The AI Agent Node is the core building block of any intelligent conversation in Frontline.
With it, you can control:
The model your agent uses at a specific point in time
How it reasons and responds (via instructions & temperature)
What tools it can call
What happens when the conversation ends
And you can trace it all through the audit log
š§ Configuration Tab
The Configuration tab lets you configure how your AI Agent behaves within this step of the flow.
š§¾ Model
Choose which AI model powers this node (e.g. GPT-4.1). You can tailor the agentās capabilities and performance per step in the conversation.
āļø Instructions
Write clear, role-specific instructions that guide the Agentās behavior.
Examples:
āAct as a helpful assistant. Help the user with transaction history.ā
āRespond like a friendly sales rep. Recommend a plan based on user needs.ā
These instructions shape how the Agent interprets and responds to user inputs.
š„ Temperature
Control how creative or precise the agent should be:
0.0 = predictable, deterministic
1.0 = more creative and varied
For support and transactional use cases, keep it low (0.1ā0.3). For open-ended conversations, increase as needed.
šŖ Exit Paths
Define how the flow should continue after the Agent responds.
You can create custom exit paths (e.g. āuser asked about pricingā) or use a single default path like:
Exit node when the user has finished inquiring about transactions
š Tools Tab
Connect any custom Agent Tools to this node so your Agent can fetch data or trigger actions using APIs. By default all AI Agent nodes have access to file_search which power search across the agents' knowledge base.
To do this:
Navigate to the Tools tab inside the node.
Select one or more tools from your account (e.g. āOrder Statusā, āTransactionsā).
The AI Agent will automatically determine if and when to call them during the conversation.
For example:
If the user asks āCan I see my last 3 payments?ā, the Agent may call the Transactions tool and return a detailed breakdown.
š” Tools can also interpolate variables like {order_id} from captured user inputs or conversation context.
šŖµ Audit Log Visibility
Every time an AI Agent Node runs, youāll see a full Conversation Audit trail that includes:
āļø Agent instructions
š§ Reasoning path
š§° Tool calls (with inputs and outputs)
š§¾ The Agentās response
š” What influenced the answer (memory, tools, RAG, etc.)
This makes it easy to debug, improve instructions, and understand how your AI Agent came to a conclusion.
ā Best Practices
Keep instructions short but clear. Guide tone, role, and scope.
Start with low temperature for predictable outcomes.
Assign tools only when relevant to avoid over-triggering.
Use variables + exit paths to branch flows based on user goals.
Review the audit log often ā itās your best source of truth for tuning the experience.