Single agent

Retrieval (RAG)

Pull in the right documents before answering.

1 agentCore

How it works

1Index your sources as searchable chunks.
2On a query, retrieve the most relevant chunks.
3Put them in the prompt as grounding context.
4Answer only from that context, cite sources, and say "not found" when absent.

Use it when

Answers must be grounded in your own, private, or current knowledge the model wasn't trained on — docs, policies, tickets.

Reach for something else when

The model already knows the answer reliably, or your corpus is small enough to paste directly.

Where you stay in the loop

You curate the sources and the grounding rules; the agent answers only from them. Verify citations and watch for confident answers the sources don't actually support.

In the wild

A policy bot retrieves the 3 most relevant handbook sections and answers strictly from them, linking each.

Hand this to your agent

You answer only from provided context (RAG style).

For each question I'll paste relevant source snippets labeled [1], [2], etc.
Rules:
  • Answer using only those snippets.
  • Cite the snippet number after each claim.
  • If the snippets don't contain the answer, say "Not in the provided
    sources" — never guess.

Question + sources:
<...>

Replace the <…> placeholders, paste it into your agent, and it'll scaffold the workflow with you.