Using with a large document database?

[joshmouch]

Just came across this project, and it looks promising! Sorry for the question-as-an-issue, but I'm a .NET person, so it's rough going through the python.

Is this project able to train to use a function or a document to lookup data? For example, at some point, it's determined that more information on hot dogs is needed, would it be able to hit a file or API or function or even a hot-dog-specific expert agent in order to get that additional detail without weighing down the more generic prompts with that potentially huge knowledgebase?

[ShengranHu]

Thanks for your interest in our work and your question.

What you describe sounds similar to a dynamic workflow with RAG integrated into agentic system designs. ADAS could be applicable to design such systems in your specific domains. One key insight from our development process is that it's helpful to provide the system with basic functions (like RAG) and at least one example of a design pattern you consider crucial in your domain (perhaps a dynamic branch, based on your description).

While it's possible for the system to discover everything from scratch, incorporating some domain-specific insights can encourage it to explore and utilize these helpful building blocks more efficiently.