Configuring Prompts & LLM Settings

Last updated: August 20, 2025

This tutorial explains how to configure Prompting and LLM Settings in Zams. These settings let you control how your agent behaves, how it responds to users, and how it interacts with integrations.

Prerequisites

There are no prerequisites for this tutorial.

Steps

Step 1: Agent Config Tab

  1. Sign in to Zams.

  2. From the Dashboard, select your agent.

  3. Go to the Config tab.

  4. Locate the AI Settings section. Here you can set:

    • Large Language Model (LLM)

    • Background Info

    • Escalation Protocol

    • Out-of-Context Handling

    • Security Notes

Step 2: Select the LLM

1. In the Large Language Model (LLM) dropdown, select the model you want the agent to use.

  • Example: Claude Sonnet 4.

2. The chosen model will define how the agent generates responses.

Tip: Different models may have different strengths. Experiment to see which best fits your use case.

Step 3: Configure Background Info

  1. In the Background Info tab, describe the agent's role.

  • Example: This agent will be used to summarize text and retrieve specific information from given text.

  1. Think of Background Info as the system prompt—the rules the model always follows.

  2. This can be updated anytime if you want to refine the agent's behavior.

Step 4: Configure Escalation Protocol

  1. Open the Escalation Protocol tab.

  2. Define how the agent should handle sensitive information.

    • Example: Escalate customer payment issues to john@acme.com.

  3. This ensures critical cases are routed correctly.

Step 5: Configure Out-of-Context Handling

  1. Open the Out-of-Context Handling tab.

  2. Define how the agent should respond if a user asks something outside its intended purpose.

    • Example: If asked unrelated questions, politely respond: "I can only assist with email summarization".

Step 6: Add Security Notes

  1. Open the Security Notes tab.

  2. Add specific guardrails to enforce security rules.

    • Example: Never share customer passwords.

  3. The agent will follow these guidelines within the scope of its tasks.

Note: If you want to learn more about Shortcuts, follow our guide on Creating a Shortcut.