Configure AI Assistant

This article covers how Amplitude admins and builders configure AI Assistant, Resource Center, and related content and styling.

Enable AI Assistant and choose where it appears

Configuration typically involves:

1. Enable AI Assistant for your org

   Make sure you enable AI Assistant for your organization and projects. This may involve:

   • Having AI Assistant included in your Amplitude plan.
   • Enabling AI Assistant in AI Assistant > Settings > AI Assistant.

2. Configure the widget launcher

   From the Assistance / AI Assistant or Resource Center configuration area:

   • Choose a pre-built launcher, a custom element, or custom code to determine where the launcher appears. For example, on the bottom-right of all pages, or only on certain screens.
   • Specify targeting logic based on:
     • User properties (such as role or plan).
     • Cohorts defined in Amplitude.
     • URL or route rules.

3. Choose which surfaces to enable

   In the widget configuration:

   • Turn Chat (AI Assistant) on or off.
   • Turn Resource Center on or off.
   • Decide whether users land on Chat or Resource Center by default when the widget opens.

Connect content sources

AI Assistant is only as good as the content you give it. You manage that content through Content & Sources and Resource Center.

Add content sources

Examples of sources include:

• Your public docs or help center.
• An internal knowledge base.
• Product marketing pages.
• A blog with how-to content.

To connect a source:

1. Go to AI Assistant > COntent > Sources.
2. Click Add source and choose the type (for example, "Help center", "Web", or "Docs site").
3. Provide the required details such as base URL, credentials if needed, or site map.
4. Configure the sync frequency (automatic or on-demand).
5. Save and run an initial sync.

Each source ingests many content items (individual articles, pages, or documents). Amplitude indexes these for both:

• Resource Center (search and recommendations).
• AI Assistant (through its documentation search tool).

Plan limits may control how many sources and documents for each source you can connect; check your Amplitude plan if you hit a limit.

Curate Resource Center recommendations

Use the Resource Center configuration to build recommendation sets for different contexts:

• By page or feature.
• By user segment.
• As a global default.

You can include:

• Articles and documents.
• Guides & Surveys content (tours, checklists, surveys).
• Links or other resources.

AI Assistant can then:

• Link users to these items in its answers.
• Trigger guides or surveys as part of workflows.

Create custom answers

Custom answers let you fill specific gaps in your documentation or enforce exact wording for high-importance topics.

Typical use cases:

• "What is your SLA?"
• "What's the difference between the Business and Enterprise plans?"
• "How do I cancel my subscription?"

To create an answer:

1. Navigate to AI Assistant > Content > Answers.
2. Click Create New.
3. Set the following:
   • Status: Draft or Published. Determines if the answer is available to AI Assistant.
   • Title: A title or short description of the answer.
   • Answer: The text you want AI Assistant to use.
   • Answer is verbatim: Instructs AI Assistant to use the text as you provide it, rather than interpret it. This is useful when you have specific or important topics you want to convey, for example, when a user asks about a refund.
4. Click Create.

Amplitude indexes answers and makes them available to AI Assistant. Answers don't appear in Resource Center as standalone articles—they exist specifically to power AI Assistant's responses.

Configure fallbacks and handoffs

When AI Assistant can't confidently answer a question, it falls back gracefully rather than guessing.

To configure a fallback:

1. Navigate to AI Assistant > Behavior > Fallbacks.
2. Configure a fallback for each of the following scenarios:
   • AI Assistant can't find an answer to a user's question.
   • A user provides negative feedback.
   • A user requests to escalate to a human.

Each fallback includes:

• Targeting, so you can configure different fallback options for different cohorts or segments of users.

• A fallback message that you define, for example: "I'm not confident I can answer that correctly. Let me connect you with another resource."

• An action or assist:

  Action	Description
  Visit link	Supply a URL that opens along with AI Assistant's response.
  Click element	Specify an element on the page, which you can use to trigger another action.
  Run callback	Run a callback function you defined in the Guides & Surveys SDK.
  Open third-party chat	Select a chat provider to direct the user to. This action is most useful with the Escalate to human fallback.
  Show guide	Select an existing guide to launch.
  Show survey	Select an existing survey to launch.
  Open article	Select an indexed article to open in Resource Center.
  Play video	Select a video to play. Add and manage available videos in Guides and Surveys > Content > Videos.

Configure workflows and tools

Under the hood, AI Assistant can call tools (APIs) and workflows (multi-step procedures) so it can do more than just answer questions.

Built-in tools

By default, AI Assistant has access to tools such as:

• Documentation search: finds relevant help content.
• Guides search: finds relevant Guides & Surveys experiences.
• Guide launcher: starts a specific guide or checklist in the user's product.
• Fallback tools: AI Assistant uses these when it can't answer confidently or when users provide negative feedback.

AI Assistant automatically considers these tools whenever they could help answer or act on a user's request.

Workflows

Workflows string together multiple steps and tools into a reusable pattern. For example:

• "Help a user reset their password."
• "Troubleshoot why a dashboard is empty."
• "Walk through a feature adoption checklist."

A typical workflow might:

1. Ask clarifying questions in chat.
2. Call an internal API to look up data.
3. Decide between multiple branches based on that data.
4. Launch a relevant guide or checklist.
5. Exit back to AI Assistant chat with a summary.

To create a workflow:

1. Navigate to AI Assistant > Automations > Workflows and click Create.
2. Define the workflow's trigger. For example, define the words or phrases that trigger the workflow when a user enters them. Optionally, tell AI Assistant when to exit a workflow, for example, when the user changes topic or no longer responds.
3. Define the workflow. Use natural language to inform AI Assistant how it should respond. Click Insert Actions or enter @ to insert an action into the workflow.
4. Set targeting for the workflow.
5. Click Save and Enable to activate the workflow.

Example workflow

This workflow helps users troubleshoot an empty dashboard by checking their plan and launching a relevant guide.

Trigger:

Use this workflow whenever the user reports that their dashboard or charts aren't showing data, or asks why their data is missing.

Prompt:

First, ask the user whether they recently set up a new project or whether data was previously showing.

Then, check the user's account and plan status using @Get user account information.

If the account is on a free plan, let the user know that data may take up to 24 hours to appear and launch @Empty dashboard troubleshooting guide.

If the account is on a paid plan, ask whether they've verified that events are being sent using the Amplitude Debugger, then launch @Instrumentation troubleshooting guide.

Exit the workflow and return to AI Assistant chat with a brief summary of what you recommended.

Tools

Extend AI Assistant with custom tools backed by your own APIs. For each tool you define:

1. A name and description (written in plain language as if you're describing it to a human agent).
2. One or more input parameters, with types and validations.
3. A backend endpoint that receives the request.

AI Assistant then uses these tools to:

• Look up internal information (for example, account status, plan limits).
• Perform actions (for example, create a support ticket, start an internal workflow).
• Provide more personalized, contextual help.

To create a new tool:

1. Navigate to AI Assistant > Automations > Tools and click Create.
2. Enter details about the tool. Provide a name and description written in plain language, as if you're explaining the tool to a human agent. Optionally, enable a confirmation dialog that prompts the user before the tool runs.
3. Define the API call:
   • Select the HTTP method: GET, POST, PUT, PATCH, or DELETE.
   • Add the API endpoint: Enter the full URL on your server that receives the request. The endpoint must be a valid http or https URL.
   • Define the request parameters: Parameters are the inputs that AI Assistant can fill in when it calls the tool. For each parameter, set the Type (for example, string or number), Key, Required state, and Description. Write the description in plain language so AI Assistant understands when and how to use the parameter.
   • Wire parameters into the endpoint or body: Parameters aren't automatically appended to the URL or request body. To use a parameter, reference it with a placeholder in the endpoint URL, request body, or both. Any parameter you define but don't reference with a placeholder is ignored.
   • Define the request body: Supply valid JSON. The body must be a JSON object (not an array). Reference parameters in the body with placeholders. Leave the body empty for GET requests that pass all parameters through the URL.
   • Define the request headers: Supply valid JSON. For example, {"Content-Type": "application/json"} or {"Authorization": "Bearer YOUR_TOKEN"}.
4. Test the response. Enter values for each parameter you defined, send a test request to the endpoint, and review the response.

Example GET request

A simple GET request that passes a parameter as a query string value:

• Method: GET
• Endpoint: https://weather.example.com/search?city=
• Parameters: city (string, required) — "The city to look up the weather for."
• Body: leave empty

For GET requests, embed parameters directly in the endpoint URL. For example, use ?city=, not just ?city — the placeholder is what tells AI Assistant where to insert the value.

POST request with URL and body parameters

You can mix parameter placeholders in both the URL path and the request body, and include static values alongside them:

• Method: POST
• Endpoint: https://api.example.com/update_order/
• Parameters: order_id (string, required); shipping (string, required) — "The shipping method for the order, for example 'standard' or 'express'."
• Body:

{
  "shipping": "",
  "notify_customer": true
}

• Headers:

{
  "Content-Type": "application/json",
  "Authorization": "Bearer YOUR_TOKEN"
}

In this example, is wired into the URL path, is wired into the body, and "notify_customer": true is a static value that doesn't come from a parameter.

Key things to know about parameters

• Parameters define what AI Assistant is allowed to provide when it calls the tool. They don't automatically generate query strings, form fields, or body content.
• You must wire each parameter into the endpoint URL or body yourself using placeholders.
• AI Assistant validates required parameters before the tool runs. If a required parameter is missing, AI Assistant won't send the request.
• The tool sends all requests as JSON, so custom tools work best with JSON APIs.

User approvals and safety

For tools that affect user data or perform sensitive actions, you can require user approval:

• AI Assistant explains what it wants to do in plain language.
• The user confirms or rejects the action.

As you add tools:

• Start with read-only tools to minimize risk.
• Limit write-capable tools to well-scoped, high-value actions.
• Use X-ray to audit when, how, and why AI Assistant called tools.

Style AI Assistant and Resource Center

AI Assistant uses the same theme system as Guides & Surveys and Resource Center, which lets you maintain a consistent in-app brand.

Choose or create a theme

1. Navigate to AI Assistant > Themes.
2. Select an existing theme to edit, or click Create Theme.

A theme typically controls:

• Primary and secondary colors.
• Fonts and typography.
• Button styles.
• Panel backgrounds and borders.

Apply themes to Resource Center and AI Assistant

In the Resource Center configuration:

1. Select the theme you want the widget to use.
2. Confirm that the theme's components for Launcher, Panel / header, List items, and Tabs (Chat / Resources) (when available) look the way you want.

AI Assistant shares the same base theme but may expose additional configuration for:

• Chat tab styling such as chat bubble colors and background.
• AI Assistant name and avatar.
• Icon used for the launcher.