Knowledge Bases
Knowledge Hub is an Early Access release. Refer to the Early Access disclaimer for details.
A Knowledge Base (KB) is the core indexing unit in Knowledge Hub. It holds the vector embeddings generated from one or more data sources, and it serves as the retrieval target for AI agents and applications.
What a Knowledge Base does
A Knowledge Base:
- Receives chunked and embedded content from Sources
- Stores content as a single unified vector index regardless of how many Sources contribute to it
- Applies a configured retrieval mode (Semantic, Lexical, or Hybrid) when answering queries
- Exposes a retrieval endpoint that AI agents use to search content
From an AI builder's perspective, a Knowledge Base is the "library" you point your agent at. From a data engineer's perspective, it is the destination you configure in Data Integration.
Provide a clear, meaningful description when you create a Knowledge Base. AI agents use it to identify and select the correct Knowledge Base for a task.
Single-index model
All Sources connected to a Knowledge Base write into a single shared index. When an agent queries the Knowledge Base, it searches across content from all Sources simultaneously. There is no per-Source query scope within a Knowledge Base.
If you need to keep Source data queryable in isolation, use separate Knowledge Bases.
Embedding model
Each Knowledge Base has one embedding model. The model converts text chunks into vector embeddings during ingestion and converts queries into vectors during retrieval. Both must use the same model to produce comparable vectors.
Knowledge Hub uses a single, Boomi-managed embedding model at Early Access; you do not choose between models. The model is set automatically when you create the Knowledge Base and is immutable after the first ingestion run. Refer to Configuring embedding settings.
Retrieval configuration
Each Knowledge Base has a default retrieval mode (Semantic, Lexical, or Hybrid) and a Top-K setting. These defaults apply to all queries unless overridden at query time by the calling agent.
Refer to Configuring retrieval settings and Retrieval modes.
Knowledge Base states
| State | Meaning |
|---|---|
| Draft | The Knowledge Base has been created but not yet published. It cannot serve retrieval queries. Sources can be added and ingestion can run. |
| Enabled | The Knowledge Base is published and active. Retrieval queries are served. |
Disable and Archive is currently not supported.
To transition a Knowledge Base from Draft to Enabled, you must publish it. Publishing requires at least one Source in Enabled state. Refer to Publishing a Knowledge Base.
Relationship to repositories and Sources
A Knowledge Base belongs to exactly one repository. It can have many Sources. Each Source feeds its content into the Knowledge Base's shared index.
Repository
└── Knowledge Base
├── Source A (S3 bucket)
├── Source B (Confluence space)
└── Source C (Salesforce objects)
Next steps
- Refer to Creating a Knowledge Base to create your first Knowledge Base.
- Refer to Configuring embedding settings and Configuring retrieval settings to tune how it indexes and retrieves content.
- Refer to Publishing a Knowledge Base to make it available to agents.
- Refer to Managing Knowledge Base lifecycle to disable or delete a Knowledge Base.
- Refer to Sources to learn about the data connections a Knowledge Base contains.
