Skip to content

Multi-Collection Routing

When your data lives in multiple indexes, retrievalagent can route each query to the right subset automatically. Before retrieval, an LLM routing step selects which collections are relevant — only those are searched.

Setup

from retrievalagent import init_agent

rag = init_agent(
    collections=["products", "faq", "policies"],
    backend="qdrant",
    backend_url="http://localhost:6333",
    model="openai:gpt-5.4",
)

from retrievalagent import init_agent

rag = init_agent(
    collections={
        "products": "Product catalog: SKUs, prices, specs, availability",
        "faq":      "Customer-facing FAQ, troubleshooting, return policy",
        "policies": "Internal HR and compliance policy documents",
    },
    backend="qdrant",
    backend_url="http://localhost:6333",
    model="openai:gpt-5.4",
)

from retrievalagent import Agent
from retrievalagent.backend import MeilisearchBackend

backends = {
    "products": MeilisearchBackend("products"),
    "manuals":  MeilisearchBackend("manuals"),
}
rag = Agent(
    index="catalog",
    collections=backends,
    collection_descriptions={
        "products": "Product listings with prices and specs",
        "manuals":  "Installation and user guides",
    },
)

How it works

  1. The query arrives at invoke / chat.
  2. An LLM call selects the relevant collection names (using names + optional descriptions).
  3. Only the selected backends are searched — the context variable _ACTIVE_COLLECTIONS scopes retrieval.
  4. Each retrieved document gets a metadata["_collection"] tag with its source collection.

If the LLM returns an empty selection (uncertain), retrievalagent falls back to searching all collections.

Collection metadata

state = rag.invoke("What's our return policy?")
for doc in state.documents:
    print(doc.metadata["_collection"], doc.page_content[:80])

Accessing the collection map

print(rag.collections.keys())   # dict[str, SearchBackend]