Prompts

Prompts are templates used by the RAG endpoint to turn retrieved chunks into an LLM answer. Each prompt has a {{context}} placeholder where the retrieved chunks are inserted, and a {{question}} placeholder for the user's query.

RagPack ships with three built-in system prompts. You can also create your own via the admin UI.

Built-in prompts

Slug	Name	Description
basic_rag	Basic RAG	Answers naturally using only the provided context. Does not cite sources.
rag_with_citations	RAG with Citations	Answers with inline source citations (e.g. "According to [Source 1]...").
concise_rag	Concise Answer	Answers in 2–3 sentences.

basic_rag is the default when no promptSlug is passed to collection.rag().

List prompts

const prompts = await client.prompts.list();

for (const p of prompts) {
  console.log(p.slug, p.name, p.is_system);
}

System prompts (is_system: true) are listed first, followed by any custom prompts you've created.

Get a prompt

const prompt = await client.prompts.get("basic_rag");
console.log(prompt.content);

Using a prompt in RAG

Pass the prompt's slug to collection.rag():

const { answer } = await collection.rag({
  query: "What is the refund policy?",
  promptSlug: "rag_with_citations",
  model: "gpt-4o",
});

Custom prompts

Custom prompts can be created and managed in the admin UI. Use {{context}} and {{question}} as placeholders in your template:

Answer the question below in a formal tone using only the provided context.

<context>
{{context}}
</context>

Question: {{question}}

Once created, use the prompt's slug in collection.rag() the same way as a built-in prompt.