RAG and Search - Onyx Documentation

Onyx answers questions from your connected sources with retrieval-augmented generation (RAG). When a user asks a question, Onyx searches the OpenSearch document index, selects the most relevant documents with a model, and streams an answer with numbered citations. Each user sees only documents they are allowed to see.

Settings that shape search

Setting	Where	What it does
Connectors	Admin Panel → Connectors	Decide which sources Onyx indexes. See Connectors.
Document Sets	Admin Panel → Document Sets	Group connectors into filters for users and Agents. See Document Sets.
Index Settings	Admin Panel → Index Settings	Choose the embedding model and options such as Contextual Retrieval, Multipass Indexing, Extract & Caption Images, and Vector Quantization. Not shown on Onyx Cloud, where Onyx manages the index. See Search Configs.
Search Selection Model	Admin Panel → Chat Preferences	The model that selects and checks retrieved documents.
Search Mode	Admin Panel → Chat Preferences	A quick document search view. Needs a Business or Enterprise plan, or Onyx Cloud.
Auto-Detect Search Filters	Admin Panel → Chat Preferences	Applies source and time filters that Onyx infers from the query.

See Chat Preferences for the chat-level settings.

Search types

Internal Search searches documents that connectors indexed.
Federated search queries a source live at question time, for example federated Slack, and merges the results with indexed documents.
Web Search searches the public web through the provider you configure under Admin Panel → Web Search. See Web Search.

Users can narrow a search by source, Document Set, and date range in chat. Users can explore what Onyx indexed with Internal Search.