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

# Create a Whatsy persona that matches your writing style

> Build a Whatsy persona by defining your tone, writing style instructions, and example reply pairs so the LLM matches how you actually write.

A persona is the core of Whatsy's intelligence. It tells the local LLM who it is replying as, how that person writes, and what their replies should sound like. A well-crafted persona makes automated replies feel natural and personal; a vague one produces generic-sounding responses that don't match your style. Spending a few minutes getting the persona right makes a significant difference in quality.

## What goes into a persona

Every persona is made up of four components:

* **Name and description** — a label and summary for your own reference (e.g. "The Professional", "Polite, concise, focused on scheduling")
* **Tone** — a short phrase describing the communication style (e.g. "Formal / Direct", "Warm / Casual", "Helpful / Patient")
* **Writing instructions** — a markdown system prompt that describes exactly how you write, what to avoid, and any special rules
* **Example reply pairs** — real or representative message exchanges showing an incoming message and the reply you would send

## Creating a persona in the app

<Steps>
  <Step title="Open the Personas view">
    Open the Whatsy app and click **Personas** in the left sidebar. You'll see your persona library. Any existing personas are listed here as cards showing their name, status (active or draft), and total chats handled.
  </Step>

  <Step title="Create a new persona">
    Click the **+ New Persona** button in the top-right corner. Whatsy creates a new draft persona and opens the persona editor automatically.
  </Step>

  <Step title="Enter a name and description">
    In the **Name** field, give the persona a memorable label (e.g. "Out of Office", "Weekend Mode"). In the **Description** field, write a one-sentence summary of the persona's purpose and audience — this is only for your reference.
  </Step>

  <Step title="Describe the tone">
    Fill in the **Tone** field with a short phrase that captures the communication style. Examples: `Casual / Warm`, `Formal / Direct`, `Helpful / Patient`. This feeds directly into the LLM prompt and shapes the overall register of replies.
  </Step>

  <Step title="Write your persona instructions">
    The **System Prompt** field is where you describe how you actually write. Be specific — the more detail you give, the better the output. See the example below for a starting point.
  </Step>

  <Step title="Add example reply pairs">
    Scroll to the **Example Replies** section and click **Add** to insert a pair. Each pair has an **Incoming** field (the message someone sends you) and a **Reply** field (what you would write back). Add at least five pairs to give the model enough signal.
  </Step>

  <Step title="Save and activate">
    Click **Save** to save the persona as a draft. Review your instructions and examples, then click **Set as Active** in the editor header to make this the persona Whatsy uses for auto-replies.
  </Step>
</Steps>

## Example persona instructions

Write your system prompt in markdown. Here is a solid starting point you can adapt:

```markdown theme={null}
You are replying as the user in their personal WhatsApp chats. Match their style and tone exactly.

- Write in all lowercase.
- Be warm and natural, not stiff or formal.
- Keep replies short — one to three sentences at most.
- Use emojis sparingly, only when they feel natural.
- Never reveal that you are an AI or an automated system.
- If you don't know the answer to something specific, say you'll follow up later.
- Don't use bullet points or headers in replies — write conversationally.
```

## Example reply pairs

Include pairs that cover the kinds of messages you actually receive. Here are three examples to illustrate the format:

| Incoming message                         | Your reply                                                                         |
| ---------------------------------------- | ---------------------------------------------------------------------------------- |
| Hey, are you free to catch up this week? | yeah for sure! how does thursday afternoon work for you?                           |
| When will the report be ready?           | Should have it over to you by end of day Friday. I'll ping you when it's done.     |
| Can you recommend a good plumber?        | i used a guy called Mike last year — super reliable. want me to dig up his number? |

<Tip>
  The more specific and varied your examples, the better the LLM can calibrate to your voice. Aim for at least ten pairs that cover different topics and message lengths. Include a mix of short casual replies and slightly longer ones so the model learns when to be brief and when to say more.
</Tip>

## Activating a persona

Only one persona can be active at a time. To switch which persona handles auto-replies, open the persona you want in the editor and click **Set as Active**. The previously active persona automatically reverts to draft status. The dashboard always shows which persona is currently active.

## Working with multiple personas

You can maintain as many draft personas as you like — for example, a "Vacation Mode" persona for when you're travelling, a "Professional" persona for business contacts, and a casual one for friends. Switch between them by opening the relevant persona and clicking **Set as Active**. Use [behavior rules](/guides/behavior-rules) to route specific contacts or message types to a particular persona automatically.