RAG Query

Answer a question from your own documents, not the model's training.

RAG Query asks a question against a private knowledge base — a RAG Container — and generates an answer grounded in what it finds there. It retrieves the relevant material first, then answers from it.

A general LLM knows what it was trained on. This is how an agent answers from your credit policy instead.

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: answering advisors' policy questions

Financial advisors need quick answers about internal lending policy. Rather than searching long technical documents, they ask the agent: "What is our policy on SBLOC loans for new businesses?"

The problem. The policy documents are far too long to put in a single prompt, and a keyword search doesn't understand what the advisor is actually asking.

The approach. Send the question to a RAG Container pre-loaded with the policy documents. The skill finds the relevant sections and uses them to generate a policy-based answer.

1. Choose the knowledge base. Select the container holding your policy documents from the RAG Container ID dropdown.

2. Write the instruction. Instruction to LLM sets the role:

You are a helpful and knowledgeable financial policy assistant. Provide clear and concise answers based on the provided context.

3. Set the query. Point Query at the advisor's question — $input.userQuestion.

4. Filter (optional). Filters takes a stringified JSON object that narrows the search before retrieval: {"loan_type": "SBLOC"} searches only documents tagged for that loan type. Narrowing the haystack usually beats improving the needle.

Advanced configuration

Setting	What it does
Response Format	Text, or a structured JSON object for use by later skills.
Temperature	Controls creativity. Lower values (around 0.2) are better for factual, grounded answers — which is the entire point of RAG.
Num of Conversation Turns	Lets the model consider previous messages for conversational context.

Output

• content — the generated answer, as text or JSON.

• statusCode — 200 success · 400 bad request · 500 internal error.

---

For how retrieval works and why, see Knowledge & Context. To configure ingestion, see RAG Knowledge Pipeline. To add this skill to an agent, see Adding a Skill to the Agent.