> ## 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.

# Whatsy: autonomous WhatsApp replies in your own voice

> Whatsy is a local-first macOS app that reads WhatsApp messages and replies as you — powered by a local AI model, with no cloud data sharing.

Whatsy is a macOS app that keeps your WhatsApp conversations alive when you're unavailable. It reads incoming messages, thinks in your voice, and sends replies from your number — all without touching the cloud. Everything runs on your Mac using a local language model through [Ollama](https://ollama.com), so your words, your contacts, and your persona never leave your device.

## What is Whatsy?

Whatsy acts as your personal messaging stand-in. When a message arrives on WhatsApp, the agent picks it up, weighs your persona and any rules you've set, and sends back a reply that sounds like you. The whole loop runs locally:

* **Autonomous replies** — the agent handles incoming WhatsApp messages while you focus on something else.
* **Local LLM via Ollama** — you choose and run the model yourself (we recommend `llama3.2`). No API key, no subscription, no cloud endpoint.
* **Runs on your Mac** — a native Swift/SwiftUI app (no Electron) that stays in your menu bar and keeps the agent alive even when the window is closed.
* **Your number, your tone** — Whatsy links to WhatsApp via a QR scan on your existing account and replies from your real number, styled to match the way you actually write.

## Key features

<CardGroup cols={2}>
  <Card title="Persona Builder" icon="user-pen">
    Write a persona that captures your voice, vocabulary, and communication style. The agent uses it as its core identity when crafting every reply.
  </Card>

  <Card title="Behavior Rules" icon="sliders">
    Define conditional rules — such as "if the message is from an unknown contact, reply with this" — to control exactly how the agent responds in different situations.
  </Card>

  <Card title="Privacy-First" icon="shield-halved">
    No telemetry, no cloud AI, no third-party servers. Your message history, persona, and WhatsApp session stay entirely on your Mac.
  </Card>

  <Card title="Activity Dashboard" icon="chart-bar">
    Review every automated reply, see which rule or persona triggered it, and inspect the full conversation thread — all from inside the app.
  </Card>

  <Card title="Response Delay" icon="clock">
    Add a configurable delay between receiving a message and sending a reply so responses feel natural rather than instant.
  </Card>

  <Card title="Spam Filtering" icon="ban">
    Built-in keyword detection silently drops obvious spam before it ever reaches the AI model, keeping your activity log clean.
  </Card>
</CardGroup>

## What stays private

Whatsy is designed so that sensitive data never has a reason to leave your Mac. Here's what stays local:

* **Message content** — every incoming and outgoing message is stored only in an on-device database at `~/Library/Application Support/Whatsy/`.
* **Your persona** — the persona file that shapes the agent's voice lives on disk, not on any server.
* **WhatsApp session** — the authentication session created when you scan the QR code is saved locally and is never transmitted.
* **Conversation history** — the context window the agent uses to craft replies is assembled in memory on your machine and discarded when the agent stops.

<Note>
  Whatsy uses [Ollama](https://ollama.com) to run the language model entirely on your hardware. No OpenAI key, no Anthropic account, no external inference endpoint — replies are generated locally and only locally.
</Note>

## Coming soon

Whatsy currently supports **WhatsApp only**. Support for **Telegram** and **Discord** is actively planned and will be added in a future release. You'll see the option appear in the app as each platform ships.