Generates a response to the user's query, incorporating memory and context.
(self, query: str, relationship_with_user="friend", style_hint="", user_conversation_meta_data: dict = None)
| 244 | self._trigger_profile_and_knowledge_update_if_needed() |
| 245 | |
| 246 | def get_response(self, query: str, relationship_with_user="friend", style_hint="", user_conversation_meta_data: dict = None) -> str: |
| 247 | """ |
| 248 | Generates a response to the user's query, incorporating memory and context. |
| 249 | """ |
| 250 | print(f"Memoryos: Generating response for query: '{query[:50]}...'") |
| 251 | |
| 252 | # 1. Retrieve context |
| 253 | retrieval_results = self.retriever.retrieve_context( |
| 254 | user_query=query, |
| 255 | user_id=self.user_id |
| 256 | # Using default thresholds from Retriever class for now |
| 257 | ) |
| 258 | retrieved_pages = retrieval_results["retrieved_pages"] |
| 259 | retrieved_user_knowledge = retrieval_results["retrieved_user_knowledge"] |
| 260 | retrieved_assistant_knowledge = retrieval_results["retrieved_assistant_knowledge"] |
| 261 | |
| 262 | # 2. Get short-term history |
| 263 | short_term_history = self.short_term_memory.get_all() |
| 264 | history_text = "\n".join([ |
| 265 | f"User: {qa.get('user_input', '')}\nAssistant: {qa.get('agent_response', '')} (Time: {qa.get('timestamp', '')})" |
| 266 | for qa in short_term_history |
| 267 | ]) |
| 268 | |
| 269 | # 3. Format retrieved mid-term pages (retrieval_queue equivalent) |
| 270 | retrieval_text = "\n".join([ |
| 271 | f"【Historical Memory】\nUser: {page.get('user_input', '')}\nAssistant: {page.get('agent_response', '')}\nTime: {page.get('timestamp', '')}\nConversation chain overview: {page.get('meta_info','N/A')}" |
| 272 | for page in retrieved_pages |
| 273 | ]) |
| 274 | |
| 275 | # 4. Get user profile |
| 276 | user_profile_text = self.user_long_term_memory.get_raw_user_profile(self.user_id) |
| 277 | if not user_profile_text or user_profile_text.lower() == "none": |
| 278 | user_profile_text = "No detailed profile available yet." |
| 279 | |
| 280 | # 5. Format retrieved user knowledge for background |
| 281 | user_knowledge_background = "" |
| 282 | if retrieved_user_knowledge: |
| 283 | user_knowledge_background = "\n【Relevant User Knowledge Entries】\n" |
| 284 | for kn_entry in retrieved_user_knowledge: |
| 285 | user_knowledge_background += f"- {kn_entry['knowledge']} (Recorded: {kn_entry['timestamp']})\n" |
| 286 | |
| 287 | background_context = f"【User Profile】\n{user_profile_text}\n{user_knowledge_background}" |
| 288 | |
| 289 | # 6. Format retrieved Assistant Knowledge (from assistant's LTM) |
| 290 | # Use retrieved assistant knowledge instead of all assistant knowledge |
| 291 | assistant_knowledge_text_for_prompt = "【Assistant Knowledge Base】\n" |
| 292 | if retrieved_assistant_knowledge: |
| 293 | for ak_entry in retrieved_assistant_knowledge: |
| 294 | assistant_knowledge_text_for_prompt += f"- {ak_entry['knowledge']} (Recorded: {ak_entry['timestamp']})\n" |
| 295 | else: |
| 296 | assistant_knowledge_text_for_prompt += "- No relevant assistant knowledge found for this query.\n" |
| 297 | |
| 298 | # 7. Format user_conversation_meta_data (if provided) |
| 299 | meta_data_text_for_prompt = "【Current Conversation Metadata】\n" |
| 300 | if user_conversation_meta_data: |
| 301 | try: |
| 302 | meta_data_text_for_prompt += json.dumps(user_conversation_meta_data, ensure_ascii=False, indent=2) |
| 303 | except TypeError: |