Use it in Workflows & Agents

A knowledge base is only useful once something queries it. ByteChef exposes it through the Knowledge Base component, which can be used directly as a workflow action or attached to an AI agent as its retrieval layer.

Search from a workflow step

Add the Knowledge Base Search Data action to a workflow to fetch the most relevant chunks for a query. The action takes:

• Knowledge base — which knowledge base to query.
• Query — the natural-language text to match against. Leave it empty for a filter-only search.
• Tags — restrict the search to documents carrying one of the given tags.
• Top K — how many chunks to return.
• Similarity threshold — a floor on the relevance score (0.0–1.0); lower-scoring chunks are dropped.
• Metadata filter — an optional list of metadata key-value pairs applied before ranking. Entries within a group are ANDed; groups are ORed.

It returns an array of matching chunks, each with its content, relevance score, and metadata — ready to feed into a prompt, a condition, or any downstream step.

Ground an AI agent (RAG)

Attach a knowledge base to an AI agent so the agent can retrieve context on its own:

• As the agent's vector store — the agent searches the knowledge base automatically when answering, grounding its responses in your documents.
• As a tool — the agent decides when to call the knowledge base, issuing a search only when it needs supporting facts.

Either way the pattern is the same:

User question → embed → nearest chunks from the knowledge base → chunks added to the prompt → model answers

This is what keeps an agent's answers grounded in your content instead of the model's training data alone.

Filtering by tag

When a single knowledge base holds distinct kinds of content, pass a tag filter so a given workflow or agent only retrieves from the documents you intend. Tag matching is inclusive — any document carrying one of the requested tags is eligible — so an agent scoped to policy never surfaces a chunk from an untagged product spec in the same knowledge base.

Feeding a knowledge base from a workflow

The Knowledge Base component also has a write side used by sources: a workflow can upsert records into a knowledge base as documents, which are then chunked and embedded like any uploaded file. This is how external systems keep a knowledge base current without manual uploads.