Join the redcache-ai discord server | Email me

RedCache-ai: A memory framework for Large Language Models and Agents

What ? 🤖

REmote Dictionary cache - Artificial Intelligence (RedCache-AI). While developing a chat application, every solution I came across was either expensive, closed source or lacked extensive support for external dependencies. Redcache-ai provides a dynamic memory framework for Large Language Models, enabling developers to build wide ranging applications, from AI-powered dating apps to healthcare diagnostics platforms.

Quickstart 😄

Installation

Install redcache-ai as a python package

pip install redcache-ai

Initialize Redcache-ai
Redcache-ai provides two options. Initialize to disk or sqlite. By default redcache-ai initializes to disk. This where memories are stored.

Option 1: initialize to disk
Make sure to import all the required dependencies

from redcache_ai import RedCache, load_config, set_openai_api_key

from redcache_ai.storage import DiskStorage, SQLiteStorage


storage = Disktorage()

Option 2: Initialize to sqlite

storage = SQLiteStorage(db_path='my_cache.db')

Store text as memory

Note: the output visual representation below is for disk storage only. The input is the same for sqlite

Input

memory_cache = storage.add("England is nice country", user_id = "James", metadata={"category": "facts"})

Output

{"James": {"7bbfbcbf-da9e-44ca-9cbb-ab558c64b36a": {"id": "7bbfbcbf-da9e-44ca-9cbb-ab558c64b36a", "text": "\"England is a nice country\"", "metadata": {"data": "\"England is a nice country\"", "category": "facts"}, "vector": [0.4472135954999579]}}}

Retrieve Memory

# Get all memories
memories = storage(user_id = "James")
print(all_memories)

Output:

# Get all memories
[
    {
        "id": "7bbfbcbf-da9e-44ca-9cbb-ab558c64b36a",
        "text": "England is a nice country",
        "metadata": {
            "category": "facts"
        }
    }
]

Search Memories

# Get all memories
results = storage.search("country", user_id="James", num_results=1)

Output:

[
    {
        "id": "7bbfbcbf-da9e-44ca-9cbb-ab558c64b36a",
        "text": "England is a nice country",
        "metadata": {
            "category": "facts"
        },
        "score": 0.849
    }
]

Update Memory

updated_memory = storage.update("7bbfbcbf-da9e-44ca-9cbb-ab558c64b36a", "England is a beautiful country", user_id="James") 

Output:

{
    "id": "7bbfbcbf-da9e-44ca-9cbb-ab558c64b36a",
    "text": "England is a beautiful country",
    "metadata": {
        "category": "facts"
    }
}

Delete Memory

storage.delete("7bbfbcbf-da9e-44ca-9cbb-ab558c64b36a", user_id="James")

Output:

# No output, memory is deleted if successful

Delete all memories

storage.delete_all(user_id="James")

# No output, all memories for the user are deleted if successful 

Enhance a memory using LLMs 🔥

Large Langugae Model Integradtion
For now, redcache-ai only supports OpenAI. To integrate redcache-ai memories into openAI, you have to set an OPENAI API Key.

Once you have the key, you'll integrate your text memories like this:

from redcache_ai.config import set_openai_api_key

set_openai_api_key("your-openai-api-key-here") 

from redcache_ai import RedCache, load_config

# Load the default configuration
config = load_config()

# Initialize RedCache with OpenAI LLM
redcache = RedCache.from_config(config)

The output for load_config is below:

{
    "llm": {
        "provider": "openai",
        "config": {
            "model": "gpt-4",
            "temperature": 0.2,
            "max_tokens": 1500,
        }
    }
}

enhanced_memory = redcache.enhance_memory("England has a rich history", user_id="James", category="facts")
summary = redcache.generate_summary(user_id="James")

These examples demonstrate the basic usage of redcache-ai. For more detailed information and advanced features, please refer to the full documentation. If implementation details are unclear please see the test_redcache.py file in examples. For more information, visit the discord channel.

What's Next ✈️
Redcache-ai is still early. Priority is to fix any and all existing bugs. Improve documentation then:
1.Add integration with more LLM providers including Llama, Mixtral, Claude etc.
2.Add functionality for AI Agents.
3.Provide support for a hosted version.
Happy memory building. ❤️