Definition: Vector index
A vector index is a KERNSTACK search index with index_type "vector", declared over the tables and buckets that hold the source records. Retrieval runs inside the project, fed by embeddings from catalogue models such as text-embed-v1 (3 credits per 1,000 tokens) or a custom inference endpoint.
An index, not another database
A standalone vector database holds a copy of your data that a pipeline must keep honest. A declared vector index derives from the source itself: list the table or bucket, name the fields, and the platform maintains the index as records change. Deleting the record deletes the passage; there is no orphaned vector at a second vendor to explain during an audit.
Where embeddings come from
Catalogue models cover the common path — text-embed-v1 for embeddings at 3 credits per 1,000 tokens, sentiment-multilingual-v1 at 2 — and a custom inference_endpoint can supply a model you already run. Text and vector indexes can be declared over the same sources for hybrid queries that need literal terms and meaning at once.
Frequently asked questions
No. The index is declared over tables and buckets inside the project, in the project’s European zone. Your records never leave to be embedded and copied elsewhere.
With text-embed-v1, 3 credits per 1,000 tokens — embedding 10 million tokens costs 30,000 credits. Re-embedding after a model change is an index operation over the same sources, not a data migration.
At retrieval-as-the-product scale: extreme vector counts or specialised latency budgets. Product corpora — catalogues, policies, support archives — are the declared index’s home ground, and the source data staying home keeps a later move cheap.
Related capabilities, architectures, and comparisons
Ship your stack on the Software Lifecycle Cloud.
Closed preview on dedicated European infrastructure. Bring a repository or start from a blueprint.