You want your agent to help customers check their order status. You add your order management system as a source. The agent learns about orders in general - but when a customer asks about their specific order, it still cannot tell them. It can describe what an order status is. It cannot look one up.

That is the difference between a source and an action. A source gives your agent knowledge. An action gives your agent the ability to do something with it - look something up live, send a message, update a record. For order status, you need an action that calls your order management API, not a source that describes how orders work. Understanding which one to reach for will save you a lot of time spent wondering why something is not working.

In this article, you'll learn:

• What makes Sources and Actions different
• How to decide which one you need
• Real examples of when to use each

The Core Difference

Think of it this way:

• Knowledge Sources give your agent something to know.
• Actions give your agent something to do.

A Source is static content - articles, documents, past conversations - that your agent reads and uses to form answers. An Action is a live task your agent can execute during a conversation.

When to Use a Knowledge Source

Use a Knowledge Source when the answer already exists somewhere as written content - and your agent just needs to find it and share it.

Use a Source when:

• The answer is in your knowledge base, a document, or a webpage
• The information is relatively stable and doesn't change in real time
• You want your agent to explain, summarize, or guide based on existing content

Examples:

• "How do I reset my password?" → answered from a knowledge base article
• "What's your refund policy?" → answered from a policy document
• "How does the billing cycle work?" → answered from an FAQ snippet

When to Use an Action

Use an Action when answering the question requires doing something - fetching live data, triggering a process, or interacting with an external system.

Use an Action when:

• The answer depends on real-time data (e.g., order status, account balance, availability)
• Something needs to happen as a result of the conversation (e.g., send an email, book a meeting)
• The agent needs to interact with an external tool or API

Examples:

• "What's the status of my order?" → needs to look up live data via an API
• "Can I book a demo for Thursday?" → needs to check availability and create a calendar event
• "Send me a summary of this conversation" → needs to trigger an email

The Grey Area: When You Might Need Both

Sometimes the same question requires both a Source and an Action working together.

Example: A user asks about upgrading their plan.

1. The agent uses its Knowledge Source (your pricing documentation) to explain the available plans and what each one includes.
2. If the user decides to upgrade, the agent uses an Action (a CRM or billing API) to initiate the upgrade process.

The Source handles the information part. The Action handles the doing part. Together they create a conversation that goes from question to resolution without the user having to go anywhere else.

A Quick Decision Guide

Ask yourself: "Does my agent need to look something up or do something right now - or does it just need to know the answer?"

• Just needs to know the answer → Knowledge Source
• Needs live data or to trigger something → Action
• Needs to explain something AND do something → Both

Best Practices

• Start with Sources. Get your agent answering questions well before adding Actions - a well-sourced agent without Actions is more useful than a poorly sourced one with lots of Actions.
• Don't use an Action where a Source will do. Actions add complexity. If the answer is already in your knowledge base, you don't need an Action to retrieve it.
• Think of Actions as automation opportunities. Any time your support team manually does something after a conversation ends - sending a follow-up, logging a ticket, checking an account - that's a candidate for an Action.