This document provides a comprehensive guide on configuring an agent within the Carrick Health system. It outlines the necessary steps to set up an agent with a knowledge base, ensuring it can effectively assist contacts with information and booking inquiries.

Once an agent has been created, it is essential to configure it with a knowledge base before it can assist contacts effectively. After creating an agent, you will be automatically directed to the configuration screen. If you are not automatically redirected, you can navigate to the "Workflows" tab in the left-hand sidebar, scroll to the bottom, click "Agents," and select an existing agent to configure.

The first screen you will encounter is the "General" tab, which provides a high-level overview of your agent's capabilities.

• Agent Identity: This section contains the basic information you entered when creating your agent, along with a new "Conversation Flow" option. This option determines whether the agent should follow a specific conversation template. If left blank, the agent will respond to questions without a structured flow.

• Default Messages: This section includes your agent's replies for transfers, errors, escalations, and guardrails. Carrick has configured all agents with sensible defaults, so this section can generally be left unchanged. Additionally, you can select default assignees for agent handoffs, which are the users to whom the agent will escalate conversations if it cannot resolve the issue independently.

The "Clinic & Knowledge" tab is where you will upload files to your agent's knowledge base, which it will utilize to answer questions from your contacts.

This area contains the name of your clinic and its basic information, including the clinic's name, address, and contact information.

This section allows you to determine how your agent retrieves information and what actions it should take when it is uncertain about how to respond to a question. It is generally recommended that every agent has an enabled knowledge base, with the specific actions determined on a clinic-by-clinic basis.

To upload documents to an agent with an enabled knowledge base, click "Choose File" to select a file. Acceptable formats include PDF, Word (DOCX), plain text, Markdown, or CSV, with a maximum file size of 10 MB. Choose a title that is easily understandable for your users and optionally provide a short description of the file's contents. After entering this information, click "Upload" to send your file to the knowledge base.

Uploaded files should contain detailed information about your clinic, including hours of operation (including temporary closures), physician information, insurance and billing details, and answers to commonly asked questions.

This section contains your uploaded documents. After a file is uploaded, it will take a moment to process before it is ready for publication. If the processing seems to be taking too long, click the "Refresh" button. Once processed, you can preview the contents of the file. If everything is satisfactory, click "Publish" to enable your agent to reference this document when responding to contacts.

The "Behavior" section allows you to configure your agent's personality, topics to avoid, suggestions, situations for handoffs to team members, and other general information.

In this section, you can choose your agent's tone and speaking style. In most cases, both can be left as "Auto," allowing Carrick to automatically select a style based on the conversation context and the contact's replies.

Here, you can specify any topics that your agent should avoid discussing, typically for safety or privacy reasons. Your agent can have up to 10 forbidden topics at any time, which may include detailed billing information, personal information about staff, and confidential medical information.

This section allows you to define situations in which your agent should always transfer a conversation to a staff member. Common scenarios include when a contact expresses significant frustration, discusses forbidden topics, or directly requests to speak with a team member.

In this area, you can enter topics that your agent should proactively mention without being prompted. A common example would be providing links to direct contacts to your online portal and forms.

This section allows you to input information specific to your clinic, such as confirming a patient's preferred pharmacy before discussing prescriptions or avoiding promises of specific appointment times when they are not guaranteed.

The "Actions" section is where you can configure the actions your agent is permitted to perform, including appointment lookups, bookings, and topic rules.

If enabled, this feature allows your agent to look up a contact's appointments without needing to transfer the conversation to a team member. You can set the maximum time window for the agent's search, specifying the metric (minutes, hours, days, weeks).

If enabled, your agent will be able to search for open slots with physicians and book appointments for contacts without needing to pass the conversation to a team member. Similar to appointment lookups, you will need to set a maximum time window for the agent's search and select which services and physicians your agent can utilize for booking appointments.

This section allows you to create custom topic rules for your agent to follow. Basic rules may include human handoffs, billing inquiries, and rescheduling assistance. Each rule requires a name, description, minimum confidence level, positive and negative examples for triggering the rule, and actions that should occur when the rule is activated.

This section allows you to test your agent's knowledge and behavior before making it live.

In this area, you can ask your agent questions to test its knowledge base and responses. Enter a message that a contact may send and click "See Agent Reply" to view the agent's response. You can continue the conversation to test it further.

Similar to the previous section, this area allows you to ask questions to see your agent's replies and the sources of its information. This feature is more suited for one-off inquiries rather than continuous conversations.