Prompt Library

The Prompt Library provides reusable prompts for an AI Agent. It helps users begin common tasks without composing every request from scratch.

Prompt Library entries are organized by category and can be displayed through the AI Agent conversation interface. They guide what the user sends to the Agent; they do not change the Agent's System Instructions, Knowledge Sources, or permissions.

When to Use the Prompt Library

Use Prompt Library entries for recurring, well-understood requests such as:

• Summarizing a document or permitted set of records.
• Explaining a supported process.
• Starting a guided intake or planning conversation.
• Producing a response in a useful format.
• Invoking an established Agent workflow with the required details clearly requested.

Do not use a saved prompt to conceal important prerequisites or authorization rules. If a workflow requires particular inputs, the Agent instructions and Function implementation must still validate them.

Add a Prompt

To add an entry:

1. Open AI Agents in the Project Editor.
2. Open the Agent's actions and select Settings.
3. Expand Prompt Library.
4. Select the add control.
5. Enter a Category.
6. Enter the Prompt text.
7. Save the entry.
8. Test it through the intended AI Agent Widget.

Both Category and Prompt are required.

The Prompt Library supports searching, refreshing, and moving between pages of entries. Use categories consistently so users can scan related prompts together.

Write Effective Prompts

A library prompt should be understandable without relying on hidden Project Editor context.

Good prompts:

• State a clear task.
• Ask the user for information that must be supplied.
• Use terminology familiar to the application's users.
• Match the selected Agent's actual Knowledge Sources and functions.
• Avoid promising an action the Agent cannot perform.
• Make important output requirements explicit.

For example:

Summarize the open support requests I am permitted to view. Group them by priority and identify any request that has not been updated in the last seven days.

This is more useful than a vague prompt such as Check my requests.

For a consequential action, the prompt can begin the workflow but should not bypass confirmation or authorization:

Help me prepare a cancellation request. First identify the order and explain whether it is eligible. Do not cancel it until I confirm.

Categories

Categories help organize the library for users. Choose short, stable labels such as:

• Getting Started
• Orders
• Support
• Reports
• Content

Avoid creating several categories that mean the same thing. Renaming a category on one entry does not automatically rename it on other entries, so update related entries consistently.

Prompt Library and the Widget

The AI Agent Widget can expose the Prompt Library as a conversation control. The Widget presents the configured entries; it does not define them.

Test each prompt in the Page where users will see it. Page context, the Agent's client Access (Security), user identity, available Knowledge Sources, and existing chat messages can all influence the result.

The Prompt Library may not appear when no entries are configured or when the current interface does not expose that control. See AI Agent Widget.

Editing and Deleting Prompts

Use the Prompt Library actions to edit an entry's Category or Prompt, or to delete the entry.

Before changing a prompt:

• Check whether its wording still matches the Agent's instructions and tools.
• Verify any names, fields, time periods, or output requirements it mentions.
• Retest it with representative users.
• Consider whether existing documentation or training refers to the old wording.

Deleting a Prompt Library entry removes it from future listings. It does not delete prior conversations that began with that prompt.

Prompt Library vs. Agent Instructions

Capability	Prompt Library	System Instructions
Visible starting point for users	Yes	No
Defines the Agent's enduring role	No	Yes
Sent as a user request	Yes	No
Enforces permissions	No	No—permissions must be enforced by trusted controls
Appropriate for optional task templates	Yes	No
Appropriate for global Agent boundaries	No	Yes

If every conversation must follow a rule, place it in System Instructions or enforce it in application logic as appropriate. Do not depend on users selecting a particular library prompt.

Testing

For each Prompt Library entry, verify:

• The correct Agent is selected in the Widget.
• The prompt is understandable to its intended users.
• Required information is requested or already available.
• The Agent selects the expected Knowledge Sources.
• Unauthorized users cannot obtain protected information or perform restricted actions.
• Tool failures and unavailable data are reported clearly.
• The result remains useful in a new chat without prior context.
• Long output remains usable in the Widget.

When Agent configuration changes, retest the affected Prompt Library entries. A prompt that worked with one source or Function schema may become misleading after that dependency changes.

Good Practices

• Keep one clear task per prompt.
• Use consistent categories.
• Match prompts to capabilities the Agent actually has.
• Ask for clarification rather than encouraging guesses.
• Keep authorization and validation in trusted Project controls.
• Test prompts in a new conversation and with representative existing history.
• Remove obsolete prompts rather than leaving users with broken workflows.

For Agent behavior settings, see Creating and Configuring AI Agents. For Widget presentation and conversation controls, see AI Agent Widget.