> ## Documentation Index
> Fetch the complete documentation index at: https://docs.autosage.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Prompts

> Shape the assistant's voice and rules with a persona, instructions, and custom prompts.

Every knowledge base ships with capable, built-in answering behavior. Prompt settings let you shape that behavior — giving the assistant a voice, layering on rules, or taking complete control for a fully branded experience.

## The prompt fields

<AccordionGroup>
  <Accordion title="Persona" icon="masks-theater">
    Who the assistant is — its role, tone, and personality. For example, "a friendly support specialist for Acme's billing team."
  </Accordion>

  <Accordion title="Instructions" icon="list-check">
    Rules and guidance layered on top of the persona — what to always do, what to avoid, how to format answers, which topics to stay within.
  </Accordion>

  <Accordion title="Custom prompt" icon="wand-magic-sparkles">
    A complete replacement for the built-in answering instructions, for white-label and fully branded assistants that need total control of behavior.
  </Accordion>
</AccordionGroup>

## How the fields combine

This is the most important thing to understand about prompts, and it keeps the behavior predictable:

<CardGroup cols={2}>
  <Card title="Persona + Instructions" icon="layer-group">
    **Additive.** They layer on top of AutoSage's built-in answering behavior as authoritative overrides — the assistant still benefits from grounded retrieval and citations, now in your voice and within your rules.
  </Card>

  <Card title="Custom prompt" icon="pen-ruler">
    **A full replacement.** It takes over the built-in answering instructions entirely. When a custom prompt is set, the Persona and Instructions fields are not applied.
  </Card>
</CardGroup>

<Warning>
  Persona and Instructions are mutually exclusive with a full Custom prompt. Use **Persona + Instructions** for most cases — it is the simplest way to shape the assistant while keeping AutoSage's grounded answering intact. Reach for a full **Custom prompt** only when you need total control of the assistant's behavior, and remember it takes over from Persona and Instructions. Put everything you need — voice, rules, and answering guidance — directly in the custom prompt.
</Warning>

<Tip>
  Saving a field empty clears it. To go back from a full custom prompt to persona-and-instructions behavior, clear the custom prompt and the Persona and Instructions fields apply again.
</Tip>

## Tuning Quick and Deep modes separately

AutoSage answers in two modes — **Quick** for fast responses and **Deep** for thorough, multi-step reasoning (see [Answering modes](/chat/modes)). You can tune each independently:

<ParamField path="Quick mode custom prompt" type="text">
  The standard custom prompt applies to **Quick** mode answers.
</ParamField>

<ParamField path="Deep mode custom prompt" type="text">
  The main custom prompt applies to **Deep** mode answers.
</ParamField>

<Note>
  Because each mode has its own custom prompt, you can give a brief, snappy voice to Quick answers and a more detailed, methodical one to Deep answers within the same knowledge base.
</Note>

## Advanced fields for connected data

When a knowledge base has database connections, two more fields let you guide how the assistant works with live data:

<ParamField path="SQL instructions" type="text">
  Guidance for how the assistant queries your connected data — which tables to prefer, conventions to follow, and constraints to respect.
</ParamField>

<ParamField path="Chart instructions" type="text">
  Guidance for how query results are visualized — preferred chart types, formatting, and labeling.
</ParamField>

## Next steps

<CardGroup cols={2}>
  <Card title="Answering modes" icon="bolt" href="/chat/modes">
    See how Quick and Deep modes differ and when to use each.
  </Card>

  <Card title="Build an agent" icon="robot" href="/agents/overview">
    Package your prompt settings into a deployable agent.
  </Card>
</CardGroup>