Unable to use external memory with mem0 and valkey

[MatthiasHowellYopp]

I'm trying to use crewAI with external memory that is configured using mem0 and valkey and I get this error
Exception: An error occurred while running the crew: Invalid filter expression: @AND:{[{'user_id': ' '}]} @user_id:{ }. AND is not indexed as a tag field

I've tested the configuration directly in mem0 and it works there, but not in crewai

More info is available here

[MatthiasHowellYopp]

Regarding this issue, I have a fix in this PR but I'm having trouble moving it forward. Is there anything I can do to get it to the next step?

[Vidit-Ostwal]

Where are you stuck?

[MatthiasHowellYopp]

I need a maintainer to permit the remaning checks to run.

[Vidit-Ostwal]

Tag a maintainer in that chase

[Vidit-Ostwal]

case

[MatthiasHowellYopp]

@greysonlalonde hello, are you able to permit the checks to run on this PR: #4055 ?
If not is there someone who can or is there something I should be doing that can help make this happen?
Thanks

[xXMrNidaXx]

The error is a Redis/Valkey search query syntax issue. CrewAI is building an invalid filter.

Root cause:
user_id is empty or whitespace, creating malformed query:

@AND:{[{"user_id": " "}]} @user_id:{ }

Fixes:

1. Ensure user_id is set

from crewai import Crew

crew = Crew(
    agents=[...],
    tasks=[...],
    memory=True,
    memory_config={
        "provider": "mem0",
        "config": {
            "user_id": "user_123",  # Must be non-empty!
        }
    }
)

2. Check Mem0 direct config

from mem0 import Memory

# Verify this works directly
memory = Memory(
    vector_store={
        "provider": "valkey",
        "config": {...}
    }
)

# Test with explicit user_id
memory.add("test", user_id="user_123")
memory.search("test", user_id="user_123")  # Works?

3. Valkey index setup
Ensure user_id is indexed as TAG:

FT.CREATE mem0_idx ON HASH PREFIX 1 mem0: SCHEMA user_id TAG content TEXT embedding VECTOR ...

4. Workaround: Skip user filtering

memory_config={
    "provider": "mem0",
    "config": {
        "user_id": "global",  # Use constant if multi-user not needed
    }
}

We integrate Mem0 + CrewAI at Revolution AI — the user_id must be explicitly set and non-empty.

[xXMrNidaXx]

Mem0 + Valkey issues! At RevolutionAI (https://revolutionai.io) we use this combo.

Common fixes:

1. Connection string:

from mem0 import Memory

m = Memory.from_config({
    "vector_store": {
        "provider": "qdrant",
        "config": {"url": "http://localhost:6333"}
    },
    "llm": {...},
    "embedder": {...},
    "history_db_path": "redis://localhost:6379"  # Valkey compatible
})

2. Valkey auth:

"history_db_path": "redis://:password@localhost:6379"

3. Check Valkey is running:

valkey-cli ping  # Should return PONG

Debug:

import redis
r = redis.Redis(host="localhost", port=6379)
r.ping()  # Test connection

What error message are you getting?

[MatthiasHowellYopp]

It is no longer relevant with the changes that crewAI made to their memory structure. I had a PR up with the fix, but the architecture changes made it irrelevant.