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

# In-App Chat

> Use Inflection's in-app agent to ideate journeys, build segments, and more.

The Inflection in-app chat lets you build marketing assets by describing what you want in plain language. Describe an audience, email, or journey — the agent plans the work, builds it, and shows you the result inline, ready to review.

## Getting Started

Open **AI chat** inside Inflection. There is nothing to install. On your first message, the agent loads your business context (industry, ICP, common segments) so what it builds fits your account. The same agents are also available over the MCP connector for users working from Claude or ChatGPT.

<Info>
  **Try this in your first 10 minutes:** Type *"How many contacts opened an email in the last 30 days?"* to see the agent ground in your data, then *"Draft a re-engagement email for the ones who didn't."* You'll have a previewable email and a sense of the flow within minutes.
</Info>

## What You Can Do Here

### Overview

| Function      | Ideate | Create | Edit   | Outcome                                                          |
| ------------- | ------ | ------ | ------ | ---------------------------------------------------------------- |
| **Audiences** | ✅      | ✅      | ✅      | A saved, auto-refreshing dynamic list — reusable across journeys |
| **Segments**  | ✅      | ✅      | via UI | A mutually-exclusive category for lifecycle/tier messaging       |
| **Email**     | ✅      | ✅      | ✅      | An on-brand, previewable email saved to your library             |
| **Tokens**    | ✅      | ✅      | ✅      | Verified personalization — including computed scores             |
| **Journeys**  | ✅      | ✅      | ✅      | A multi-step draft, ready for your review and activation         |
| **Analytics** | ✅      | —      | —      | Dashboard-accurate performance, rendered inline                  |

### Build Audiences

Describe who you want to reach. The agent grounds in your real field values and saves a reusable dynamic list.

Consistent, reusable targeting without learning the filter UI.

<Info>
  **Example prompts:**

  * "Build a list of enterprise accounts in California who haven't logged in for 30 days."
  * "How big is that audience?" · "Save it as 'Enterprise – dormant'."
</Info>

### Create and Edit Email

Draft from a brief, restyle elements in plain language, or edit a saved template **by name** — then preview before anything sends.

Review a thumbnail preview before you ship.

<Info>
  **Example prompts:**

  * "Create a welcome email with a 20% offer and one clear CTA."
  * "Edit the 'Trial Welcome' email — make the CTA green and tighten the hero copy."
</Info>

### Build Segments and Tokens

Create ordered, mutually-exclusive segment categories (lifecycle, tier, engagement) and verified personalization tokens — including arithmetic scores such as a weighted engagement metric.

<Info>
  **Example prompts:**

  * "Create lifecycle segments: New, Active, At Risk, Churned."
  * "Build a score = 3×demo\_requests + 1×email\_opens over 90 days."
</Info>

### Build Journeys

Describe a multi-step flow — emails, delays, branches, Salesforce sync, webhooks — and the agent assembles a **draft** you review and activate in the app.

<Info>
  **Example prompt:**

  * "Draft a 3-step onboarding journey: welcome email, 2-day delay, branch on opened."
</Info>

### Ask About Performance

Pull campaign and journey results in the same chat — dashboard-accurate numbers, rendered as a clean artifact.

<Info>
  **Example prompt:**

  * "How did the Q1 newsletter perform — delivered, opens, clicks?"
</Info>

### The Chat Experience

* **Live plan** — the agent shows its plan as a checklist and ticks items off as it works.
* **Streaming activity** — see which specialist and which tool is running, live.
* **Parallel-agent tree** — fan-out work (for example, several emails) renders as a tree, not a wall of text.
* **Inline artifacts and previews** — emails show a thumbnail; journeys render as a Mermaid diagram.
* **File uploads** — drop in a brief or asset and the agent works from it.

## Use Cases

**Brief to reviewable journey, in one sitting.** Paste a campaign brief; the agent sizes the audience, drafts the emails, and assembles the journey draft — review and activate when ready. Lifecycle teams run their whole program this way — onboarding, activation, re-engagement, win-back — even with a small team.

**The weekly newsletter, faster.** Draft and restyle the email, confirm the audience, drop it into the send journey.

**Live audience sizing on a working call.** Ask counts and compositions to pressure-test an idea before you build it — useful for demos and planning sessions.

## Limitations

**Drafts, not auto-activation.** You decide when a journey or send goes live.

**Segment sizes settle after save.** A new category populates within approximately 10 minutes; previews before then can overlap.

**Custom CSS lives in the editor.** Plain-language style edits cover common cases; bespoke `@media`/`<style>` rules go in the BEE editor.

**One build per request.** Ask for one email or audience at a time for clean, reviewable steps.

### What the Agent Can Access

Your data schema and business context, your existing assets (journeys, emails, segments, tokens), filter types and token functions, and analytics rollups — enough to build things that fit your account.

### What It Cannot Access

Individual PII beyond a contact lookup you request, other users' chats, or anything outside your Inflection workspace.

## Prompting Best Practices

**Lead with the goal, not the mechanics.** "Re-engage trial users who stalled after signup" works better than a list of filters.

**Let it ground first.** The agent checks your real values before composing — give it the business term.

**Iterate in conversation.** Start rough, refine over follow-ups.

**Say "save" when you mean it.** Nothing is committed until you ask.

## Report Unexpected Outcomes

If the agent is not delivering after following these tips, tell your CS team in Slack: include a link to the conversation, your goal, and screenshots of the prompt and the output.

## Share Your Use Cases

We tune the agents on how you use them — tell us what worked, where it struggled, and what would make it more useful.