Creating and Loading Multi-Doc Agents in the Cloud

You are already using an object index to "retrieve" agents right?

You should be able to use azure as a backing for that

If you really wanted to boil it down though, an agent is just

• the system prompt
• the tools it has access to

You could store that information as nodes in any vector index. And then on the fly, retireve, and construct those agents. If those agents have query engine tools, those can also be backed by azure vector store, and are basically a no-op to re-create on the fly

Btw, super curious what your overall goal/use-case is. We are actively trying to document and publicize agentic use cases with case studies where possible

Thanks for the feedback, yeah I have a RAG system that ingests municipal building and zoning code data and I have scope for 3600+ municipalities US wide.
I have it working but it has to load all the Agents (per doc.. each muni code has a few hundred docs etc.) and top Agent every time I run the fastAPI backend. These are assigned to each municipality as an array of agents. Municipality is selected when a property address is searched.
I am going to test out Azure CosmoDB as the vector store.