feat: Add RAG query rewriting for contextual follow-up questions

When a user asks "Wer ist Mieter in diesem Haus?" after discussing
a specific house, the raw message lacks context for RAG search.
Now uses a quick LLM call to resolve pronouns/references before
searching WildFiles (e.g. "diesem Haus" -> "Mieter Haus Coburg").

Also moved history fetch before RAG search so context is available.

Refs: WF-90

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

bot.py

@@ -440,19 +440,8 @@ class Bot:
     async def _respond_with_ai(self, room, user_message: str):
         model = self.room_models.get(room.room_id, DEFAULT_MODEL)
 
-        # Build conversation context from room timeline
-        messages = [{"role": "system", "content": SYSTEM_PROMPT}]
-
-        # WildFiles document context
-        doc_results = await self.rag.search(user_message)
-        doc_context = self.rag.format_context(doc_results)
-        if doc_context:
-            logger.info("RAG found %d docs for: %s", len(doc_results), user_message[:50])
-            messages.append({"role": "system", "content": doc_context})
-        else:
-            logger.info("RAG found 0 docs for: %s", user_message[:50])
-
-        # Fetch last N messages from room via API
+        # Fetch conversation history FIRST (needed for query rewriting)
+        history = []
         try:
             resp = await self.client.room_messages(
                 room.room_id, start=self.client.next_batch or "", limit=10
@@ -462,10 +451,27 @@ class Bot:
                     if not hasattr(evt, "body"):
                         continue
                     role = "assistant" if evt.sender == BOT_USER else "user"
-                    messages.append({"role": role, "content": evt.body})
+                    history.append({"role": role, "content": evt.body})
         except Exception:
             logger.debug("Could not fetch room history, proceeding without context")
 
+        # Rewrite query using conversation context for better RAG search
+        search_query = await self._rewrite_query(user_message, history, model)
+
+        # WildFiles document context
+        doc_results = await self.rag.search(search_query)
+        doc_context = self.rag.format_context(doc_results)
+        if doc_context:
+            logger.info("RAG found %d docs for: %s (original: %s)", len(doc_results), search_query[:50], user_message[:50])
+        else:
+            logger.info("RAG found 0 docs for: %s (original: %s)", search_query[:50], user_message[:50])
+
+        # Build conversation context
+        messages = [{"role": "system", "content": SYSTEM_PROMPT}]
+        if doc_context:
+            messages.append({"role": "system", "content": doc_context})
+        messages.extend(history)
+
         # Add current user message
         messages.append({"role": "user", "content": user_message})
 
@@ -485,6 +491,43 @@ class Bot:
             logger.exception("LLM call failed")
             await self._send_text(room.room_id, "Sorry, I couldn't generate a response.")
 
+    async def _rewrite_query(self, user_message: str, history: list[dict], model: str) -> str:
+        """Rewrite user message into a standalone search query using conversation context."""
+        if not history or not self.llm:
+            return user_message
+
+        # Build a compact history summary (last 4 messages max)
+        recent = history[-4:]
+        context_lines = []
+        for msg in recent:
+            prefix = "User" if msg["role"] == "user" else "Assistant"
+            context_lines.append(f"{prefix}: {msg['content'][:200]}")
+        context_text = "\n".join(context_lines)
+
+        try:
+            resp = await self.llm.chat.completions.create(
+                model=model,
+                messages=[
+                    {"role": "system", "content": (
+                        "You are a search query rewriter. Given conversation history and a new user message, "
+                        "produce a single standalone search query that resolves all pronouns and references "
+                        "(like 'this house', 'that document', 'it') using context from the conversation. "
+                        "Reply with ONLY the rewritten search query in the same language as the user message. "
+                        "No explanation, no quotes. If the message is already self-contained, return it as-is."
+                    )},
+                    {"role": "user", "content": f"Conversation:\n{context_text}\n\nNew message: {user_message}"},
+                ],
+                max_tokens=100,
+            )
+            rewritten = resp.choices[0].message.content.strip().strip('"\'')
+            if rewritten and len(rewritten) < 500:
+                logger.info("Query rewritten: '%s' -> '%s'", user_message[:50], rewritten[:50])
+                return rewritten
+        except Exception:
+            logger.debug("Query rewrite failed, using original", exc_info=True)
+
+        return user_message
+
     async def _auto_rename_room(self, room, user_message: str, ai_reply: str):
         """Generate a short topic title and set it as the room name."""
         try: