Originally published at lumbox.co
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.
