Integration with Existing Vector Databases

[XXNS]

LightAgent currently supports memory capabilities through custom implementations like mem0, which is excellent. However, many enterprise users already have existing vector database setups (Pinecone, Weaviate, Milvus, Chroma, etc.) for their knowledge bases and would benefit from direct integration options.

Proposed Solution

I'd like to request support for connecting LightAgent to existing vector databases through a standardized interface. This would allow users to:

1. Leverage their existing knowledge bases without migrating data
2. Use their preferred vector database technology
3. Maintain consistent memory across different agent frameworks

Implementation Suggestions

• Create an abstract VectorStore interface class that different database connectors can implement
• Include basic connectors for popular vector databases (similar to how mem0 offers Qdrant support)
• Allow passing custom vector DB instances to the agent during initialization
• Document best practices for vector dimensions and embedding models compatibility

Example Usage (Concept)

from LightAgent import LightAgent
from my_vector_db_connector import MyVectorDB

# Initialize custom vector DB connection
vector_db = MyVectorDB(connection_params)

# Pass to agent
agent = LightAgent(
    model="deepseek-chat",
    api_key="your_api_key",
    base_url="your_base_url",
    vector_store=vector_db
)

Benefits

This enhancement would make LightAgent more accessible to enterprise users with existing AI infrastructure and create a more flexible integration path for organizations with specific database requirements or compliance needs.

Additional Context

I've been using LightAgent in several projects and find it incredibly well-designed. This addition would help our team standardize on LightAgent across more use cases.

[wxai-space]

If you already have a RAG (Retrieval-Augmented Generation) system in place, you can seamlessly integrate your existing vector database (such as Pinecone, Weaviate, Milvus, etc.) into LightAgent by creating a custom tool. This tool will retrieve results from the vector database based on the user's query and return the results to LightAgent for summarization and response. Below is an example implementation:

def search_data(
        query: str,
        max_results: int = 5
) -> str:
    """
    Search the vector database based on the provided query and return the retrieved results.
    
    :param query: The search keyword or phrase.
    :param max_results: The maximum number of results to return (default is 5).
    :return: Formatted search results.
    """
    # Assume this calls the vector database's search API
    results = vector_db.search(query, max_results)  # Replace with actual vector database retrieval logic
    formatted_results = "\n".join([f"- {result}" for result in results])  # Format the results
    return f"From searching '{query}', I found the following relevant information:\n{formatted_results}"

# Define tool metadata within the function
search_data.tool_info = {
    "tool_name": "search_data",  # Tool name
    "tool_name": "Vector Knowledge Base Search",  # Tool name in Chinese
    "tool_description": "Search the vector database based on keywords and return relevant results.",  # Tool description
    "tool_params": [  # Tool parameter definitions
        {"name": "query", "description": "The search keyword or phrase", "type": "string", "required": True},
        {"name": "max_results", "description": "The maximum number of results to return", "type": "int", "required": False},
    ]
}

Key Points:

1. Tool Functionality:

   • The search_data tool retrieves data from the vector database based on the user's query and returns formatted results.
   • It supports customizing the number of results returned (max_results), with a default value of 5.

2. Tool Metadata:

   • tool_info defines the tool's name, description, and parameters, ensuring LightAgent can correctly identify and use the tool.
   • The query parameter is required and specifies the search keyword, while max_results is optional and controls the number of results returned.

3. Integration:

   • Add this tool to LightAgent's tools list to make it available for use in conversations.
   • LightAgent will automatically invoke this tool based on the user's query, retrieve data from the vector database, and generate a summarized response.

[wxai-space]

Take a look at how tools are generated
https://github.com/wxai-space/LightAgent?tab=readme-ov-file#3-tool-generator