Skip to content

100+ AI Agents, Agent Skills and RAG Apps - Free and Open Source.

Store only the candidate's message in Mem0, not the coach's reply

ManpreetLikhariShorthillsAI56fb5ea9c35a
diff --git a/advanced_llm_apps/llm_apps_with_memory_tutorials/adk_career_coach_agent_memory/adk_career_coach_agent_memory.py b/advanced_llm_apps/llm_apps_with_memory_tutorials/adk_career_coach_agent_memory/adk_career_coach_agent_memory.py
index 76f1d027..3035d166 100644
--- a/advanced_llm_apps/llm_apps_with_memory_tutorials/adk_career_coach_agent_memory/adk_career_coach_agent_memory.py
+++ b/advanced_llm_apps/llm_apps_with_memory_tutorials/adk_career_coach_agent_memory/adk_career_coach_agent_memory.py
@@ -300,8 +300,12 @@ if google_api_key:
st.session_state.messages.append({"role": "assistant", "content": answer})
# Write this exchange back to long-term memory for future sessions.
+ # Only the candidate's own message goes in: one memory.add() call
+ # instead of two (half the extraction latency/cost, since infer=True
+ # runs an LLM call per add()), and the fact store stays candidate
+ # facts only -- the coach's own advice text never gets treated as a
+ # fact about the candidate.
memory.add(prompt, user_id=user_id, metadata={"role": "user"})
- memory.add(answer, user_id=user_id, metadata={"role": "assistant"})
elif not user_id:
st.error("Please enter your email to start chatting.")
else:
diff --git a/advanced_llm_apps/llm_apps_with_memory_tutorials/adk_career_coach_agent_memory/idea.md b/advanced_llm_apps/llm_apps_with_memory_tutorials/adk_career_coach_agent_memory/idea.md
index 970c3466..79b49cdb 100644
--- a/advanced_llm_apps/llm_apps_with_memory_tutorials/adk_career_coach_agent_memory/idea.md
+++ b/advanced_llm_apps/llm_apps_with_memory_tutorials/adk_career_coach_agent_memory/idea.md
@@ -127,8 +127,11 @@ Step by step, in plain terms:
tool to look something up (a sample interview question, a salary
benchmark, etc.) before answering.
5. The specialist's answer comes back through the orchestrator to the app.
-6. The app saves both your message and the answer into Mem0, so this
- exchange is available to recall in future conversations.
+6. The app saves your message into Mem0 -- just your side of the exchange,
+ not the coach's reply. The long-term store is meant to hold facts about
+ you (the candidate), not the advice the coach gave, so only your message
+ goes in. This also means only one Mem0 write per turn instead of two,
+ which matters because each write triggers its own LLM call internally.
7. The answer is shown to you in the chat window.
## Two very different kinds of "remembering"