An AI Agent is an intelligent partner built on Large Language Models (LLMs).

It understands user goals, makes its own decisions, and performs real tasks.

Agents can use memory to maintain the context of conversations, and they support chat-based testing to provide human-like interactions.

Step 1: Create a Project

Click the Create Project button on the main page to start a new project.

If needed, you can pin the project to the top of the left sidebar using the pin icon for quick access.

Step 2: Select Agent

Within the project sidebar, you can choose between Ability and Agent.

For this guide, we’ll select Agent.

Step 3: Add an Agent

Inside the project, click the Create button to add a new Agent.

Step 4: Enter the Canvas

Click on the created Agent to open the Canvas Editor.

The Canvas is where you design and build your workflow.

Just like with Abilities, Agents come with one INPUT node and one OUTPUT node by default.

Step 5: Use Agent-Specific Nodes

Unlike Abilities, Agents can make use of AI Agent nodes.

These include:

• Long-term Memory

• Short-term Memory

• Agent Tools (to execute other agents)

• ReAct Agent Node

Choose the right node for your scenario.

Example: If you want to build a word-chain game agent, you’ll need the Short-term Memory node to remember the user’s last input.

Step 6: Configure and Connect Nodes

Click Add Node at the bottom of the canvas to add nodes.

Drag and drop the nodes you need, then connect them with edges to define data flow.

Double-click a node to configure input/output variables, and add code or prompts if necessary.

(Tip: See [Working with Nodes] and [Building Abilities] for more details.)

After connecting edges, make sure to declare the final output in the OUTPUT node.

Step 7: Design Conversational Logic

Unlike simple function execution, Agents maintain conversational context and interact like humans.

• Use Short-term Memory to recall recent dialogue.

• Use Long-term Memory to bring back user preferences or past records.

• Add conditional branches so the Agent can choose different responses or actions depending on the situation.

Step 8: Expand with AI Features

Add an LLM node to generate conversational responses or handle complex tasks.

If needed, you can also use Agent Tool nodes to combine external tools or exploration features.

This allows your Agent to move beyond simple conversation and take real decision-making actions.

Step 9: Chat-Based Testing

Ability tests focus on checking sequential execution, but Agent tests are done through a chat interface.

Chat with your Agent directly to see how it behaves.

If you notice awkward or incorrect responses, adjust the nodes and test again.

Step 10: Improve and Scale

Run tests repeatedly to check how well the Agent maintains context, uses memory, and makes decisions.

Through iteration, you’ll refine your Agent into a more natural and intelligent conversational partner.