Onboarding Manual
Arkus AI Agent Builder

This guide explains how to create simple agents in the Arkus AI Agent Builder.

An agent is an intelligent assistant that understands requests, reasons about them, and performs actions by using connected tools (for example, Google Calendar) or consulting defined knowledge bases (for example, PDF files).

In this guide, you will learn how to build two examples:

• Scheduling Assistant Agent – to learn how to connect your agent to external tools.
•  Guideline Agent – to learn how to connect your agent to files and use them as a knowledge base.

You can later adapt these same steps to build other agents, such as those supporting clinical workflows, administrative tasks, or research.

1. Understanding Key Terms

Term	Description
Arkus AI Agent Builder	A visual environment where you can create AI assistants by connecting blocks.
Agent	The central “brain” that interprets messages, retrieves information, or uses connected tools.
Node (or Block)	A component that performs a specific task (for example, Chat Input, File Loader, or Google Calendar).
API Key	A secure code that allows your agent to communicate with external services.
Composio	A connector that lets agents interact with external tools such as Google Calendar.
Tool Mode	A setting that lets a connected service or tool be used by an agent, not only read.
Knowledge Base	One or more uploaded documents that the agent uses to find and summarize information.

2. General Tips

• The Arkus AI Agent Builder does not require coding. You build by visually connecting blocks.
• When you see the term “API,” think of it simply as a bridge that connects one system to another.
• It is best to complete one connection at a time.
• Always test your setup using sample data before using real information.
• If something does not connect properly, refresh the page and check your keys and logins.
• Start small (one tool or one file) and expand as you learn

3. Accounts and Access Required

Before starting, make sure you have access to. If not, please create your account.

1. Arkus AI Agent Builder – your workspace for creating and testing agents.
   https://langflow.arkus.ai/
   
2. OpenAI – provides the AI model used to understand and respond to messages.
   https://platform.openai.com/ 
   
3. Composio – used to connect the agent with external tools such as Google Calendar.
   https://platform.composio.dev/
   

4. Getting the required API Keys

a. OpenAI API Key

1. Visit https://platform.openai.com.
2. Log in or create an account.
3. In the left menu, select View API keys.
4. Click Create new secret key.
5. Copy the generated key and store it securely.
6. You will later paste this key into the Agent block.

b. Composio API Key

1. Visit https://composio.dev.
2. Log in or create an account.
3. Go to Settings → General.
4. Copy your API key (not the project ID).
5. You will later paste this key into the Google Calendar block and other tools.
   

5. Building Your First Agent – Scheduling Assistant

This agent will help you schedule meetings by accessing Google Calendar availability. Later, you can replace Google Calendar with other tools to create different types of agents with different functions.

The purpose of this example is to help you learn how to connect your agent to tools.

Step 1. Create a new flow

1. From the Project Page, click + New Flow.
2. Give your flow a clear name, such as Scheduling Assistant.
3. An empty workspace will open, where you can start adding blocks.

Step 2. Add the main blocks

1. In your new flow workspace, click + Add Node in the upper bar.
2. A list of available nodes will appear.
3. You can scroll through or use the search bar to find the ones you need.
4. Add the following blocks to your flow by clicking on each of them:
   
   • Chat Input
   • Agent
   • Google Calendar
   • Chat Output

Each node represents one step in your agent’s workflow logic. Here is what each block does:

Node	Purpose in the workflow	Example
Chat Input	This is where the user types a message or question. It captures the input and sends it to the agent.	A nurse types: “Can you schedule a consultation with Dr. Andersson next week?”
Agent	This is the “brain” of the workflow. It understands the user’s request, decides what action is needed, and communicates with connected tools or files.	The agent interprets the message and asks clarifying questions about time and duration.
Google Calendar	This is a tool node that connects the agent to your Google Calendar through Composio. It allows the agent to look for available times, create new events, or check existing ones.	The agent checks for free slots on Tuesday and Thursday.
Chat Output	This block displays the agent’s response back to the user. It completes the communication loop.	The agent replies: “You have available time on Tuesday at 10:00 or Thursday at 14:30. Which one works for you?”

Together, these four nodes form a complete communication flow:

User message (Chat Input) → Agent reasoning (Agent) → Action or data check (Google Calendar) → Response shown (Chat Output)

Once these nodes are in place, you can continue connecting and configuring them as described in the next steps.

Step 3. Turn Google Calendar into a Tool

1. Click the Google Calendar block.
   
2. Enable Tool Mode.
   After clicking the Calendar block, look for a small switch called “Tool Mode” and activate it.
   This allows the agent to use your calendar as a tool (for example, to check or create events).

Step 4. Connect the blocks

This step defines how information flows between the components.

1. Connect Chat Input → Agent (Input dot).
   This connection allows the agent to receive the user’s message sent through the chat.
   
2. Connect Agent → Chat Output.
   This connection allows the agent to send its responses back to the user
   
3. Connect Google Calendar → Agent (Tools dot).
   This connection gives the agent access to your Google Calendar, enabling it to check availability and create or update events when needed.
   

Step 5. Configure the Agent

1. Click the Agent block.
   
2. Under Settings, paste your OpenAI API key.
   (this should have been set in “Getting the required API Keys” section, but you can also set it now if you haven’t done so previously)
   
3. In the field named Instructions, you will define what your agent should do — its purpose, responsibilities, and behavior.

• Think of the instructions as a short description of your agent’s goal and rules.
• This is where you tell the AI who it is, what task it performs, and how it should communicate.
• You can copy the example below or write your own version to fit your needs.

Example: Scheduling Assistant Instructions

You are a professional scheduling assistant. Your task is to help users organize meetings by checking available time slots and collecting all required details.

Ask about:

• meeting purpose
• duration
• participants
• preferred days and times
• online or in-person format
  

After collecting this information, check the connected Google Calendar, suggest two or three time slots, confirm all details, and create the event.

Use a friendly and professional tone.

Step 6. Add and connect Composio

1. Click the Google Calendar block.
   
2. Paste your Composio API key.
   (this should have been set in “Getting the required API Keys” section, but you can also set it now if you haven’t done so previously)
   
3. If you see a yellow alert icon:
   1. Click it and log in using the same Composio account.
   2. Approve Google Calendar access.
   3. Once connected, the alert will disappear.
      
4. After connecting, click Actions, then select all available options.

Step 7. Test the agent

1. Click Playground at the top right corner to start the flow.
   
2. Type in the chat:
   I need to schedule a team meeting next week.
   
3. The agent should ask follow-up questions, find available times, and suggest options.
   
4. Confirm and check your Google Calendar for the new event.

Step 8. Your Scheduling Assistant is ready

Congratulations — you have completed your first agent. Your Scheduling Assistant is now fully functional and connected to an external tool (Google Calendar).

You can now:

• Test it by asking different scheduling questions.
  
• Adjust the Instructions if you want the agent to use a different tone or focus (for example, schedule internal meetings, patient visits, or research discussions).
  
• Replace Google Calendar with another tool later if you want to create a different type of agent.
  

This completes the process of connecting an agent to a tool and making it work with real data.

In the next section, you will learn how to create another type of agent. One that connects to files and uses them as its knowledge base.

6. Building Your Second Agent – Guideline Assistant

This agent helps you search and answer questions based on uploaded documents, such as clinical guidelines or research files. Later, you can replace these files with your own materials to create agents that assist in your specific area of work.

The purpose is to help you learn how to connect your agent to files and knowledge bases.

Step 1. Create a new flow

1. From the Project Page, click + New Flow.
2. Name it Guideline Assistant.
3. A new workspace will open for building the flow.
   

Step 2. Add the main blocks

1. In your new flow workspace, click + Add Node in the upper bar.
2. A list of available nodes will appear.
3. You can scroll through or use the search bar to find the ones you need.
4. Add the following blocks to your flow by clicking on each of them:
   • Chat Input
   • Agent
   • File
   • Chat Output

Each node represents one step in your agent’s workflow logic. Here is what each block does:

Node	Purpose in the workflow	Example
Chat Input	The starting point of the conversation. It receives the user’s question or message and sends it to the agent.	A genetic counselor types: “What are the criteria for BRCA testing in women under 40?”
Agent	The main reasoning unit of the flow. It understands the user’s question, decides what information to look for, and uses the connected files as its reference source.	The agent analyzes the message and checks the uploaded PDF to find the relevant section.
File	This block stores and reads uploaded files (for example, guideline PDFs). It gives the agent access to these documents so it can retrieve and summarize answers from them.	The agent reads the Breast Cancer Clinical Guidelines.pdf to locate the correct recommendations.
Chat Output	The final part of the flow that displays the agent’s response to the user.	The agent replies: “According to the uploaded guideline, BRCA testing is recommended for women diagnosed before 40.”

These four nodes together create a complete workflow where the user can ask questions, the agent processes them, consults the uploaded files, and returns an answer.

User message (Chat Input) → Agent reasoning (Agent) → Document lookup (PDF Loader) → Response shown (Chat Output)

Once the blocks are in place, you can proceed to upload your documents and configure the agent, as explained in the next steps.

Step 3. Connect the blocks

1. Connect Chat Input → Agent (Input dot).
   
2. Connect Agent → Chat Output.
   
3. Connect PDF Loader → Agent (Knowledge dot).
   This allows the agent to access and reason over your uploaded documents.
   

Step 4. Upload your document

1. Click the File block.
   
2. Upload one or more documents (for example, Breast Cancer Clinical Guidelines.pdf).
   
3. Wait for the upload to complete before continuing.
   

Step 5. Configure the Agent

1. Click the Agent block.
   
2. Under Settings, paste your OpenAI API key.
   (this should have been set in “Getting the required API Keys” section, but you can also set it now if you haven’t done so previously)
   
3. In the field named Instructions, you will define what your agent should do — its purpose, responsibilities, and behavior.
   

• Think of the instructions as a short description of your agent’s goal and rules.
• This is where you tell the AI who it is, what task it performs, and how it should communicate.
• You can copy the example below or write your own version to fit your needs.

Example: Guideline Assistant Instructions

You are a guideline assistant that helps users understand and find information in uploaded medical or research documents. Use the connected files to provide accurate answers and refer to specific sections when possible.

When answering:

• Summarize clearly and avoid unnecessary details.
• Use professional but simple language.
• If the answer is not found in the files, say so clearly.
  

Step 6. Test the agent

1. Click Playground at the top right corner to start the flow.
   
2. In the chat area, type a question related to your uploaded file, for example:
   What are the screening recommendations for patients under 40?
   
3. The agent should read the document, extract relevant information, and summarize it.
   

Step 7. Your Guideline Assistant is ready

Well done, you have finished creating your second agent. Your Guideline Assistant is now fully functional and connected to a document source.

You can now:

• Ask it questions about the uploaded files and see how it summarizes or retrieves information.
  
• Update the Instructions to adjust the tone, detail level, or focus (for example, clinical, academic, or educational use).
  
• Replace or add new files to expand the knowledge base the agent uses.

This completes the process of connecting an agent to a knowledge base and teaching it to use information from files.

7. After Building

Now that you have completed both examples, you have learned how to build two main types of agents:

1. Agents that connect to tools
2. Agents that connect to files

These examples demonstrate the two key capabilities of the Arkus AI Agent Builder: integrating external systems and reasoning over internal knowledge. From here, you can continue experimenting and even combine both approaches by creating an agent that reads clinical guidelines and automatically schedules follow-up consultations based on recommendations, for example.

To make the most of your next builds:

• Duplicate existing flows when creating new agents; it saves time and keeps a working structure.
  
• Check API keys and logins whenever a connection doesn’t work.
  
• Expand step by step — start with one file or one tool before adding more integrations.
  
• Explore new connections such as databases, hospital systems, or internal research repositories once you’re comfortable with the basics.