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

# Talking to Your AI Assistant

> Learn how to communicate effectively with your UltraClaw AI assistant across iMessage, Telegram, WhatsApp, and web chat.

# Talking to Your AI Assistant

Your UltraClaw agent is designed to feel like texting a knowledgeable friend. There's no special syntax to learn — just send a message in plain language and your agent will respond.

## How Your Agent Processes Messages

When you send a message, your agent:

1. **Reads your message** along with recent conversation context
2. **Recalls relevant memories** from past conversations
3. **Checks available tools** — calendar, email, files, integrations
4. **Thinks through its response** — optionally using extended reasoning
5. **Replies to you** on the same channel you messaged from

Your agent maintains a continuous conversation thread. It remembers what you discussed yesterday, last week, and last month. You never need to repeat yourself.

## Tips for Getting the Best Results

### Be specific

The more context you give, the better your agent performs.

| Instead of...        | Try...                                                             |
| -------------------- | ------------------------------------------------------------------ |
| "Schedule a meeting" | "Schedule a 30-min meeting with Sarah next Tuesday at 2pm"         |
| "Write an email"     | "Draft a follow-up email to the client about the Q1 proposal"      |
| "Look something up"  | "What were Apple's revenue numbers in their last earnings report?" |

### Give your agent standing instructions

Use [workspace files](/assistant/workspace-files) to tell your agent about your preferences once, so you don't have to repeat them:

* Your preferred communication style (formal, casual, concise)
* Time zone and scheduling preferences
* How you like emails and documents formatted
* Topics or areas you work in frequently

### Use multi-step requests

Your agent can handle complex, multi-part tasks in a single message:

> "Check my calendar for tomorrow, draft a prep doc for my 10am meeting based on the last email thread with that client, and remind me tonight at 9pm to review it."

### Correct and refine

If your agent's response isn't quite right, just say so naturally:

* "Make it shorter"
* "Actually, use a more formal tone"
* "Good but add a section about pricing"

Your agent learns from corrections and applies your preferences to future responses.

## Multi-Channel Support

Your agent is the same across every channel. Conversations, memory, and context are shared — you can start a conversation on iMessage and pick it up on Telegram or WebChat without missing a beat.

<CardGroup cols={2}>
  <Card title="iMessage" icon="message" href="/channels/imessage">
    Text your agent from your iPhone, iPad, or Mac. No app needed.
  </Card>

  <Card title="Telegram" icon="paper-plane" href="/channels/telegram">
    Chat with your agent in a Telegram bot conversation.
  </Card>

  <Card title="WhatsApp" icon="phone" href="/channels/whatsapp">
    Reach your agent on WhatsApp for international messaging.
  </Card>

  <Card title="WebChat" icon="browser" href="/channels/webchat">
    Chat from your browser via the UltraClaw dashboard.
  </Card>
</CardGroup>

<Info>
  Slash commands, directives, and stop phrases work identically across all channels. See the [Slash Commands](/assistant/slash-commands) reference for the full list.
</Info>

## Sending Files and Images

You can send images, PDFs, and documents to your agent on supported channels. Your agent can:

* **Read and summarize** documents and PDFs
* **Analyze images** — charts, screenshots, photos, receipts
* **Extract data** from files and use it in responses

<Note>
  File support varies by channel. iMessage and WebChat support the widest range of file types. Check your [channel's page](/channels/comparison) for specifics.
</Note>

## When Your Agent Can't Help

Your agent will tell you honestly if it can't complete a task. Common reasons include:

* The task requires an integration that isn't connected yet
* The request involves real-time data your agent doesn't have access to
* The action would need your explicit approval (your agent will ask first)

If you're ever stuck, send `/help` for a quick overview of what your agent can do, or visit our [troubleshooting guide](/troubleshooting/common-issues).