Mem0

A guide to integrating Mem0 with Keywords AI.

Mem0 is a self-improving memory layer for LLM applications, enabling personalized AI experiences that save costs and delight users.

Mem0 + Keywords AI provides a powerful combination for building AI applications that can remember user interactions over time and get complete LLM observability.

Check out the Mem0 documentation for more information.

Quickstart

Prerequisites

• A Keywords AI API key
• A Mem0 API key

Add memory with Mem0 SDK

Install the Mem0 Python client

pip install mem0

Add Keywords AI gateway to the Mem0 client as an LLM provider. You can go to the Models page to find the available models.

Python

from mem0 import Memory
import os

api_key = os.getenv("MEM0_API_KEY")
keywordsai_api_key = os.getenv("KEYWORDSAI_API_KEY")
base_url = os.getenv("KEYWORDSAI_BASE_URL") # "https://api.keywordsai.co/api/"

config = {
    "llm": {
        "provider": "openai",
        "config": {
            "model": "gpt-4o-2024-08-06",
            "temperature": 0.0,
            "api_key": keywordsai_api_key,
            "openai_base_url": base_url,
        },
    }
}

m = Memory.from_config(config_dict=config)

result = m.add(
    "I like to take long walks on weekends.",
    user_id="alice",
    metadata={"category": "hobbies"},
)

print(result)

Add memory with OpenAI SDK

Currently we only support adding memory to your AI products using the OpenAI SDK integration.

Once you integrate Keywords AI gateway with the OpenAI SDK, you can add memory to your AI product by following the code example below.

OpenAI Python SDK

client = OpenAI(
    api_key=os.environ.get("KEYWORDSAI_API_KEY"),
    base_url=os.environ.get("KEYWORDSAI_BASE_URL"),
)

def plain_json_version():
    # Using plain JSON with the same structure
    messages = [
        {"role": "system", "content": "You are a helpful assistant."},
        {
            "role": "user",
            "content": "Context: I like eating carrots, and I like to play basketball.",
        },
    ]
    response = client.chat.completions.create(
        model="openai/gpt-4o",
        messages=messages,
        extra_body={
            "mem0_params": {
                "user_id": "user_1",
                "org_id": "org_1",
                "api_key": os.environ.get("MEM0_API_KEY"),
                "add_memories": {
                    "messages": messages,
                },
            }
        },
    )

In this case, the messages will be added to the memory and the query will be used to search the memory.

Add memory through Keywords AI SDK

We recommend using the Keywords AI SDK for better type checking and autocomplete.

Install the Keywords AI SDK

pip install keywordsai-sdk

Keywords AI Python SDK

from keywordsai_sdk.keywordsai_types.services_types.mem0_types import (
    Mem0Params,
    AddMemoriesParams,
    SearchMemoriesParams,
)
from openai import OpenAI


client = OpenAI(
    api_key=os.environ.get("KEYWORDSAI_API_KEY"),
    base_url=os.environ.get("KEYWORDSAI_BASE_URL"),
)


def test_mem0_generation():
    try:
        # Create a request with linkup_params that will fail
        mem0_memory = [
            {"role": "system", "content": "You are a helpful assistant."},
            {
                "role": "user",
                "content": "Context: I like eating carrots, and I like to play basketball.",
            }
        ]
        response = client.chat.completions.create(
            model="openai/gpt-3.5-turbo",
            messages=mem0_memory,
            extra_body={
                "mem0_params": Mem0Params(
                    user_id="user_1",
                    org_id="org_1",
                    api_key=os.environ.get("MEM0_API_KEY"),
                    add_memories=AddMemoriesParams(
                        messages=mem0_memory,
                    ),
                ).model_dump(exclude_none=True),
            },
        )
        print(json.dumps(response.model_dump(), indent=4), "response")
        assert response.choices[0].message.content is not None

    except Exception as e:
        assert False

Search memories

Once you pass the params like mem0_search_memories_response and mem0_add_memories_response to your prompt, you can view the responses in the side panel of Logs page.

You can search memories by using the search_memories parameter in the Mem0Params object.

Python

response = client.chat.completions.create(
            model="openai/gpt-3.5-turbo",
            messages=[
                {"role": "system", "content": "You are a helpful assistant."},
                {
                    "role": "user",
                    "content": "Based on the {{mem0_search_memories_response}}, what is the user's favorite food?",
                },
            ],
            extra_body={
                "mem0_params": Mem0Params(
                    user_id="user_1",
                    org_id="org_1",
                    api_key=os.environ.get("MEM0_API_KEY"),
                    add_memories=AddMemoriesParams(
                        messages=mem0_memory,
                    ),
                ).model_dump(exclude_none=True),
            },
        )