append(f"- {label}: {txt}")
+                        if lines:
+                            parts.append("Recent game exchanges:\n" + "\n".join(lines))
+                except Exception:
+                    pass
             return " ".join(parts)
     except Exception:
         return ""
@@ -456,6 +479,7 @@ class QueryRequest(BaseModel):
     history: list
     game_context: Optional[Union[str, Dict[str, Any]]] = None
     mode: Optional[str] = None
+    game_memory: Optional[Dict[str, Any]] = None
 
 class TTSRequest(BaseModel):
     text: str
@@ -481,6 +505,13 @@ async def predict(request: Request, query_request: QueryRequest):
     history = query_request.history
     game_context = query_request.game_context
     mode = (query_request.mode or '').lower() or None
+    # Attach optional game_memory into context for prompt formatting
+    if query_request.game_memory is not None:
+        if isinstance(game_context, dict):
+            game_context = dict(game_context)
+            game_context['__memory'] = query_request.game_memory
+        else:
+            game_context = { 'context': game_context, '__memory': query_request.game_memory }
     return StreamingResponse(stream_response(query, history, game_context, mode), media_type="text/event-stream")
 
 @app.post("/api/tts")