Retrieve overview

note

Knowledge Hub is an Early Access release. Refer to the Early Access disclaimer for details.

Retrieve is the capability in Knowledge Hub that serves content to AI agents and applications. When an agent queries a Knowledge Base, Knowledge Hub searches the vector index, ranks the matching content, and returns the most relevant chunks for the agent to use.

How retrieval works

When an agent submits a retrieval query:

1. Knowledge Hub converts the query text into a vector using the same embedding model the Knowledge Base uses.
2. Knowledge Hub searches the vector index for chunks whose embeddings are closest to the query vector, using the configured retrieval mode (Semantic, Lexical, or Hybrid).
3. Knowledge Hub ranks the results and returns the top-N chunks, where N is the Top-K setting on the Knowledge Base.
4. The agent receives the ranked chunks and uses them to ground its response.

What the agent receives

Each chunk that Knowledge Hub returns includes:

• The text content of the chunk
• Metadata fields mapped from the source data (for example, document title, author, source URL, record ID)
• A relevance score indicating how closely the chunk matched the query

The agent uses this content to formulate a grounded response and, when configured, cites the source metadata in its output.

Retrieval modes

The retrieval mode controls how Knowledge Hub searches the index:

• Semantic: converts both the query and the indexed chunks into vectors and finds the nearest matches by meaning. Best for natural-language questions where exact wording does not matter.
• Lexical: matches on exact keywords in the query. Best for technical queries, product names, codes, and identifiers where precision matters.
• Hybrid: combines Semantic and Lexical results. Recommended for most use cases.

Refer to Retrieval modes for detail on each mode.

Knowledge Base configuration for retrieval

Retrieval behavior is configured at the Knowledge Base level:

• Retrieval mode: Set the default mode (Semantic, Lexical, or Hybrid). Agents can override this per query.
• Top-K: Set the maximum number of chunks Knowledge Hub returns per query. Agents can override this per query.
• Embedding model: The model that embeds chunks during ingestion and queries at retrieval time. The embedding model must be the same for both operations.

Connecting an agent to Knowledge Hub

Knowledge Hub does not generate responses. It returns source content, and the agent is responsible for synthesizing the retrieved chunks into a response. To connect an Boomi Agentstudio agent to a Knowledge Base, refer to Connecting Knowledge Bases in the Boomi Agentstudio documentation.