After creating your AI agent, review and configure how it should behave, what information it can use, and what it can do during customer conversations.

Your agent’s setup starts with two core areas that are required for every agent:

• “Rules” define how the agent should behave across conversations.
• “Knowledge” gives the agent the information it can use when responding to customers.

If your agent needs to follow a structured workflow for a specific scenario, you can also add “Procedures”. Procedures are optional and define the steps the agent should follow, such as what information to collect, what actions to take, and when to stop or escalate.

You can also configure additional settings to control what the agent can do and how it handles conversations:

• Actions let the agent perform supported tasks during a conversation.
• Context and response controls what customer information the agent can access and how it responds.
• Exit conditions define when the agent should stop handling a conversation.

If you created the agent using conversational setup, AgentFlow’s Co-pilot can generate the initial Rules and, when relevant to your use case, Procedures. Review and refine the generated content before testing your agent.

Before you begin

Before configuring your agent:

• Create an AI agent using Create an AI agent with conversational setup.
• Make sure you have access to the SleekFlow AI module.

To open an existing agent, go to SleekFlow AI and select the agent you want to configure.

Edit the agent name and playbook

The playbook defines how your agent should behave and what it should do during customer conversations.

You can update the agent name and refine the instructions generated during conversational setup.

Rename your agent

The agent name helps your team identify the agent within SleekFlow. It is used internally and is not shown to customers.

To rename the agent, click the agent name at the top of the page and edit it directly.

Choose a clear name that reflects the agent’s purpose or use case.

Configure Rules

Rules define how the agent should behave across all conversations.

Use Rules to document:

• The agent’s responsibilities and boundaries
• How it should respond to customers
• Its tone of voice
• Information it should or should not provide
• Business context it should follow

For example:

Answer questions about store locations and opening hours using assigned Knowledge. Do not guess information that is not available. If the requested information cannot be verified, tell the customer and hand the conversation over when appropriate.

Review any Rules generated by Co-pilot and edit them to make sure they accurately reflect your business requirements.

Configure Procedures

Procedures define the steps your AI agent should follow for a specific scenario or workflow. They are optional. Use them when the agent needs to follow a structured sequence instead of relying only on general Rules and Knowledge.

A procedure can tell the agent what information to collect, which steps or actions to perform, and when to continue, stop, escalate, or exit. For example, a lead qualification procedure could instruct the agent to ask about the customer’s product interest, budget, and purchase timeline, calculate a lead score, and hand over the conversation when the lead meets your criteria.

Review procedures generated by Co-pilot and remove, edit, or add instructions as needed.

Add a scenario and its steps

Follow the steps below to configure scenarios: 

1. Expand “Procedure”
   
2. Click “Add scenario”
3. In the Applies when field, describe the situation that should trigger the scenario. For example: “The customer asks about an order status.”
4.  
5. Click “Add step” and describe what the agent should do. Type / in a step to insert an action.
6. If you need another workflow, click “Add scenario” and configure its trigger and steps.

Insert and configure actions

Actions let your agent perform supported tasks during a conversation. Insert an action at the step where you want the agent to perform it. Once added, it appears as an action chip within the procedure.

To insert an action:

1. Click Insert action, or type / in the procedure step.
2. Select the action you want to add.
3. Configure the action if required.

An action chip may be ready to use or marked as needing configuration. You can still save or publish the agent if an action needs setup, but AgentFlow may skip that action when it runs until the required configuration is complete.

Available actions

The actions shown in the command menu depend on your agent setup and connected integrations.

Use /source when a specific step should rely on selected Knowledge sources. Shopify actions appear after Shopify is connected, and Custom API actions appear after the API integration is configured. See Shopify integration and Custom API integrations in SleekFlow AI for setup instructions.

Set up Knowledge base

Knowledge provides the information your agent can reference when answering customers. Your workspace has a global Knowledge library, and each agent uses the sources assigned to it. Sources you add while configuring an agent are added to the global library and assigned to that agent.

If website information was used during conversational setup, sources may already be assigned. Review them before adding more.

For instructions on choosing, adding, assigning, and managing sources, see Manage Your AI Knowledge Base.

Set up Knowledge base with Co-pilot

Use Co-pilot to help identify and add Knowledge your agent may need. Describe the information your agent should know or the types of questions it needs to answer.

Co-pilot can:

• Add website sources
• Create Custom Q&A entries
• Assign existing documents from the global Knowledge library

Co-pilot cannot upload images or other files from the setup chat. To upload a supported file, add it manually from the agent’s Knowledge settings.

Review all sources and Custom Q&A entries suggested or added by Co-pilot. Make sure they are relevant, accurate, and appropriate for the agent.

Choose and add Knowledge sources

On the “Knowledge Base” page, you can:

• Open “Assigned” to review sources currently assigned to the agent.
• Open “Library” to find and assign existing sources from your workspace’s global Knowledge library.
• Click + to create and assign a new source.

Choose sources that are relevant to your agent’s purpose. For source types, supported formats, limits, and instructions on assigning or removing sources, see Manage Your AI Knowledge Base.

Check what was imported

Check source status and extraction quality before testing your agent. If the agent cannot answer a question, confirm that the relevant content was imported successfully before changing its playbook.

For instructions on checking extraction quality, resolving processing failures, and retrying incomplete article generation, see Manage Your AI Knowledge Base.

Configure Context and response

Context and response controls which customer data the agent can read and how it generates replies.

The page contains two sections:

• Data access
• Response engine

Configure Data access

Data access controls which properties the agent can read during customer conversations. You can grant access to contact properties and custom object properties.

To configure Data access:

1. Under Objects, select “Contact” or a custom object.
2. Search for the properties you want the agent to access.
   
3. Select each property the agent needs. To grant access to every property for the selected object, click “Select all”.
4. Repeat these steps for any other objects the agent needs to access.

The number beside an object shows how many of its properties are selected. Click “Clear all” to remove access to all selected properties for the current object.

Only select information the agent needs for its role. For example, a support agent could use a customer’s subscription plan or order status to provide more relevant assistance.

If you do not select a property, the agent cannot read its value through “Data access”.

Configure the Response engine

The Response engine controls how the agent generates and paces replies.

Calculate lead score

Enable “Calculate lead score” when you want the agent to evaluate a contact’s sales potential during a conversation. 

The agent scores each conversation from 0 to 100 based on intent signals and the criteria you define.

When enabled, the section contains pre-filled scoring criteria. Each criterion has:

• A percentage that determines how much it contributes to the lead score
• A description of the intent signals the agent should evaluate

To configure lead scoring:

1. Review the pre-filled criteria.
   
2. Edit each criterion and its percentage to match your qualification requirements.
3. Click “Add criteria” to add another scoring criterion.
4. To remove a criterion, click its delete icon.
5. Make sure the total weight equals 100%.

New criteria have a weight of 0% until you update them. Click “Use default” to restore the pre-filled criteria and weights.

Enable fast-track simple questions

Turn on Enable fast-track simple questions to use the fast-track response path for simple questions that the agent can answer directly from its Knowledge Base.

Fast-track responses do not consume AI credits. Other requests follow the standard response process and consume AI credits.

Turn off this setting if you want all requests to follow the standard response process.

Configure when the agent should exit

When to exit defines when the agent should stop handling a conversation.

AgentFlow provides three pre-filled exit conditions:

• Timeout
• Human Escalation
• Customer goal completed

Review each condition and update its settings for your use case. You can also use AI Assist or add an exit condition manually.

Review the pre-filled exit conditions

Timeout

Timeout ends the conversation after the contact has not responded for a specified period.

To configure it:

1. Expand “Timeout”.
   
2. Under Time out after, set how long the agent should wait.
3. Select the time unit.

The “Time elapsed without contact’s response” trigger is pre-filled and cannot be changed.

Human Escalation

Human Escalation ends the agent’s handling when the customer asks for human assistance or meets the escalation criteria you define.

To configure it:

1. Expand “Human Escalation”.
   
2. Review the “Human/escalation requested” trigger.
3. In the “Describe” field, explain which requests or situations should cause the agent to exit. For example, specify that the agent should exit when a customer asks to speak with a team member.
4. To change the condition’s internal name, click the edit icon beside “Human Escalation”.

Customer goal completed

Customer goal completed ends the agent’s handling when the customer’s goal or the agent’s task has been completed.

To configure it:

1. Expand “Customer goal completed”.
   
2. Review the “Goal achieved” trigger.
3. In the “Describe” field, define what must happen for the goal to be considered complete. For example, the customer has received the requested information and no longer needs help from the agent.
4. To change the condition’s internal name, click the edit icon beside Customer goal completed.

Make each description specific enough for the agent to determine consistently when it should exit.

Set up exit conditions with AI Assist

Use AI Assist when you want Co-pilot to guide you through creating another exit condition.

To set up an exit condition with AI Assist:

1. Click “AI Assist”. Co-pilot opens on the left and asks questions about when the agent should exit.
   
2. Answer each question by selecting a suggested option. To provide a different response, select “Other” and enter your answer.
3. Click “Submit” to send each answer and continue through the guided setup.
4. Review the exit condition created by Co-pilot and edit it if needed.

Create an exit condition manually

Add another exit condition when the pre-filled conditions do not cover a situation in which the agent should stop handling the conversation.

To create an exit condition:

1. Click Add exit condition.
   
2. Click the edit icon beside the condition name and enter a clear internal name.
3. Select a “Trigger”.
4. In the “Describe” field, explain when the agent should exit.
5. Review the condition to make sure it does not overlap unnecessarily with an existing condition.

The following triggers are available:

Describe the exit criteria clearly and specifically so the agent can apply the condition consistently.

In addition to the triggers used by the pre-filled conditions, you can select “Detected sentiment”. Use this trigger when the agent should exit after detecting a particular customer attitude or emotional state.

For example, configure the agent to exit when:

• The customer appears angry or highly frustrated.
• The customer repeatedly expresses dissatisfaction.
• The customer shows distress or concern that requires human support.
• The conversation becomes sensitive and should be handled by a team member.

Describe the relevant signals clearly. Avoid broad instructions such as “when the customer is unhappy,” which may cause the agent to exit conversations unnecessarily.

Manage exit conditions

Click an exit condition to expand or collapse its settings.

Depending on the condition, you can:

• Click the  icon to rename it.
• Change its trigger.
• Update its description or trigger-specific settings.
• Click “Delete exit condition” to remove it.

Deleting an exit condition removes the rule that tells the agent to exit in that situation. Review the remaining conditions to make sure the agent can still exit or escalate when needed.

Save and apply configuration changes

Changes are not saved automatically.

Before your agent has a deployment, click “Save as draft” to save your configuration.

After your agent has a deployment, click “Publish change” to save and apply your updates.

Changes to the agent configuration, such as Rules, Procedures, Knowledge, and response settings, take effect after you publish the changes.

Exit-related changes are handled differently. When publishing changes that affect exits, you can choose how the update should apply to conversations that are already enrolled in the agent.

Review your configuration

Before testing your agent, review the complete configuration.

Check that:

• The Rules accurately describe how the agent should behave.
• Any Procedures contain clear and complete steps.
• Assigned Knowledge is relevant and up to date.
• No assigned source has failed, and any source with extraction gaps still contains the information the agent needs.
• Required action prerequisites are complete.
• Data access includes only the properties the agent needs.
• Response settings match your intended experience.
• Exit conditions cover the situations where the agent should stop handling the conversation.

Next steps

After configuring your agent, test its behavior before deploying it to customers.

See Test and deploy your AI agent to learn how to validate the agent’s responses and deploy it to your messaging channels.

Deployment settings also let you configure what happens after the agent exits a conversation, including supported post-exit actions.