# AI Chat

import { TileAiChat } from "../../../src/tiles/TileAiChat";

**AI Chat** in the sidebar is the main entry point to everything the AI can do, with
access to your library, your knowledge and your brand.

<figure className="gai-mock">
	<TileAiChat
		agentName="Brand Assistant"
		messages={[
			{ role: "user", text: "What's our refund window for annual plans?" },
			{
				role: "ai",
				text: "Sixty days from the invoice date for annual plans, prorated after the first thirty.",
				citations: ["Refund Policy", "Billing FAQ"],
			},
			{ role: "user", text: "Draft a reply to a customer asking about it" },
		]}
		placeholder="Ask a follow-up..."
	/>
	<figcaption>
		The <b>citations</b> under the answer name what the agent read. An answer with no
		citations came from the model rather than from your knowledge.
	</figcaption>
</figure>

## Start a conversation

Choose **New chat**, pick an agent, and type. If you have set a default agent in your
preferences, new chats start with it.

The starters under the box cover the common openings:

- *Create a 4K image of…*
- *What's the latest news about…*
- *Search my knowledge base for…*
- *Brainstorm ideas for…*
- *Write a first draft of…*
- *Summarize the key points of…*

## What an agent can reach

While a reply is being written, the app shows what the agent is doing: retrieving your
organization's knowledge, reasoning through the request, reviewing your brand identity.

Under a finished reply, **References** lists what the agent drew on and **Used n tools**
shows what it ran. Check them. An answer with no references came from the model's
general knowledge rather than from your material, which is worth knowing before you act
on it.

## Attach files and documents

**+** attaches files. Typing **@** mentions a document from your knowledge bases so the
message is answered against that document specifically. A message can carry a limited
number of reference documents; remove one to add another.

Typing **/** opens the **tool palette**: every tool available to you, searchable, with
your favorites and recents first. See [tools](/guide/ai/tools).

## The canvas

When a reply is something you will edit rather than read (a document, an infographic, a
piece of HTML), it opens in the **canvas** beside the conversation.

In the canvas you can edit directly, and select text to run **Polish**, **Summarize** or
**Expand** on just that part. **Accept** and **Reject** apply or discard a suggested
change.

Documents export as **HTML** or **DOCX**; infographics download as **PNG** or **HTML**.
**Add to media library** puts a generated image into the library so the rest of the
product can use it.

## Spaces

**Spaces** group related chats, share a system prompt, and let teammates in. A space for
a campaign, a project or a team gives everyone working on it the same context without
setting it up each time.

- **Personal** holds your private chats
- **Shared with me** holds chats teammates have shared with you
- A space can set a **default agent**, so new chats in it start with the right one

## Share a conversation

**Share** on a conversation adds people and then gives you a link. Only the people you
added can open it: the link alone is not access. Each person is a **Viewer** (read), a
**Member** (read and post), or an **Owner** (full access, including sharing).

## Find an earlier conversation

**Search chats** searches titles and message content, grouped into Today, Yesterday,
Previous 7 days, Previous 30 days and Older. Recent chats are searched instantly, and
**Search older chats** extends the range.

## Troubleshooting

| | |
|---|---|
| **Streaming interrupted** | The connection dropped mid-reply. **Retry** re-runs the message. |
| **Agent is no longer available** | The agent was deleted or you lost access to it. Start a new chat with another one. |
| **You're out of credits** | The workspace pool is empty; see [usage and credits](/guide/admin/usage-and-credits). |