LLM Prompts
Form ID: (ML202000)
You use this form to create a prompt for the LLM. After you create the prompt and test it, you select the Active check box and save your changes. This causes the model to generate an action for the prompt and add the corresponding command to the More menu of the source form. When a user clicks this command, the model executes a data generation task associated with the prompt.
This form is available only if the AI Automation feature is enabled on the Enable/Disable Features (CS100000) form.
Form Toolbar
The form toolbar has only standard buttons.
Summary Area
In this area, you can specify or modify the basic settings of the LLM prompt.
| Element | Description |
|---|---|
| Prompt ID |
Required. The internal name of the prompt. |
| Active |
A check box that you select to make the prompt active and to add the LLM command to the source form. (By default, it’s cleared for a new prompt.) If you need to remove the generated LLM command from the form, you clear this check box. |
| Prompt Name |
Required. The display name of the prompt. This name is also used for the command (and button) if the Button Name box is left blank. |
| LLM Connection |
The name of the connection to be used for the LLM prompt. An LLM connection is defined on the LLM Connections (ML201000) form. You can create the prompt without a connection; however, you need a connection to test and use the prompt. The lookup table for this box displays only connections that have been tested—that is, connections with the Active status. |
| Source Form |
Required. The form for which the command should be created. Attention: Only data entry forms are available for
selection.
|
| Button Name |
The name of the command on the source form’s More menu that the model will generate. (This name will also be used for a button on the form toolbar if a user selects the command as a favorite.) When a user clicks the command or button, the system executes the generated action. |
| Element | Description |
|---|---|
| Example Record |
The record that’s used to test the prompt. After you select a record, the system adds to the Prompt tab the full prompt that it will send to the model. In this prompt, the system replaces the input fields from the prompt with the data of the record that you have selected in this box. |
| Input Tokens |
The number of tokens—that is, small units of text—sent to the LLM. |
| Output Tokens |
The number of tokens received from the LLM. |
| Total Tokens |
The total number of tokens. |
|
The section also contains the following button. |
|
| Test Prompt |
Sends the prompt with the data of the example record to the LLM and displays the results on the Preview and Response tabs. This button is available if a connection is selected in the LLM Connection box. |
Instructions Tab (Bottom Left Pane)
On this tab, you add instructions for the LLM in the text area. By default, the tab
contains three sections: Context Instructions,
Instructions with Input Data, and Output Data
Field. You can add your own sections to refine the instructions, modify
existing sections, or
remove
them (not recommended). You can use the ## notation or
any other notation to mark the headers.
The tab contains the following elements.
| Element | Description |
|---|---|
| Context Instructions |
General instructions for the model. This could be the business context, the description of the role for which the prompt is executed, or the tone of voice and the style of the text. Attention: Context instructions provide additional
information that affects the results and facilitates the correct execution of the
prompt. Avoid removing these instructions; instead, expand them according to the
user’s task.
|
| Instructions with Input Data |
The general description of the user’s task, the description of the actions that the model should execute, and the input data that it should use to perform the task. As input data, you use placeholders—that is, internal names of the data
fields from the source form enclosed in double parentheses, such as
|
| Output Data Field |
The list of fields that the model should return, information on how it should fill in those fields, and any limitations relevant to the task the command will help the user perform. You must specify the fields of the source form for which the command is generated. You use placeholders for the output data, such as
Important: If you don’t specify any output fields, the system displays
an error when you try to generate an example or test the prompt.
|
|
The formatting toolbar has the following buttons. |
|
| Insert Instructions |
Opens the Insert Instructions dialog box, where you can select a set of system instructions to be added to the prompt. You create system instructions on the LLM System Instructions (ML203000) form. |
| Insert |
Displays commands you can use to select a data field of a particular type:
|
| Element | Description |
|---|---|
| Prompt ID |
The identifier of the set of system instructions. |
| Description |
The description of the set of system instructions. |
| Insert Above Prompt |
A check box that you select to insert the system instructions at the beginning of the prompt. |
| Insert Below Prompt |
A check box that you select to insert the system instructions at the end of the prompt. |
|
The dialog box has the following buttons. |
|
| Insert |
Adds the set of system instructions to the prompt and closes the dialog box. |
| Cancel |
Closes the dialog box without adding the set of system instructions to the prompt. |
Output Format Tab (Bottom Left Pane)
On this tab, you specify how the data returned by the model be formatted.
You can use only the output fields from the source form. To cause the system to fill in those fields correctly, you must use the special format that the system generates. We recommend that you not modify this format.
When the system generates the example, it uses real data from the database. If it detects a field that has a masked version on the Instructions tab, the system replaces this field with its masked version. You can replace the data in the example with any arbitrary values.
The tab contains the following elements.
| Element | Description |
|---|---|
| Output Format Example |
An example of the data the model should return. This example indicates to the model the structure of the returned data, which should be in JSON format with the same fields as those specified in the Output Data Field section of the Output Format tab. |
| Important Note |
Information about how the returned data should be formatted. Attention: Avoid removing this section. It
provides guidance on what data the model should return and how this data should be
formatted, which significantly improves the quality of the results.
|
| The tab also contains the following button. | |
| Generate Example |
Generates an example of the returned data. The system generates an output example for the model by using the output fields specified on the Instructions tab. The system uses arbitrary values for these fields from the database to show how the response can look like. If you want to add more details to the response, edit the generated example accordingly. |
Preview Tab (Bottom Right Pane)
This tab shows how the generated data will be displayed on the appropriate form.
Prompt Tab (Bottom Right Pane)
This tab displays the full prompt that the system will send to the model. This prompt is a combination of the data from the Instructions and Output Format tabs. The system replaces the input fields in the prompt with the actual data of the record that you’ve selected in the Example Record box.
If you’ve added a masked field to the prompt, the system displays the masked data on this tab instead of the real data. The system will send the masked data to the LLM.
Response Tab (Bottom Right Pane)
This tab displays the data that the model has returned in JSON format. The system uses this format to update the fields on the source form.
When you test a prompt that contains masked fields and the LLM sends the processed data back to MYOB Acumatica, the system replaces the masked content with the actual data.