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

# Messaging basics

> How to have a conversation with WednesdayAI: context, memory, sharing files, and group behaviour.

# Messaging basics

Talk to WednesdayAI the same way you would with any contact - just send a message. There is no special syntax for normal conversation.

## Starting a conversation

Send any message to the bot on your connected channel (WhatsApp, Telegram, etc.). The agent responds using your conversation history as context.

> [!NOTE]
>
> **First time?** See [Pairing your first channel](https://docs.wednesdayai.dev/next/users/pairing/first-channel.md) if you have not yet been connected to the gateway.

## Context vs memory

These are two different things, and the difference matters:

* **Context** is the running conversation in your current **session** - a window of recent history the agent can see. Context is bounded: very long conversations eventually drop their earliest messages, and starting a fresh session clears it.
* **Memory** is durable. When you ask the agent to *remember* something, it writes it to a file that survives across sessions, so it still knows it tomorrow or next week.

> [!TIP]
>
> If you want a fact or preference to stick, ask the bot to remember it: "Remember that I prefer TypeScript." See [Memory](https://docs.wednesdayai.dev/next/users/memory.md) for how this works.

**To start a fresh conversation** (clears context for the new session):

```
/reset
```

`/new` does the same. You can start fresh on a specific model:

```
/new <model-name>
```

(The name can be a short alias your admin has set, or a full model id like `provider/model-name` - or just send `/model` to see what is available.)

## Getting status

Send `/status` at any time to see the active model, gateway connectivity, and provider usage or quota (when tracking is enabled).

## Sharing files, images, and voice

You can send the agent attachments directly in the chat - a photo, a screenshot, a PDF, a document, or a voice note - and ask questions about them:

> (attach a photo) "What is in this picture?"
>
> (attach a PDF) "Summarise this report in five bullet points."

See [Sharing files](https://docs.wednesdayai.dev/next/users/files.md) for what each channel supports.

## Steering the conversation

You can guide the AI with natural language - no special commands needed. To be explicit, use directives:

* `/think high` - ask the model to reason carefully before responding
* `/verbose on` - show detailed tool output and reasoning steps
* `/model <name>` - switch to a specific model for this session

Directives can lead a message:

```
/think high
Explain how database indexes work and when to use them.
```

A directive sent on its own (nothing else in the message) sticks for the rest of the session. A directive in front of a real message applies to that message only.

### Thinking levels

`/think` controls how hard the model reasons before replying. Levels:

`off` `\|` `minimal` `\|` `low` `\|` `medium` `\|` `high` `\|` `xhigh` `\|` `adaptive`

Higher levels are slower but better for hard problems. `adaptive` lets the model pick its own budget (the default on newer models). `xhigh` is only supported by certain models.

## In a group chat

When the bot is in a group, it does **not** reply to every message. By default it only wakes up when you **mention or ping it** (for example `@YourBot what's the weather?`) or reply to one of its messages. Each group has its own separate session, so commands and context in a group do not affect your private DM with the bot.

## Long conversations

If a conversation is getting long and the agent is losing track or making mistakes, summarise and continue:

```
/compact
```

You can add focus instructions:

```
/compact Focus on the authentication module we discussed.
```

Or just start fresh with `/reset`.

## Tips

* **Tool use is automatic** - if the agent needs to search the web or run code, it will. Just ask naturally.
* **Quick messages are batched** - if you fire off several short messages in a few seconds, they are bundled into one turn rather than triggering separate replies. Attachments and commands are handled immediately.
* **One device for long chats** - your history lives on the gateway, not your phone. If you use multiple devices, stick to one for a long conversation so context does not diverge.
* **Inline shortcuts** - a few commands work mid-message: `Hey /status, what's the weather in Sydney?` triggers a status reply and then answers the rest.
* **Voice** - if your admin has enabled it, ask for spoken replies with `/tts on` (on Discord the command is `/voice`).

## Quick command reference

| Command | Effect |
| - | - |
| `/reset` or `/new` | Start a fresh conversation |
| `/status` | Current model, gateway status, quota |
| `/help` | Show available commands |
| `/think <level>` | Adjust reasoning depth (`off`-`xhigh`, `adaptive`) |
| `/model <name>` | Switch AI model |
| `/compact` | Summarise context and continue |
| `/verbose on` | Show tool output and reasoning |
| `/usage tokens` | Show token count on each reply |

Full command reference: [Slash commands](https://docs.wednesdayai.dev/next/users/commands.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.