Is it bad practice to create_agent all the time?

OSS Product Help LangChain
,
1

Hey guys,

I cant seem to wrap this around my head

When I use PostgresSQL for memory, it seems that I have to do create_agent all the time since PostgresSQL is required to be wrapped with the with statement

i.e:


with PostgresSaver.from_conn_string(SUPABASE_DB_URL) as checkpointer:
   agent = build_agent("gpt-4o-mini", checkpointer) thread_id = "user-123"
   config = {"configurable": {"thread_id": "user-123"}}

    print(invoke_agent(agent, thread_id, "What's the capital of the moon?"))

So does that mean every time I ask a follow up question, I’ll need to call build_agent again? There has to be a better way to do this (other than refactoring to a function) because won’t it be redundant and create memory issues?

Thanks

1 Like
2

hi @MartinYTSo

try this:

# Create the context manager
_checkpointer_cm = PostgresSaver.from_conn_string(os.getenv("POSTGRES_URI"))
# Enter it once and keep the real PostgresSaver
checkpointer = _checkpointer_cm.__enter__()
checkpointer.setup()  # once on startup
# ...
agent = create_agent("gpt-4o-mini", checkpointer=checkpointer)
1 Like
3

Thank you it worked! may I ask what is your logic?

4

your welcome @MartinYTSo :slight_smile:
What do you mean by “my logic”?

Huge favour: could you mark this post as Solved if it solves your issue?

5

I mean how did you know what to do. I know this is kind of trivial but I’m kind of new to langchain

6

No need to call the build agent each time. Create the agent once inside the with block and reuse it for all your questions.

7

Hmmm I read articles, study the codebases, help here on the forum solving problems, code… I dunno, I just like programming and Langchain specifically :slight_smile:

1 Like