Configuring Custom APIs
Custom Actions connect the AI Agent to external APIs, enabling it to retrieve data or trigger processes from third-party systems during a conversation. When a customer query matches a configured action, the AI Agent calls the associated API endpoint and uses the response to assist the customer.
Built-In Knowledge Search Action
Every AI Agent includes a built-in Knowledge Search action, which appears as a card under Enabled Custom APIs on the Custom APIs tab. This action searches the agent's configured knowledge sources for documents relevant to the customer's query and is enabled by default.

-
To edit the action's description, click the Edit
icon on the card. This opens the Edit Custom API panel, where you can update the Description fields and click Save. -
To turn Knowledge Search off, use the toggle directly on the card. Unlike other Custom APIs, the Knowledge Search action cannot be deleted — it can only be enabled or disabled.
Disabling Knowledge Search turns the AI Agent into an actions-only agent that responds solely through its configured actions (Custom APIs or Guided Help) and the conversation itself, without searching the knowledge base. Use this when the agent's job is to perform tasks — such as checking order status, resetting a password, or booking an appointment — where knowledge-base answers aren't needed, or when no knowledge base is connected to the agent at all.
When Knowledge Search is disabled:
-
The AI Agent no longer searches the knowledge base, and knowledge-based answers, certified answers, and source citations no longer appear in responses.
-
The agent still greets customers, asks follow-up questions, and runs its configured actions normally, grounding its responses only in action results and the conversation history — never in general knowledge.
-
If a customer asks a content question the agent cannot address through its configured actions, the agent routes to the existing off-topic path and returns the topic-identification-failure message already configured for the agent — the same message shown today when a question falls outside the agent's topics.
-
A knowledge base is no longer required to configure or run the agent — this is the supported way to build a pure-action agent.
Knowledge Search is enabled by default for every agent, so existing agents keep their current behavior unless you deliberately turn it off.
Prerequisites
-
The AI Agent must be saved before enabling actions.
-
The Enable Actions toggle must be turned on.
-
API credentials must be prepared before configuring authentication.
Creating a Custom Action
To create a custom action:
- From the My AI Agents page, select an AI Agent you want to edit and navigate to the Enable Actions section.
- Turn on the Enable Actions toggle if it is not already on.
- Select the Custom Actions node.
- Click + Add Custom Actions. The Create Guided Help panel opens.
- Complete the fields as described below:
- Action Name: A unique name for this action as it appears in the actions list. Use only letters, numbers, spaces, periods (.), dashes (-), colons (:), and underscores (_).
- Description: A short description of this action as it appears in the actions list. Helps administrators identify the action's purpose.
- Sample Questions: Example customer queries that should trigger this action. The AI Agent uses these to determine when to invoke the action.
- Method: The HTTP method used to call the API. Select GET or POST from the dropdown.
- GET: Retrieves data from the API endpoint. No request body is sent.
- POST: Sends data to create a resource. A Body section is available to define the JSON payload and its variables.
- PUT: Sends data to update a resource. A Body section is available, identical in behavior to POST.
.
- URL Endpoint: The full URL of the external API endpoint (for example, https://api.example.com/action).
- Accept Header: The media type the API is expected to return. Defaults to application/json.
- Content Type: The media type of the request body sent to the API. Defaults to application/json.
-
Authentication: The credential used to authenticate API calls. Credentials are sourced from External Credentials, which are available across all workflows and actions. If no credential is selected, authentication defaults to none (unauthenticated).
-
To use an existing credential, select it from the dropdown. For more information, see Configuring External Credentials.
-
To create a new credential, select Create New Credential. A panel opens on the right with the following fields:
-
Credential Name: A descriptive name to identify the credential.
-
Authentication Type: The authentication method for the credential. Select from the following options:
-
API Key: Provide the Header Name and API Key Value.
-
Bearer Token: Provide the Token.
-
Basic Auth: Provide the Username and Password.
-
OAuth 2.0 (Client Credentials): Provide the Token URL, Client ID, Client Secret, and Scopes.
-
OAuth 2.0 (Authorization Code PKCE): Provide the Authorization URL, Token URL, Client ID, Redirect URI, and Scopes.
-
-
Base Endpoint URL (optional): The base URL of the API the credential applies to.
-
Click Test Connection to verify the credential, then click Save Credential. Saved credentials automatically appear in the credential dropdown wherever credentials are supported.
- Headers: Custom HTTP headers to include in the API request. Click + Add to add a header row. Each row contains the following fields:
- Header Key: The name of the HTTP header.
- Header Value: The value for the header.
- Query Params: Query parameters to append to the API request URL. Click + Add to add a parameter row. Each row contains the following fields:
- Query Param Key: The name of the query parameter.
- Query Param Description: A description of what the parameter represents.

- Click Save. The action appears under Active Guided Help once saved.
Managing Custom Actions
Active Actions
Lists all custom actions currently enabled for the AI Agent. Each action card shows the action name, its source (eGain for built-in actions), and its enabled status. You can Edit, Delete, or Disable the actions.
Available Actions
Lists actions that have been created but are not yet enabled. You can Edit or Delete the actions. Click the Enable toggle to make them active in the chat widget.