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

# AI & Replies Overview

> Speed up your inbox with AI-drafted replies, quick canned responses, and a connected AI provider.

Xobito gives your agents three AI-adjacent tools that all live in the Live Chat composer. Used together, they let an agent reply to a common question in a couple of clicks, and draft a nuanced reply to a less-common one in a few seconds.

<Note>
  None of these tools *send* a message automatically. They draft or insert text into the composer; your agent reviews, edits, and sends. Humans stay in the loop.
</Note>

## The three tools

<CardGroup cols={3}>
  <Card title="AI Prompts" icon="sparkles" href="/ai-features/ai-prompts">
    Reusable, named prompts ("Friendly reply", "Polite decline") that draft a response using AI.
  </Card>

  <Card title="Canned Replies" icon="clipboard" href="/ai-features/canned-replies">
    Saved static text snippets your agents insert from the composer picker — no AI needed.
  </Card>

  <Card title="AI Integration" icon="plug" href="/ai-features/ai-integration">
    Connect an OpenAI-compatible provider so the AI Prompts feature can actually call a model.
  </Card>
</CardGroup>

## When to use which

| Situation                                                                            | Best tool                    |
| ------------------------------------------------------------------------------------ | ---------------------------- |
| Same answer every time ("Our hours are Mon–Fri 9–6")                                 | **Canned Reply**             |
| Needs a human-sounding reply tailored to the customer's question                     | **AI Prompt**                |
| You want agents to pick from a library of tones ("friendly", "formal", "empathetic") | **AI Prompt** (one per tone) |
| Agent needs a one-click way to insert fixed text like office hours                   | **Canned Reply**             |
| Customer's question is new — no AI prompt or canned reply fits                       | Agent writes manually        |

<Tip>
  Canned Replies are faster and don't cost AI credits. Use them for anything static. Reserve AI Prompts for situations where the reply needs to actually read the customer's message.
</Tip>

## How they work in Live Chat

When an agent opens a conversation, the composer has three controls:

1. **AI draft button** (sparkles icon) — opens a menu of your AI Prompts. Pick one, AI drafts a reply, agent edits and sends.
2. **Canned-reply icon** (chat-bubble icon, next to the attachment paperclip) — opens the canned-reply picker. Search by title and click to insert.
3. **Manual typing** — always available.

## Setup order

If you're starting from scratch, set things up in this order:

<Steps>
  <Step title="Connect an AI provider">
    Go to **Settings → AI Integration** and add your API key. Without this, AI Prompts can't call a model.

    See [AI Integration](/ai-features/ai-integration).
  </Step>

  <Step title="Build a few AI Prompts">
    Start with 3–5 general-purpose prompts: "Friendly reply", "Acknowledge and escalate", "Polite decline".

    See [AI Prompts](/ai-features/ai-prompts).
  </Step>

  <Step title="Build your canned-reply library">
    List out the 20 most common static answers (hours, address, price list, etc.) and save each as a canned reply.

    See [Canned Replies](/ai-features/canned-replies).
  </Step>

  <Step title="Train your agents">
    Show your team the sparkles button and the canned-reply icon in the composer. Most agents get fast within a day.
  </Step>
</Steps>

## Cost and privacy

* **Canned Replies** are stored in your Xobito workspace. They never leave — no third party is involved.
* **AI Prompts** send the conversation context to your chosen AI provider. The provider's privacy policy applies to that data. If the conversation contains sensitive information and you're unsure, use Canned Replies instead.
* **AI Integration** billing is handled by your AI provider directly — Xobito just calls their API with your key. See [AI Integration](/ai-features/ai-integration) for cost-control tips.

<Warning>
  If you operate under strict privacy regulations (HIPAA, GDPR health data, etc.), review your AI provider's terms before enabling AI Prompts. Canned Replies remain a safe alternative.
</Warning>

## A sensible workflow

Even the best AI prompts don't replace human judgement — they just save typing. Here's a pattern that works well across teams:

<Steps>
  <Step title="Customer sends a message">
    Agent opens the conversation in Live Chat.
  </Step>

  <Step title="Try a canned reply first">
    Click the canned-reply icon in the composer. If a saved reply matches, click to insert, edit lightly, send. Done in seconds.
  </Step>

  <Step title="Fall back to an AI prompt">
    No canned reply fits? Click the sparkles icon and pick a prompt that matches the tone you want. AI drafts, agent reviews.
  </Step>

  <Step title="Write from scratch if nothing fits">
    The first time you hand-write a reply, consider whether a new canned reply or AI prompt would save you time next time.
  </Step>
</Steps>

## Permissions

Control access with these permissions, set per role in [Roles & Permissions](/team/roles-permissions):

| Permission                                         | What it unlocks                               |
| -------------------------------------------------- | --------------------------------------------- |
| `tenant.ai_prompt.view`                            | Use existing AI prompts from the composer     |
| `tenant.ai_prompt.create` / `.edit` / `.delete`    | Create, edit, delete AI prompts               |
| `tenant.canned_reply.view`                         | Use existing canned replies from the composer |
| `tenant.canned_reply.create` / `.edit` / `.delete` | Create, edit, delete canned replies           |
| `tenant.whatsmark_settings.edit`                   | Configure the AI Integration                  |

<Tip>
  Agents typically have the `view` permissions for prompts and replies. Managers and admins have `create` / `edit` / `delete`. Only admins need `tenant.whatsmark_settings.edit` to touch the AI Integration.
</Tip>

## Measuring impact

After a month or two of use, you should see:

* **Faster first response times** — canned replies cut typical 2–3 minute responses down to seconds.
* **More consistent tone** — agents using the same library produce noticeably more consistent replies.
* **Fewer "not my area" hand-offs** — a good AI Prompt library means even agents outside their specialty can draft reasonable replies.

Track this informally — for example, by spot-checking response times in the [Live Inbox](/chat/overview) before and after rolling out a canned-reply library.

## What to read next

<CardGroup cols={2}>
  <Card title="AI Prompts" icon="sparkles" href="/ai-features/ai-prompts">
    Build your library of reusable AI-drafted replies.
  </Card>

  <Card title="Canned Replies" icon="clipboard" href="/ai-features/canned-replies">
    Save and organise your team's quick answers.
  </Card>

  <Card title="AI Integration" icon="plug" href="/ai-features/ai-integration">
    Connect OpenAI, Anthropic, or any OpenAI-compatible provider.
  </Card>

  <Card title="Live Inbox" icon="comments" href="/chat/overview">
    Where these tools actually get used.
  </Card>
</CardGroup>