router: add tool manager runtime and memory retrieval updates

services/router/memory_retrieval.py

@@ -18,6 +18,7 @@ Collections per agent:
 import os
 import json
 import logging
+import re
 from typing import Optional, Dict, Any, List
 from dataclasses import dataclass, field
 from datetime import datetime
@@ -453,7 +454,7 @@ class MemoryRetrieval:
                     # Higher threshold for messages; even higher when user asks about docs to avoid pulling old chatter.
                     msg_thresh = 0.5 if is_doc_query else 0.4
                     if r.score > msg_thresh:
-                        text = r.payload.get("text", r.payload.get("content", ""))
+                        text = self._extract_message_text(r.payload)
                         # Skip very short or system messages
                         if len(text) > 20 and not text.startswith("<"):
                             if is_doc_query and topic_keywords:
@@ -501,7 +502,7 @@ class MemoryRetrieval:
             seen_texts = set()
             unique_results = []
             for r in all_results:
-                text_key = r.get("text", "")[:50].lower()
+                text_key = self._canonical_text_key(r.get("text", ""))
                 if text_key not in seen_texts:
                     seen_texts.add(text_key)
                     unique_results.append(r)
@@ -511,6 +512,40 @@ class MemoryRetrieval:
         except Exception as e:
             logger.warning(f"Memory search failed for {agent_id}: {e}")
             return []
+
+    @staticmethod
+    def _extract_message_text(payload: Dict[str, Any]) -> str:
+        """
+        Normalize text across both payload schemas:
+        - memory-service: content/text (+ role/channel_id)
+        - router: user_message + assistant_response (+ chat_id)
+        """
+        if not payload:
+            return ""
+
+        text = (payload.get("text") or payload.get("content") or "").strip()
+        if text:
+            lower = text.lower()
+            marker = "\n\nassistant:"
+            idx = lower.rfind(marker)
+            if lower.startswith("user:") and idx != -1:
+                assistant_text = text[idx + len(marker):].strip()
+                if assistant_text:
+                    return assistant_text
+            return text
+
+        user_text = (payload.get("user_message") or "").strip()
+        assistant_text = (payload.get("assistant_response") or "").strip()
+        if user_text and assistant_text:
+            return f"User: {user_text}\n\nAssistant: {assistant_text}"
+        return user_text or assistant_text
+
+    @staticmethod
+    def _canonical_text_key(text: str) -> str:
+        if not text:
+            return ""
+        normalized = re.sub(r"\s+", " ", text.strip().lower())
+        return normalized[:220]
     
     async def get_user_graph_context(
         self,
@@ -619,6 +654,8 @@ class MemoryRetrieval:
                 query=message,
                 agent_id=agent_id,
                 platform_user_id=identity.platform_user_id,
+                chat_id=chat_id,
+                user_id=user_id,
                 limit=5
             )
             brief.relevant_memories = memories