AgentFlow provides a guided setup experience that helps you create, configure, test, and deploy an AI agent without manually building the logic in Flow Builder, with additional controls to personalize responses, guide knowledge usage, and manage deployment more precisely.

The setup process uses a 3-step wizard that walks you through creating the agent, configuring its knowledge and behavior, testing responses, and deploying it to your selected channels.

This workflow helps teams:

• Set up AI agents faster using use-case templates
• Test responses before going live
• Understand how the agent generates answers through built-in debugging tools

Typical use cases include:

• Commerce: product advisor or sales assistant
• Service booking: answering enquiries and guiding customers to complete bookings
• Education: helping users explore programs or services

The setup process includes the following steps:

1. Create the AI agent
2. Configure the agent and test responses
3. Deploy the agent to channels

Accessing the “Create AI agent” page

You can follow the steps below to access the “Create AI agent” page:

1. Click the icon on the left-sided panel to go to the SleekFlow AI page
2. Click on “AgentFlow” in the top navigation bar
   

Step 1: Create the AI agent

There are 2 ways to create your AI agent

1. If this is the first time you create an AI agent, click “Start now” to get started
   
2. If you have already created AI agents, click Create new agent in the top right corner of the page.
   

The “Create” step allows you to choose a template and define basic information for your AI agent.

Choose a template

Start by selecting a template that matches your intended use case.

Available options include templates designed for common scenarios such as:

• e-Commerce
• Service booking
• Education
• Legal intake agent

Templates provide a structured starting point with predefined Playbook instructions that you can modify later. You can also choose “Custom” and set up your AI agent from scratch.

Name your agent

Next, provide the basic details for the agent.

Required field:

• Agent name: Used internally to identify the agent. This name is not visible to customers.

Optional field:

• Description: A short internal description to help your team understand the agent’s purpose.

 If you have selected a template to start with, these fields will be prefilled. You can edit them to make sure the AI agent name and description is easy for internal team members to recognize.

Click “Create” at the top right corner of the page to continue.

Step 2: Configure and test your AI agent

In this step, you define how your AI agent behaves and provide the information it uses to answer customer enquiries.

You will:

• Add knowledge sources
• Configure response behavior
• Define conversation instructions in the Playbook
• Enable lead scoring (optional)
• Test the agent using the Testing Playground

The Testing Playground allows you to simulate conversations and evaluate responses before deploying the agent.

Set up the agent-level Knowledge Base

Each AgentFlow AI agent has its own Knowledge Base. This controls what information the agent can reference when replying to customers.

Add or assign sources that are relevant to the agent’s purpose, such as product information, policies, or answers to common customer questions.

You must add at least one knowledge source before testing the agent.

To manage the agent’s knowledge sources, click “Manage knowledge” in the “Configure” step.

For instructions on uploading files, indexing web pages, adding live web search sources, creating custom answers, and assigning existing sources, see Manage Your AI Knowledge Base.

Source processing and article generation happen in the background. You can continue configuring your agent while processing is in progress.

Before evaluating the agent’s responses, check that the sources contain the information it needs. For instructions on reviewing source statuses, checking extraction quality, and resolving processing or article-generation failures, see Manage Your AI Knowledge Base.

Allow the agent to reference contact properties

You can allow the AI agent to access selected contact properties to personalize responses.

To configure contact property access:

1. Under Customer data access, click “Add property”
2. Select the contact properties the agent can reference
3. Click “Save properties”

The agent can then use this information when generating replies.

Set response priority

Response priority controls how the agent balances response speed and answer quality.

You can choose from three options:

• Ultra fast: prioritizes speed for quick responses
• Balanced: balances response speed and accuracy
• High quality: prioritizes more detailed and accurate responses

Write the Playbook

The Playbook defines how the AI agent should conduct conversations with customers.

Use the Playbook editor to write instructions that guide how the agent responds in different scenarios and what actions it can take during a conversation.

When writing the Playbook, you may want to include instructions such as:

• Tone and communication style: Define how the agent should communicate with customers (for example: friendly, professional, concise).
• General instructions: Describe the overall role of the AI agent and how it should assist customers.
• What the agent should avoid: Define any restrictions, such as avoiding speculation, not providing unsupported information, or escalating certain requests.
• Scenario handling: Provide instructions for how the agent should respond in specific situations, such as refund requests, product inquiries, or booking questions. This can also include what information the agent should collect in each scenario, such as order numbers, product preferences, or booking details.

If you selected a template in Step 1, the Playbook will already include predefined instructions that you can review and modify.

Using Playbook commands

The Playbook editor supports slash (/) commands, which allow you to insert structured actions directly into the conversation flow.

These commands allow the AI agent to perform actions such as:

• Applying labels
• Triggering integrations
• Calling custom APIs
• Performing other configured actions

To insert a command:

1. Type / in the Playbook editor.
2. Select the action you want to insert from the command list.

The Playbook defines how the AI agent should conduct conversations with customers.

Reference a specific knowledge source with /source

You can use /source in the Playbook to direct the AI agent to use a specific knowledge base file for certain instructions.

To add a source as an action:

1. In the Playbook editor, enter /source
2. A list of existing data sources will appear
   
3. Select data source in the modal
   
4. Click “Insert”

This is useful when:

• The agent must rely on a specific document in a specific scenario
• You want more controlled and consistent answers

Example:

Use a refund policy file to ensure all refund-related responses follow the same source.

Using Shopify commands in the Playbook

If your company has Shopify connected in SleekFlow, you can add Shopify commands in the Playbook to define when the AI agent should perform Shopify-related actions.

This allows you to:

• Control when the agent retrieves product information
• Guide how the agent handles cart-related actions
• Ensure consistent handling of store-related queries

Available Shopify actions

The AI agent can use the following Shopify actions during conversations:

Calculate lead score (optional)

You can enable “Calculate lead score” to evaluate customer intent during conversations.

When enabled, the agent assigns a score based on the criteria you define.

To configure lead scoring:

1. Toggle Calculate lead score ON
2. Click on the icon to configure scoring rules
3. A side panel will appear
   
4. In the side panel, the fields in the side panel are pre-filled with suggested criteria. You can edit the criteria and adjust the weights based on your business needs. You can define scoring criteria using the following fields:
   1. Lead score weight (%): Enter the weight for each criterion. The weight determines how much the criterion contributes to the final lead score.
   2. Criteria: Describe the signals the AI agent should evaluate when assigning a score. For example, whether the lead is asking about product features, comparing options, or showing signs of purchase intent.
   3. You can also click “Add criteria” to include additional scoring conditions.

If you wish to revert back to the pre-filled content, click “Use default” at the top right corner.

After configuring the criteria, click “Save” at the top right corner to apply the lead scoring rules.

Once enabled, the AI agent will automatically calculate lead scores during conversations based on the defined criteria. Lead scoring helps your team identify high-potential leads and prioritize follow-up actions.

Test your agent in the Testing Playground

The Testing Playground allows you to test your AI agent and review how responses are generated before deployment.

The testing panel appears on the right side of the setup page.

Testing becomes available after:

• You have added at least one source to the agent-level Knowledge Base
• You have configured the Playbook

If sources are still processing, chat testing may still be available. The agent can reference the uploaded source material while generated articles are being prepared. Once the generated articles are ready, responses may reference the generated articles instead.

Test responses via chat

Once you have added at least 1 knowledge source and have configured the Playbook, you can use the chat testing interface to simulate conversations with the AI agent.

To test responses:

1. Click “Test responses via chat.”
2. Use the suggested conversation starters, or enter your own message.
   
3. In the chat testing view, you can see the following:
   1. AI agent response: 
      1. The main response card shows how the AI agent would answer the customer’s message.
      2. This allows you to check whether the response is accurate, clear, and aligned with the behavior defined in the Playbook.
   2. Actions taken
      1. The Actions taken section shows the steps the AI agent performed before generating the response.
      2. This may include actions such as:
         1. Searching the knowledge base
         2. Calling an integration or API
         3. Calculating lead score, if enabled
      3. This section helps you understand how the AI agent arrived at its answer.
   3. Sources
      1. The Sources section shows the knowledge sources referenced in the response.
      2. For each source, you can see the file name and a short text snippet used by the AI agent. This helps you verify whether the response is grounded in the correct source material.
   4. Response feedback
      1. Below the response, you can indicate whether the generated reply matches your expectation.
      2. Use this to quickly review whether the response quality is acceptable during testing.
   5. Suggested follow-up questions
      1. The chat panel may also display suggested follow-up questions based on the conversation.
      2. You can click these suggestions to continue testing how the AI agent responds to related customer questions.
   6. Message input field
      1. At the bottom of the panel, you can enter additional messages to continue the conversation and test different scenarios.
      2. This allows you to simulate multi-turn conversations and evaluate how the AI agent responds across the full interaction.

Run response batch test

Response batch test allows you to evaluate how confidently the AI agent answers a set of auto-generated test questions based on your knowledge sources.

Instead of testing one message at a time, batch testing creates multiple questions from your uploaded knowledge and runs them through the AI agent. This helps you quickly identify what the agent answers well and which areas may need improvement before deployment.

To run a response batch test:

1. Click “Run response batch test” in the Test playground
2. Select the testing language.
   
3. Click “Run performance test”.
4. After the test is complete, you can review the results in the following ways:
   1. Overall confidence score
      1. At the top of the results page, you can see the overall confidence score for the batch test. 
      2. This summary shows how well the AI agent performed across all generated questions, including the number of responses that fall into each confidence category.
   2. Confidence categories
      1. The test results are grouped into confidence categories, such as:
         1. Excellent
         2. Needs attention
      2. This helps you quickly identify which answers are performing well and which ones may need improvement.
   3. Individual test results
      1. Each test result shows:
         1. The generated test question
         2. The AI agent’s answer
         3. The confidence score for that answer
         4. The knowledge source used to generate the question
      2. This allows you to review how the AI agent responded to each question and whether the answer is supported by the expected source material.

Once you have completed the configuration and testing your AI agent, click “Save” at the top right corner to go to Step 3.

Step 3: Deploy your AI agent

After configuring and testing your AI agent, you can deploy it so it begins responding to customer conversations.

In this step, you will:

1. Select the channel where the AI agent should respond
2. Choose whether replies are delivered as one message or several.
3. Optionally define when the agent is allowed to reply
4. Configure exit conditions that determine when the agent stops responding. In this step, you can also add follow-up action on how you would like the conversation to be handled after the AI agent exits the conversation.

Select channel

First, choose the channel where the AI agent will receive and respond to messages.

To configure this:

1. In the Deploy step, locate the Channel section.
   
2. Select the channel from the dropdown list.
3. This defines the channel where the AI agent will handle incoming messages.

Configure reply delivery

The Reply delivery section controls whether the agent sends a reply as one message or several chat bubbles.

Split replies into separate messages is turned on by default for new deployments.

• Turn it on to send replies of 150 characters or more as one message per paragraph the agent writes. There is no maximum number of messages per reply.
• Turn it off to send the reply as a single message.
• Replies shorter than 150 characters are sent as a single message regardless of this setting.

Turning splitting off changes delivery only. The agent’s Knowledge, Playbook, and response content stay the same.

Configure delivery for each deployment

The setting is available for every deployment, including before you select a channel and when no WhatsApp channel is selected.

One setting applies to all channels in the deployment. For example, a deployment with WhatsApp and Instagram uses the same reply delivery setting for both channels.

To use different delivery settings for different channels, create separate deployments. There is no per-channel setting within a deployment or workspace-level default.

Review WhatsApp message billing

From October 1, 2026, each WhatsApp message is billed separately, so a split reply costs more than the same reply sent as one message. See WhatsApp Business API Pricing for details.

When splitting is on and the deployment includes a WhatsApp channel, click See example to compare delivery options.

For example, the same reply can be delivered as follows:

The billing information is available when you turn splitting on for a deployment with WhatsApp, or add WhatsApp to a deployment that already has splitting on.

You can publish the deployment with splitting enabled. The billing information does not require confirmation or block publishing.

When splitting is off or the deployment has no WhatsApp channel, the billing information is hidden. The reply delivery setting remains available.

When changes take effect

Changes apply to conversations that reach the agent after you publish the deployment. Conversations the agent is already handling keep their existing reply delivery setting.

Set specific reply time (optional)

You can control when the AI agent is allowed to reply to customers by enabling Set specific reply time.

To configure reply time:

1. Toggle Set specific reply time ON.
2. Select the days of the week when the AI agent should be active.
3. Set the From and To time.

This allows you to control when the AI agent is allowed to respond. For example, you may choose to activate the AI agent only outside of business hours.

The system supports overnight reply windows, such as 6:00 PM to 9:00 AM.

Set exit conditions

The “When to exit” section allows you to define conditions that determine when the AI agent should stop responding in a conversation.

By default, the AI agent includes predefined exit conditions for common conversation scenarios. These conditions determine when the AI agent should stop responding in a conversation.

The default exit conditions include:

• Confidence is low: when the AI agent’s response confidence is low
• Time out: when the conversation becomes inactive for a period of time
• Human agent takes over: when a human agent joins the conversation
• System error: when the AI agent encounters a system error

You can review and edit these default exit conditions based on your business needs.

To configure an exit condition:

1. Locate the “When to exit” section.
2. Click open the exit condition
3. Define the trigger criteria.
4. Optionally add a follow-up action after the AI agent exits. Follow-up actions define what happens next after the AI agent stops responding. In the deployment page, you can configure a limited set of follow-up actions, such as:
   1. Assign the conversation to a team member
   2. Add labels to the conversation for tracking or categorization
   3. If you need more advanced exit logic or additional actions, you can continue editing the generated flow later in Flow Builder.
5. You can click “Add exit condition” to configure additional exit rules.

These conditions help ensure that conversations are appropriately handed over or ended when the AI agent is no longer able to assist effectively.

Define target audience for deployment

You can limit who can interact with the AI agent during deployment. This helps you test the agent with a smaller audience before rolling it out more broadly.

To define a target audience:

1. Under Target Audience, select “Custom audience”
   
2. Click “Add condition”
3. Choose a condition type:
   1. Phone number
   2. Keyword
   3. Contact label
4. Enter the condition values
   

Deploy the AI agent

Once the channel and optional settings are configured:

1. Click “Save” and deploy in the top right corner.

After deployment:

• The AI agent becomes active
• The agent will start responding to conversations on the selected channel
• You will be redirected back to the AI agent list
• The system automatically creates a corresponding flow in Flow Builder

The generated flow can be used for more advanced configuration, such as refining exit logic or adding additional actions beyond what is available in the deployment page.

You can return later to update the agent configuration or adjust the deployment settings if needed.

Manage an existing AI agent

After creating and deploying an AI agent, you can return at any time to update its configuration, re-test responses, or adjust deployment settings.

To manage an existing AI agent:

1. Click on the icon on the left navigation bar to go to SleekFlow AI page
2. Under the “AgentFlow” tab, locate the agent you want to manage
3. Within the card, you can:
   1. Edit the AI agent by clicking on the “Edit” button
   2. Delete the AI agent by clicking on the icon

Update configuration and testing

In the “Configure” step, you can:

• Add, remove, or update sources in the agent-level Knowledge Base
• Update the Playbook instructions
• Adjust response priority and behavior settings
• Update lead scoring settings, if enabled
• Run chat testing and response batch tests again before saving changes

Manage exit conditions

In the “Exit” tab, you can manage the AI agent’s global exit conditions and follow-up actions.

You can:

• Add a new exit condition
• Edit an existing exit condition
• Delete an existing exit condition. Please note that when you delete an exit condition, it is removed from all of the existing  flows.
• Update follow-up actions linked to exit handling

Other updates, such as editing or deleting existing exit conditions, apply without redeployment.

Update deployment settings

In the Deploy tab, you can view the AI agent’s current deployments and create or update flows based on the deployment type.

A deployment with a WhatsApp channel and reply splitting enabled includes Multiple messages per reply alongside its audience and active-hours summary.

The indicator is absent when splitting is off or the deployment has no WhatsApp channel. An absent indicator is expected and does not mean there is an error.

Create a new flow

To deploy the AI agent to an additional flow, create a new flow in the Deploy tab and complete the deployment setup.

Depending on the deployment type, you can configure settings such as:

• Connected channel
• Reply schedule

After setting up the flow, click Save and deploy to apply the deployment.

Update an existing flow

For an existing deployment that is not managed in Flow Builder, you can:

• Change the connected channel
• Update the reply schedule

After making changes, click “Save and deploy” to apply the updates.

Fields that cannot be edited

The following fields can only be set when the AI agent is first created:

• Agent name
• Description

If you need to use a different agent name or description, create a new AI agent instead.