how to

Create agents using the conversational builder

Arkus allows you to create agents through an interactive conversation. Instead of configuring everything manually, you can describe what you want the agent to do, and it will guide you through the process. During the conversation, the builder asks a few key questions to understand how the agent should behave. Based on your answers, it generates the instructions and configuration needed to build the agent.

What information will you be asked?

To create an agent, the conversation focuses on a few essential elements. These questions help define how the agent should operate within your workflow.

1. What the agent should do

The first step is to describe the purpose of the agent. Explain the task or problem the agent should help solve. This helps the builder understand the main objective of the agent and generate the appropriate instructions.

For example, you might describe tasks such as:

  • answering questions based on a guideline
  • analyzing documents
  • summarizing information
  • identifying signals in patient reports

The clearer the objective, the easier it is for the builder to generate the correct configuration.

2. Who will use the agent

Next, the builder will ask who the agent is intended for. This helps define how the agent should communicate and structure its responses.

Examples of users may include:

  • clinicians
  • patients
  • internal teams
  • researchers
  • customer support teams

Understanding the audience allows the builder to adjust the tone, level of detail, and type of response produced by the agent.

3. Tools or external services

If the agent needs to interact with external tools or services, you can specify them during the conversation.

Examples may include:

  • email services
  • document platforms
  • databases
  • APIs
  • scheduling systems

These tools allow the agent to perform actions beyond simple text analysis. If the agent does not need any external tools, you can skip this step.

4. Knowledge sources

You can also specify the knowledge sources the agent should rely on. These may include documents, guidelines, internal documentation, or other reference materials.

For example:

  • clinical guidelines
  • internal policies
  • product documentation
  • research papers
  • uploaded PDFs

Providing knowledge sources allows the agent to generate responses based on trusted information.

5. Task steps

Finally, the builder may ask how the agent should approach its task. This helps define the reasoning process the agent should follow when handling a request.

For example:

  • receive the user question
  • consult the knowledge base
  • extract relevant information
  • generate the response

Defining these steps helps create more structured and reliable agent behavior.

Agent scope confirmation

Before creating the agent, the builder summarizes the information collected during the conversation. This allows you to review the configuration and make adjustments if necessary.

You will see a summary similar to the following:

  • Purpose: Consult a simple guideline and answer patient questions.
  • Users: Patients
  • Tools: None yet
  • Data sources: Guideline PDFs
  • Steps: Receive the patient’s question, consult the guideline, and answer accordingly. If the answer is not in the PDFs, explain that the information is not available.

The builder will then ask if it can go ahead and create your agent. At this stage, you can:

  • confirm the configuration and create the agent
  • refine the instructions
  • add tools or knowledge sources
  • adjust the task steps

Once you confirm the configuration, the builder generates the agent and prepares it for use.