AI Agent Node Setup Guide in Agent Studio

The AI Agent Node is the core component in Agent Studio that defines how an agent thinks, responds, and acts. It processes input from a Start Trigger or another node, understands intent, and picks the next step — generate a reply, call a tool, or capture data.

The five elements of the node

Every AI Agent Node is shaped by five settings. Get these right and the agent behaves predictably.

Prompt

The prompt sets the agent's role, tone, tasks, and tool-use rules. The agent uses it to interpret input and decide what to do. A support agent prompt might say "answer pricing queries and use the knowledge base when needed," so a question like "what does your premium plan include?" routes through the knowledge base.

Model

The model controls how well the agent understands input and generates responses. Higher-capability models reason better. Mini and Nano models are faster and lighter. Pick the model based on the use case — a stronger model spots intent in a vague message like "I'm exploring options for my business"; a lighter model is fine for direct questions.

Mode

Mode controls whether the agent talks to the user or works silently. Conversational Mode generates user-facing replies. Task Based Mode performs actions without responding — for example, extracting an email and updating the CRM behind the scenes.

Tools

Tools extend the agent beyond text generation. The agent picks the right tool based on the prompt and context. Available tools include:

• Router
• Search Knowledge Base
• Search the Web
• MCP Server
• Actions
• API Call
• Text, Image, Video, and Audio Generation

For "do you support international payments?" the agent can hit the knowledge base before answering. For ambiguous intent it can call the Router to send the flow elsewhere.

Variables

Variables let the agent use existing data and capture new data. Input variables pass context in. Runtime variables capture structured data during the conversation — for example, pulling "Name = Rahul" and "Requirement = Demo" from "Hi, I'm Rahul. I need a demo."

Node connections

The node executes when input arrives from a connected trigger or upstream node. It evaluates the input with the configured prompt, model, tools, and variables, then passes the result downstream — for example, sending captured lead details to a CRM node.

How to configure the node

1. Write a clear prompt. State the role, tasks, when to use tools, and any guardrails.
2. Pick a model that matches the complexity of inputs you expect.
3. Choose a mode — Conversational for user-facing replies, Task Based for silent actions.
4. Add only the tools you need. Fewer tools means faster and more predictable decisions.
5. Define variables. List the input variables you need and the runtime variables you want to capture.
6. Connect it up. Wire input from a trigger or upstream node and outputs to the right downstream nodes.
7. Test with real inputs before publishing the agent.

Tips for reliable behavior

• Keep prompts specific. Vague instructions produce vague agents.
• Avoid stacking tools the agent never uses — each one adds decision overhead.
• Use Task Based Mode for backstage work like CRM updates and tagging.
• Use Router when intent is unclear and you need branching logic.
• Capture only the variables that downstream nodes actually consume.