Yes—Zilliz Cloud's real-time upsert API ingests new documents within seconds, Scout queries immediately without retraining or model updates.
Scenario: ingest 1000 new documents/day (news, research, emails). Zilliz Cloud batch-inserts and increments indices within seconds. Scout queries immediately against fresh data—no waiting for retraining. The open-weights approach means: no model updates needed for new knowledge, just embed and insert. This is critical for time-sensitive RAG (news analysis, security incidents) where staleness causes missed insights.
For scalable ingestion with Zilliz Cloud: (1) use upsert API to add/update documents, (2) embed with a fast model (BGE-small for speed, BGE-large for quality), (3) batch embeddings (1000 at a time) to amortize compute, (4) Zilliz Cloud auto-reindexes incrementally. Scout's inference is stateless—each query is independent. Monitor embedding freshness: if docs arrive hourly but embeddings compute daily, Scout answers stale questions. Use webhooks or CDC to embed immediately upon arrival. Zilliz Cloud's multi-tenancy means you scale without managing infrastructure.
Related Resources
- Zilliz Cloud — Managed Vector Database — real-time data ingestion
- Retrieval-Augmented Generation (RAG) — freshness in RAG pipelines
- Getting Started with LlamaIndex — real-time integration patterns