How to create an AI Agent

This guide walks through every field of the AI Agent form so you know what each does and how it changes the AI Agent's behaviour. For a faster "happy-path" version, see Getting Started.

Steps for creating an AI Agent

Step 1: Getting Started

1. Name Your AI Agent: Start by giving your AI Agent a name. This could be anything that resonates with the purpose of your AI or your business.
2. Describe Your AI Agent: Provide a brief description. This could highlight what your AI Agent does or what questions it can answer.

Step 2: Configuring Your AI Agent

1. AI Model Selection:
   • Choose from the available AI models (OpenAI, Google Gemini, Anthropic Claude, and more). Start with a cost-effective option to get things working, then test how more capable models perform. It's your choice!
2. Creativity (Temperature):
   • Adjust the creativity level. A lower setting results in more predictable, straightforward answers, while a higher setting encourages more varied and inventive responses.
3. AI Agent Instructions (System Prompt):
   • This is where you guide your AI Agent. For instance, if you want it to respond with something specific if it could not respond from the trained data, you should do it here. F.e: If you can't make a response from the context provided, you should respond exactly "I'm really sorry but I can't respond to your question 😞"
   • Or, if you want the AI Agent to run tools sequentially you can do something like that In order to write me a personalized message for a company, you have to firstly search using 'Web Browser' their CEO's linkedin profile, then you have to use 'Linkedin Profile Scraper' tool to get their profile details and then you have to write me a personalized message to sell Cubeo AI to them.
4. Visibility Options:
   • Decide if your AI Agent should be Private or Public. If you wish to embed it on your website or share via a public link, choose Public.
5. Json Structure Response:
   • Define a standard structure of your AI Agent to respond. Define each parameter you want the AI Agent to return, also if it is required or not then your AI Agent will create a JSON with all the defined parameters by respecting the structure you define.
6. Data Handling Features:
   • Do Not Create Responses Outside Training Data: Check this if you want your AI Agent strictly to stick to the facts in your documents. It's ideal for ensuring your AI doesn't invent answers. It is applicable only if you have added documents in the training data
   • Display Source Documents In Public Link: Enable this to let users see the documents your AI Agent was trained on when they access it via a public link. They can view PDFs, Word documents, videos, etc., directly on the page.
   • Display Sources In AI Agent Response: If checked, your AI Agent will append links to its responses, pointing to the documents it used. This provides transparency and allows users to explore the source material themselves.
   • Display Tool Running Progress In Chat: It allows users to see the tools the AI Agent is using in real-time. For each tool, users can view the input provided and the response received upon completion, offering a peek behind the curtain of the AI Agent's thought process. This is especially useful when using Human-in-the-Loop, where you review and approve tool actions before they execute.
   • Beautify JSON Structured Response: When you define a JSON structure to be returned by your AI Agent, maybe you want to display the formatted JSON in the chat. This option enables the JSON transforming into Markdown format

success

Don't worry! The links built by AI Agent to your documents expires after a while and they can't be accessed anymore. All for your security!

warning

If you delete your AI Agent, all its training documents will be deleted

Real-Life Example

Imagine you run a cozy café and you’ve set up an AI Agent to answer frequent customer inquiries. You’ve trained it with your menu, a document on your sourcing practices, and a list of your upcoming events.

When a customer asks, Do you have any gluten-free options? your AI Agent can pull the exact section from your menu document to provide an answer. If Display Sources In AI Agent Response is enabled, the customer also gets a link to the menu PDF for further exploration.

Tips for Success

• Experiment with Creativity: Finding the right temperature setting can greatly influence the type of responses you get. Don’t hesitate to adjust this as you see fit.
• Detailed Instructions: The more guidance you give your AI Agent in the system prompt, the better tailored the responses will be. Think about the style, tone, and scope of information you want it to use. If you don't want it to respond outside training data, you must instruct it here about how to do that, how to respond if it doesn't find relevant info in context provided. F.e: You must always respond from the given context provided delimited by <context></context>. If you can't respond to a question just say exactly "I don't know the answer. What else can I do for you?"
• Utilize Tools Wisely: Integrate tools that complement your AI Agent's capabilities, enhancing its ability to provide helpful and accurate information.

Next steps

• Train your AI Agent — upload PDFs, websites, or Q&A pairs so it can answer from your content.
• Add tools — let it call APIs, search the web, scrape pages, talk to your CRM, or use any MCP server.
• Connect integrations — set up HubSpot, Tavily, MCP servers and more, once for the whole team.
• Turn on Human-in-the-Loop so sensitive actions wait for your approval.