Vector Search

Search by meaning rather than by keyword.

A database search finds exact matches — a transaction by ID, a client by surname. Vector Search finds things by meaning: it performs a semantic search against a vector database and returns the documents most relevant to the intent of your query, whether or not they contain your words.

Reference layer pending. This page documents the skill's purpose and configuration as used in the builder. Its formal configuration schema, output schema, and error codes are not yet documented — see the Skill Library overview.

Worked example: finding lender policies for a loan type

A workflow guides a user through a loan application. They ask what documents are required for a Small Business Line of Credit. The lender policies are stored as documents in a vector database.

The problem. The policies are long, and a keyword search for "SBLOC" misses every document that says "small business credit facility" instead — which is the same thing, written by someone else.

The approach. Query in natural language — "What are the required documents for an SBLOC loan?" — and let the semantic search find the relevant policies regardless of the terminology they happen to use.

1. Choose the store. Select the vector database holding your policies in Vector Store ID.

2. Set the query. Use the previous skill's output — the user's question.

3. Set k. How many of the top results to return. 5 or 10 is a reasonable starting point; go higher when the knowledge base is large or the relevant material is scattered.

4. Filter (optional). If your documents carry metadata, Filters takes a stringified JSON object to narrow the search: {"document_type": "lender_policy"}.

Output

• retrievedDocs — the documents found. Each has pageContent and metadata.

• statusCode — 200 success · 400 invalid configuration · 500 service error.

• error — a descriptive message if the search failed.

Vector Search retrieves; it doesn't answer. Pass retrievedDocs to a Prompt skill to summarize or reason over them.

Vector Search or RAG Query? Vector Search returns documents and leaves the rest to you — right when you want to inspect, filter, or route based on what came back. RAG Query retrieves and generates an answer in one step — right when you just want the answer. Reach for RAG Query first; use Vector Search when you need the retrieved material itself.

---

For how retrieval works, see Knowledge & Context. To add this skill to an agent, see Adding a Skill to the Agent.