> For the complete documentation index, see [llms.txt](https://help.usejimo.com/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.usejimo.com/docs/experiences/agent.md).

# Agent

Jimo's AI-powered assistant that helps your users get answers, discover features, and navigate your product through a conversational interface.

<figure><img src="https://2794860263-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FKzAcDWQbK1gKbpra7bkb%2Fuploads%2FXygIAMrs9hj6ZNU0OxOj%2Fimage.png?alt=media&amp;token=4701c069-1b52-4970-94b2-c26dc6f1df2e" alt=""><figcaption></figcaption></figure>

## What is the Agent?

The **Agent** is Jimo's built-in AI chat assistant. It turns your documentation, knowledge base, and curated answers into an interactive help layer embedded directly in your product. Users interact with it through a conversational prompt where they can ask questions in plain language and get precise, contextual replies.

Beyond answering questions, the Agent guides users on screen and completes tasks for them through [Skills](/docs/experiences/agent/skills.md). It does not start Tours or Checklists on its own. To send users to an existing experience from the chat, attach a Launch Experience button to a [Custom Answer](/docs/experiences/agent/knowledge/custom-answers.md). To hand over from an experience to the Agent, use the Launch Agent action.

***

## Overview

<figure><img src="https://2794860263-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FKzAcDWQbK1gKbpra7bkb%2Fuploads%2Fdg7XAYYmdXEcTSsMlgJP%2Fimage.png?alt=media&amp;token=54afeba6-1a31-4050-a66d-73f76969b455" alt=""><figcaption></figcaption></figure>

The Agent section in the Jimo dashboard is your central hub for configuring, training, and monitoring your AI assistant. It is accessible from the main sidebar navigation, marked with a **NEW** badge.

<figure><img src="https://2794860263-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FKzAcDWQbK1gKbpra7bkb%2Fuploads%2FUdJXP411x4GuIpPnVcuC%2F01-agent-sidebar-map.svg?alt=media&amp;token=74c5a5ea-5eaa-4c20-b4ac-3753225cbab1" alt=""><figcaption></figcaption></figure>

The Agent sidebar is organized in three groups, plus two upcoming features.

### Customize

* [**Chat**](/docs/experiences/agent/chat.md): the Agent's personality and behaviour (General Instructions), language, and the look of the chat widget.
* [**Launcher**](/docs/experiences/agent/launcher.md): the floating prompt that opens the Agent on your pages, with its own Where and Who targeting.

### Train

* [**Knowledge**](/docs/experiences/agent/knowledge.md): everything the Agent knows. Scanned pages of your product (Interface), training Sources, Custom Answers, and the user properties it may read (User Context).
* [**Skills**](/docs/experiences/agent/skills.md): reusable procedures the Agent can run on your product: guide users, execute a flow, or answer complex questions.
* [**Escalation**](/docs/experiences/agent/escalation.md): when and where the Agent hands a conversation to your support team (Intercom, Zendesk, Crisp, or email).

### Analyze

* [**Statistics**](/docs/experiences/agent/analyze.md#statistics): messages, users, success rate over time.
* [**Conversations**](/docs/experiences/agent/analyze.md#conversations): the full history of user conversations.

#### **Coming Soon**

* **Contextual Triggers**: trigger the Agent automatically from user context and behaviour.
* **Observe**: monitor user behaviour patterns to improve the Agent's responses.

<figure><img src="https://2794860263-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FKzAcDWQbK1gKbpra7bkb%2Fuploads%2F37E3lVACPGRFr95oQRi2%2Fimage.png?alt=media&amp;token=e24996c8-7615-439d-9926-606979d091a6" alt=""><figcaption></figcaption></figure>

***

### Use Cases

* 📚 **Self-Serve Support**\
  Let users ask questions in plain language and get instant answers from your knowledge base, reducing repetitive support tickets.
* 🎯 **Contextual Feature Guidance**\
  When a user asks how to do something, the Agent explains it from your sources, and, if a Guide Skill exists for that flow, offers to walk them through it on screen, step by step.
* 🚀 **Onboarding Assistance**\
  New users can ask the Agent how to get started. It answers from your documentation, offers the matching Skill, or, through a Custom Answer with a Launch Experience button, sends them to your onboarding tour or checklist.
* 🔍 **Content Discovery**\
  Replies cite the documentation they come from, and Custom Answers can link to a Changelog post, a URL or an experience, helping users discover features they didn't know existed.
* ⚡ **Automated Task Execution**\
  With an [Execute a flow Skill](/docs/experiences/agent/skills/create-a-skill.md), the Agent completes tasks on behalf of the user: clicking, filling in forms, and navigating your product.&#x20;

***

## How Users Access the Agent?

End-users can reach the Agent in three ways: the Launcher, a floating prompt on your pages; an Ask Agent block in a Resource Center; or a Launch Agent action triggered from any experience.

<figure><img src="https://2794860263-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FKzAcDWQbK1gKbpra7bkb%2Fuploads%2F1g9NdfzDXjZObHGOocPY%2F03-three-ways-to-open-agent.svg?alt=media&amp;token=cf3367ec-c988-4196-8a20-a5336c6797ec" alt=""><figcaption></figcaption></figure>

### From the Launcher

The **Launcher** is a floating input bar at the bottom of your pages. It has its own on/off switch and its own Where and Who targeting, so it can be live on every page for everyone, or only on selected pages for a segment. Users can drag it along the bottom edge to keep it out of the way. On pages you have scanned, up to four **suggestions** (predefined prompts) show as chips above the bar.

Configure it in [**Agent > Customize > Launcher**](https://i.usejimo.com/agent/launcher); suggestions are set per page in [Knowledge > Interface](https://i.usejimo.com/agent/knowledge/interface). See [Launcher](/docs/experiences/agent/launcher.md).

<figure><img src="https://2794860263-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FKzAcDWQbK1gKbpra7bkb%2Fuploads%2F3atKzcxsJ4qYU2M0sEbD%2Fimage.png?alt=media&amp;token=e3484c2b-b789-4805-b4c1-0d8837d7e935" alt=""><figcaption></figcaption></figure>

### From a Resource Center

Add an **Ask Agent** block to any Resource Center. This block displays a prompt input field in the Resource Center header. When a user types a question and submits it, the Agent chat opens with that question pre-loaded.

<figure><img src="https://2794860263-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FKzAcDWQbK1gKbpra7bkb%2Fuploads%2Fg0qwRBxQOd2bJLpjqzSU%2Fimage.png?alt=media&amp;token=7b8ccd94-70a5-48f8-8ff5-f0ea72e0b718" alt=""><figcaption></figcaption></figure>

To add it: open your Resource Center in the builder, click **+ Add block** on the Header section, and select **Ask Agent** under Interactive.

For full configuration details (layout, text styling, placeholder customization), see the Ask Agent block section in the Resource Center Components documentation.

### From a Launch Agent Action

The **Launch Agent** action opens the Agent chat from anywhere in your experiences. Like any other action in Jimo, it can be assigned to:

* **CTA buttons** (Primary or Secondary) in Tours, Surveys, Banners and Hints.
* **Resource center** actions or **Checklist** tasks.
* **Progress triggers** as the "Then" result of an If/Then condition.
* **Survey conditions** as the "Then" result based on a user's answer.

<figure><img src="https://2794860263-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FKzAcDWQbK1gKbpra7bkb%2Fuploads%2F8DwqeF9ODNmjdvZhayLs%2Fimage.png?alt=media&amp;token=6a8bb0ea-8c48-4c0d-8d2f-28acb2335f34" alt=""><figcaption></figcaption></figure>

When configuring the Launch Agent action, you can optionally pre-fill a **Message** with **dynamic attributes** (Jimo Email, Jimo Username, or any custom attribute) so the Agent immediately starts processing a contextual, personalized prompt when it opens.

<figure><img src="https://2794860263-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FKzAcDWQbK1gKbpra7bkb%2Fuploads%2FVsoQHQKhHrbz1H0eMx9e%2Fimage.png?alt=media&amp;token=216e391c-c643-4534-965b-c2669221b916" alt=""><figcaption></figcaption></figure>

For the complete configuration reference (Message field, Attributes, use cases, and combination rules with other actions), see [Launch Agent](/docs/build/actions/available-actions.md#launch-agent) in the [Available Actions](/docs/build/actions/available-actions.md) documentation.

{% hint style="info" %}
Pre-filling a message is powerful for contextual guidance. For example, at the end of an onboarding tour, you can trigger the Agent with "What should I do next?" to seamlessly transition from guided to conversational help.
{% endhint %}

***

## What the Agent can do today

The Agent answers from your knowledge, and it guides or acts on screen only through a Skill that the user confirms, with the eye on. Here is the full picture.

<figure><img src="https://2794860263-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FKzAcDWQbK1gKbpra7bkb%2Fuploads%2FgXDH88OTNase0ryCqf5B%2F07-capabilities-today.svg?alt=media&amp;token=c70a741a-ed79-4bff-a662-7c01c6f354fa" alt=""><figcaption></figcaption></figure>

| Capability                                | Status                      | What it needs                                                                 | What the user sees                                                                                                                          |
| ----------------------------------------- | --------------------------- | ----------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------- |
| **Answer questions**                      | Live                        | Sources or Custom Answers                                                     | A streamed text answer, its sources, thumbs up and down.                                                                                    |
| **Guide on screen**                       | Live                        | A Guide Skill on a scanned page, the eye on, the user confirms                | A cursor and a tooltip on one control at a time; the user clicks.                                                                           |
| **Execute for the user**                  | Live                        | An Execute Skill on a scanned page, the eye on, the user confirms             | A form for missing values, then the Agent clicks and fills. Steps that send, pay, delete or change permissions are handed back to the user. |
| **Run your own code**                     | Live, needs developers      | Tools your app registers through the SDK (25 at most)                         | Runs without confirmation and takes priority over Skills.                                                                                   |
| **Navigate to another page**              | Partial                     | A running Skill with the eye on, or a Custom Answer with a Navigate To button | No "go to URL" ability on its own: a running Skill clicks, or points at, your app's own links.                                              |
| **Launch an existing experience**         | Through Custom Answers only | A Custom Answer with a Launch Experience button                               | A button under the matching answer, for a Tour, Survey or Checklist. The user clicks.                                                       |
| **Escalate to a human**                   | Live, off by default        | One destination in Escalation: Intercom, Zendesk, Crisp or a support email    | The support chat opens with a summary, or a "Talk to a human" card appears first.                                                           |
| **Ask a clarifying question**             | Live                        | Nothing                                                                       | A question card with free text or choices.                                                                                                  |
| **Suggested prompts**                     | Live                        | Up to 4 suggestions per scanned page, see Launcher                            | Starter questions as chips above the Launcher.                                                                                              |
| **Use user attributes**                   | Live                        | Properties allowlisted in User Context                                        | Answers that fit the user's plan, role or quota.                                                                                            |
| **Reply in the user's language**          | Live                        | Language set to Auto in Chat (the default)                                    | Replies in the language the user writes in.                                                                                                 |
| **Proactive nudges, contextual triggers** | Not available               |                                                                               | Shown as Coming Soon in the dashboard.                                                                                                      |

### The eye: letting the Agent see the page

Every end user has an **eye** control in the chat widget. It is **off by default**, and remembered per browser and per project. Only the end user can switch it; there is no admin setting yet.

* **Eye off:** the Agent receives only the page URL. It answers in text, and if a Guide or Execute Skill matches the question, it describes the steps and invites the user to turn the eye on. No Start button.
* **Eye on:** the Agent reads the visible page (text and interactive elements, with emails, phone numbers, card numbers and IBANs masked, and password, email, phone and one-time-code fields hidden). Skills are offered as a **Start** button and run on screen once the user taps it or says yes.

{% hint style="warning" %}
**The dashboard test chat always behaves as if the eye is on.** What you see in Test Knowledge or Try this skill is what a user with the eye on gets. A user with the eye off, the default, gets the text version. When you test a Skill, test it live, eye on, in a new chat.
{% endhint %}

#### Requests and plans

Every message a user sends counts as one request toward your plan's monthly Agent quota, and tapping a Skill's Start button counts as one more.&#x20;

No capability is gated by plan.&#x20;

Past the quota, users see "The agent isn't available right now. Please try again later."

#### What the Agent does not know

The Agent knows nothing about your Jimo project itself: it cannot tell a user which experiences are live, why a tour did not show, or which segment they belong to. It answers those questions from your documentation, if it covers them. Changelog posts reach the Agent only if their public page is added as a URL source. Project-level questions are for admins, through the [Jimo MCP](/docs/integrations/mcp.md) in an AI client.

***

## Set Up the Agent From Claude

Everything in this section can also be done from Claude, or any AI assistant connected through the [Jimo MCP](/docs/integrations/mcp.md). The assistant maps your screens, trains the Agent on your help content, writes and tests Skills, sets launcher suggestions, and puts the launcher live, with a preview before every change your end users would see.

Skills it writes start inactive, and going live is always its own confirmation.

{% hint style="info" %}
It needs the **Build and run my Jimo Agent** permission when you connect. See [What to Ask](/docs/integrations/mcp/what-to-ask.md) for example requests, and [Permissions & Safety](/docs/integrations/mcp/permissions-and-safety.md) for what stays in the dashboard.
{% endhint %}

***

## Best Practices

<details>

<summary><strong>Training your Agent</strong></summary>

* **Scan your key pages first.** The [Interface](/docs/experiences/agent/knowledge/interface.md) tab lets the Agent read the screens your users actually work on. Start with your 3-5 most visited pages, then expand.
* **Start small and focused.** Index your help center and core product documentation first. Avoid crawling your entire website including marketing pages, blog posts, and legal pages. A focused knowledge base produces better answers than a broad but noisy one.
* **Create Custom Answers for your top 10-20 support questions.** Look at your support tickets and chat logs to identify recurring questions, then create precise Q\&A pairs for them. The Agent will prioritize these over generated answers.
* **Retrain regularly.** If your product changes often, set auto-retraining to Daily or Weekly in Agent > Knowledge > Sources. If your docs are stable, Weekly or manual retraining is enough.

</details>

<details>

<summary><strong>Building your first Skills</strong></summary>

* **Start from a real support conversation.** The best first Skill answers a question your users actually ask. Pick a recurring "how do I..." and turn the answer into a Guide user in a flow Skill.
* **Choose the right type.** Execute a flow when the Agent should do the task, Guide user in a flow when the user should stay in control, Answer complex questions when text is enough. See [Skills](/docs/experiences/agent/skills.md).
* **Scan the page before creating the Skill.** A Skill runs on the elements the Agent knows. Scanning the relevant page first ([Interface](/docs/experiences/agent/knowledge/interface.md)) gives it the vocabulary.

</details>

<details>

<summary><strong>Writing good General Instructions</strong></summary>

* **Be specific about your product.** Mention your product name, key features, and target audience so the Agent uses the right vocabulary.
* **Define the tone.** Should the Agent be friendly and casual, or professional and precise? State it explicitly.
* **Set boundaries.** List topics the Agent should and should not cover. For the moments when the Agent can't help, do not write a fallback sentence here: configure [Escalation](/docs/experiences/agent/escalation.md) instead, so the hand-off happens inside the chat with the conversation attached.
* **Use a preset as your starting point.** The five built-in presets (AI Agent, In-App Coach, Onboarding Specialist, Support, Expert Explainer) are good foundations that you can then tailor to your product.

</details>

<details>

<summary><strong>Integrating the Agent in your product</strong></summary>

* **Don't hide it.** Add the Ask Agent block to your main Resource Center so users can discover it naturally.
* **Use "Launch Agent" actions at the right moments.** Add contextual triggers at the end of a tour ("Have more questions?"), inside a checklist step ("Need help with this?"), or as a CTA in a changelog post ("Learn more about this feature"). The pre-filled message is text the Agent interprets; it cannot target a specific Skill yet.
* **Tell users about the eye.** Guide and Execute Skills only run on screen when the user has turned the eye on, and it is off by default. A line in your onboarding, or in the Agent's start screen, saves a lot of "it only gave me text" moments.
* **Pre-fill messages when you can.** Using the Launch Agent action with a pre-filled message (and dynamic attributes like the user's name or email) makes the first interaction feel personal and immediately relevant.

</details>

<details>

<summary><strong>Handing over to your support team</strong></summary>

* **Connect your support tool first.** Escalation triggers do nothing until Intercom, Zendesk, Crisp or a support email is connected in [Agent > Train > Escalation](https://i.usejimo.com/agent/escalation).
* **Keep the three triggers on to start.** Explicit request, repeated "it didn't work", and frustration. Tune the thresholds after a week of conversations.
* **List the topics that must never be automated.** Billing, refunds, account deletion: add them as topics that go directly to support.
* **Share the user's name with the Agent.** Add it in [Knowledge > User Context](https://i.usejimo.com/agent/knowledge/user-context) so your support team knows who is asking.

</details>

<details>

<summary><strong>Monitoring and improving</strong></summary>

* **Review conversations weekly.** Go to Agent > Analyze > Conversations to spot questions the Agent struggles with. These are opportunities to add new sources or Custom Answers.
* **Track your success rate.** The Success Rate metric in Agent > Analyze > Statistics tells you how often the Agent provides satisfactory answers. If it drops, your knowledge base may need updating.
* **Test in a new chat, eye on, in the live widget.** Each conversation is one thread per browser tab: asking again in the same chat is not a clean retest, since the previous question steers the answer. Use New chat for each test, and remember the dashboard test chat behaves as eye on.
* **Iterate on your instructions.** The General Instructions are not a "set and forget" configuration. As you review conversations and spot patterns, refine the persona, add constraints, or adjust the tone.

</details>

{% hint style="info" %}
**The Agent is most effective when it combines all knowledge types together.**

* Interface pages so the Agent understands what users see on screen.
* Static sources (URLs, files) for comprehensive documentation.
* Custom Answers for precise control over critical questions.
* User Context so answers fit the user's plan, role or language.
* Skills for hands-on guidance and automated task execution.
* Escalation so no user gets stuck when the Agent reaches its limit.
  {% endhint %}