LlamaIndex + Lumbox: Treat Your Inbox as a Queryable Knowledge Source

LlamaIndex made RAG something you can ship in an afternoon. Load docs into an index, attach an LLM, query. The gap most teams hit isn't the framework — it's finding a good data source. Your agent's own Lumbox inbox is a better source than most.

Why the inbox is good RAG fodder

• It's high-signal, low-noise (it's already filtered by being sent to you).
• It's structured — sender, date, subject, thread ID are all fields.
• It grows organically as the agent does its work.

Wiring it up

from llama_index.core import VectorStoreIndex, Document
from lumbox import Lumbox

lumbox = Lumbox(api_key=os.environ["LUMBOX_API_KEY"])

def load_inbox_as_documents(inbox_id: str) -> list[Document]:
    messages = lumbox.inboxes.list_messages(inbox_id=inbox_id, limit=500)
    return [
        Document(
            text=m.text,
            metadata={
                "from": m.from_address,
                "subject": m.subject,
                "date": m.received_at,
                "thread_id": m.thread_id,
            },
        )
        for m in messages
    ]

docs = load_inbox_as_documents(my_inbox_id)
index = VectorStoreIndex.from_documents(docs)
query_engine = index.as_query_engine()
print(query_engine.query("What did Acme say about the Q3 timeline?"))

Keep it fresh with webhooks

Subscribe to Lumbox's email.received webhook and incrementally upsert new messages into the index. No re-indexing, no polling. lumbox.co.