Extend the current RAG implementation

[DavdGao]

• [ ] extend the vdb store
  • [x] support milvus by providing a MilvusStore class #825
  • [ ] support MongoDB by providing a MongoDBStore class #860
  • [ ] consider supporting more complex vector database, e.g. graph vector database
• [ ] extend the reader class
  • [ ] provide Microsoft Word reader #852
  • [ ] provide Microsoft Excel reader #872
  • [ ] provide Microsoft Powerpoint reader
  • [ ] provide advanced PDF reader
• [ ] consider to improve the current implementation of SimpleKnowledge class

[DavdGao]

Contributions are welcome! Please comment here first to avoid duplicate work.

[xucailiang]

• • [ ] extend the vdb store

    • [x] support milvus by providing a MilvusStore class Add support for milvus lite vector database in the RAG module of AgentScope #825* [ ] support MongoDB by providing a MongoDBStore class Implement mongodb store #860* [ ] consider supporting more complex vector database, e.g. graph vector database* [ ] extend the reader class

    • [ ] provide Microsoft Word reader Implement word reader #852* [ ] provide Microsoft Excel reader Implement Excel Reader #872* [ ] provide Microsoft Powerpoint reader* [ ] provide advanced PDF reader[ ] consider to improve the current implementation of SimpleKnowledge class

I am very interested in integrating Graph Vector Databases. I would like to learn more about this and understand the positioning of graph databases in this context. I also want to know if the extraction and construction of nodes and relationships will involve LLM calls. There are similar open-source projects like GraphRAG; will we be adopting a similar approach?

[Tendo33]

I'm looking into using a third-party RAG system, LightRAG, as migrating our existing knowledge base to AgentScope's built-in RAG is not currently feasible.

What is the recommended way to integrate LightRAG's features into the react_agent?

Does this require manually adding tools?

For example, would a query flow involve: embedding_tool → lightrag_retrieval_tool before the agent generates the final answer?

Could you please provide an example or guidance on using AgentScope with external RAG solutions?

[M4n5ter]

I am very much looking forward to seeing support for graph databases like Neo4j and PostgreSQL + Apache AGE!

[qy-liuhuo]

I am willing to try to implement Microsoft PowerPoint reader.