Crew AI memory and chatbot

General
1

Hey all.

I have a chat app. I’m using crews to respond to user inquiries in the chat.

I would like crews to provide answers and rely on a context from a previous questions.

My expectation is that Memory feature can help with that.

I configured my crew in the following way:

dynamic_crew = Crew(
                agents=crew_agents,
                tasks=crew_tasks,
                verbose=True,
                memory=True,
                embedder={
                    "provider": "google",
                    "config": {
                        "api_key": os.getenv("GEMINI_API_KEY"),
                        "model": "models/text-embedding-004"
                    }
                },
                long_term_memory=LongTermMemory(storage=LTMSQLiteStorage(
                    db_path=
                    "./memorydb/dynamic_crew_%s/long_term_memory_storage.db" %
                    user_id)),
                short_term_memory=ShortTermMemory(storage=RAGStorage(
                    type="short_term",
                    embedder_config={
                        "provider": "google",
                        "config": {
                            "api_key": os.getenv("GEMINI_API_KEY"),
                            "model": "models/text-embedding-004"
                        }
                    },
                    path="./memorydb/dynamic_crew_%s/" % user_id)),
                entity_memory=EntityMemory(storage=RAGStorage(
                    type="short_term",
                    embedder_config={
                        "provider": "google",
                        "config": {
                            "api_key": os.getenv("GEMINI_API_KEY"),
                            "model": "models/text-embedding-004"
                        }
                    },
                    path="./memorydb/dynamic_crew_%s/" % user_id)),
            )

As you see for each user the memory is stored in a different folder.
I see the files are stored in the folder and memories are written, but at the same time I don’t see context being used.

Like I ask a simple question: How old is Nike brand? - got a response.
The next question is: “What about Columbia?” - the crew response is not about the age at all.

Are my expectations wrong? I saw this article from @zinyando that makes sense for my use case, but I’m curious if I can have built-in memory to work for this use case.

1 Like
2

I am also interested in this use case

3

This is a bit difficult to achieve with just built in memory, you need a combination of short-term memory (conversation history) and long term memories to have a proper chatbot that understand the context of the conversation.

This this thread is a bit old, did you manage to achieve what you wanted?

1 Like
4

@zinyando frankly no. I had multiple attempts, but memory is still not working as I would expect.

I just want to have some clarity about memory in CrewAI. What are the intended use cases if not the context enrichment from previous requests.

5

I’m not sure what the level of solution you work on, but there are a couple of techs for mids and enterprise grade providing long memory in specific cases - one works like a cognitive memory within the chat, another like business intent memory in specific product context.

6

@Serge_B yeah, I get it. like mem0 or smth.
I can use those.

But I’m still interested what is the goal of memory feature in CrewAI. Is it not working as intended or I’m trying to misuse it?