Knowledge and RAG data
Knowledge holds the documents you upload so that a model can answer from them.
What you upload to Knowledge is stored by HIPS and is available in whole only to HIPS's own systems and to your workspace. No model is given a knowledge base in one piece: indexing sends it passage by passage, and a question sends only the passages your search returned.
When a document is indexed, its text is split into passages and each passage is turned into a vector on its own. When a model answers a question, it receives only the passages that your search returned, never the whole base.
The store, the chunks and the search run on the platform in the European Union (Ireland), whatever region a request is kept to (Knowledge).
- Sensitive content: use the Certified pool. Then no passage ever reaches a GPU outside a certified facility.
- Content that is not sensitive, such as public help-center articles: under Auto, the Community pool may serve it, for a lower price.
When a base is indexed, each passage goes to an embedding model as it is, so a base that holds sensitive data belongs in the Certified pool.
A base has its own requirements, a pool among them (Auto by default, or certified), and they bind its ingestion and every request that searches it (Knowledge).
Create a base for sensitive documents in the Certified pool (Knowledge is on the paid plans):
import os import requests base = requests.post( "https://ai.hips.com/v1/vector_stores", headers={"Authorization": f"Bearer {os.environ['HIPS_API_KEY']}"}, json={ "name": "Contracts", # Sensitive documents: only the Certified pool's GPUs read the passages "hips": {"embedding_model": os.environ["HIPS_MODEL"], "pool": "certified"}, }, timeout=60,).json()print(base)In the Community pool the owner of a GPU could in theory see or alter the passages it processes, and that cannot be guaranteed. Read What a Community pool provider can see.
Related sheets
- Sheet
- 03 of 21
- Detail
- E
- Updated
- 2026-11-18