[Feature Request]: Customizable Component Selection within DocAgent's RAG Pipeline

[mingjun1120]

Is your feature request related to a problem? Please describe.

Currently, DocAgent's RAG (Retrieval-Augmented Generation) system appears to have a fixed set of tools and services. While this structure ensures reliability and consistency, it may limit flexibility when addressing unique or evolving user requirements. Expanding the range of tools or incorporating adaptive functionalities could enhance its versatility and better cater to diverse use cases.

Describe the solution you'd like

Currently, DocAgent's RAG (Retrieval-Augmented Generation) system appears to have a fixed set of tools and services. To enhance flexibility and adaptability for developers and users, it would be highly beneficial to allow customization of the components within the RAG pipeline. This would enable users to select their preferred tools for each stage of the process. Companies often have existing infrastructure and preferred cloud providers or technology stacks, and this customization would allow seamless integration with those systems.

Examples of customizable components include:

• Data Parsing Tools: Options such as Docling, LlamaParse, or Markitdown.
• Vector Databases: Choices like Azure AI Search, Qdrant, Weaviate, or Postgres.
• Large Language Models (LLMs): Integration with platforms like Azure OpenAI, Amazon Bedrock, or Google Vertex AI.
• Embedding Models: Support for any embedding model available on the marke.

Additional context

No response

[marklysze]

Thanks so much for detailing improvements to DocAgent that you'd like to see, @mingjun1120, they all look great. Bringing in @AgentGenie who is planning development for the DocAgent.