mem0 with bedrock

[thandavms]

I am trying to make mem0 work with bedrock embedding model. first thing. it demands a openai key ... even though i provide the config to be amazon bedrock model

The code look like this....

from mem0 import Memory

config = {
"embedder": {
"provider": "aws_bedrock",
"config": {
"model": "amazon.titan-embed-text-v1"
}
}
}

m = Memory.from_config(config)

[johnwlockwood]

The default llm is openai and by default that requires the OPENAI_API_KEY environment variable.

[thandavms]

John - thanks. but this should not force me to give an api key, if in my config i am asking it to use another embedding model. Would be great to have this mandate removed...

[johnwlockwood]

In addition to the embedding model, mem0 requires an llm, which is needed to infer facts from messages and then determine what memory update actions are needed. The default provider is openai with its "gpt-4o-mini" model, which requires an openai api key.

You might be able to avoid using a valid api key if you use Memory.add(..., infer=False), then it will skip using the llm and add the messages as literal memories.

Your other option is to configure the llm to another provider of your choice, such as a ollama with a local model.