Private embeddings
Last updated 2026-06-14
Definition
Private embeddings are how Quri turns your text into the vectors behind semantic search. By default it uses a local model, so your emails, chat summaries and recommendations never leave your environment and there is no per-word cost. You can switch to a managed embedding service later, but the private default needs no key.
How to do this in Quri
- Connect your sources so Quri writes emails, summaries and recommendations into the graph.
- Ask a meaning-based question in /app/chat — the search runs on these embeddings.
- Leave embeddings on the local default to keep everything in your environment at no per-word cost.
- Switch to a managed embedding service later for higher recall, by pointing the embedder at it.
Frequently asked
- Where does my text go when it is embedded?
- By default nowhere — Quri embeds with a local model, so your records stay in your environment and there is no per-word cost. A managed provider is opt-in.
- Do I need an API key for embeddings?
- No. The local default needs no key. You only add one if you opt into a managed embedding provider for higher search recall.