Use the Copilot Processing node to send instructions and contextual data to a large language model (LLM) during a Dialogue Tree task. The node stores the generated response in an output variable and can optionally publish the response by using a supported card template.

You can configure the prompts, data protection controls, guardrails, LLM provider, model, reasoning effort, token limit, response format, and publishing behavior.

Add a Copilot Processing Node

Follow the steps below to add the node:

• Open the required Dialogue Tree Task.
• Click the + Add icon on the canvas and select Copilot Processing node from the list.
• Configure the required prompts and output variable.
• Select Save.
• Connect the Copilot Processing node to the next node in the Dialogue Tree.
• Save and Publish the Dialogue Tree.

​

​

​

Configure the Copilot Processing Node

Configure the following fields.

Basic Configuration

​

​

​

Write the System Prompt

Use the System Prompt to define how the LLM must perform the task. Include the expected behavior, constraints, and output structure.

To insert a supported tool, enter @ in the prompt editor. To insert a variable, use the variable control in the prompt editor.

Tip: Define the expected response structure clearly. If you select the JSON response format, specify the required JSON structure and field names in the prompt.

Write the User Prompt

Use the User Prompt to provide the task input. The prompt can include system variables, user variables, and static instructions. The Copilot Processing node resolves supported variables at runtime and includes their values in the request sent to the selected LLM.

Important: Verify that every variable used in the prompt is available before the Dialogue Tree reaches the Copilot Processing node. Missing or empty values can affect the generated response.

Configure the Output Variable

In Output Variable, enter the variable that will store the generated response.

Use the output variable in subsequent nodes to:

• Evaluate the generated response.
• Apply additional processing.
• Map values to another variable.
• Publish the response.
• Include the response in a supported card template.

Choose a clear variable name that describes the stored content. For example: copilotSummary

Configure Additional Settings

Use Additional Settings to configure data protection, guardrails, and LLM settings.

To return the configurable values to their default settings, select Restore to Default.

​

​

​

Select a PII Masking Template

Use PII Masking Template to protect personally identifiable information before data is processed.

1. Open PII Masking Template.
2. Select an available template, such as Default PII Masking Template.
3. Continue configuring the node.

The selected template hides sensitive personal information covered by the template configuration.

Note: The information that gets masked depends on the rules defined in the selected PII masking template.

​

Select Guardrails

Use Select Guardrail to apply configured guardrails to user input and the AI agent's response.

1. Open Select Guardrail.
2. Select one or more available guardrails.
3. Remove an applied guardrail by selecting the remove icon next to the guardrail name.

Guardrails help enforce the input and output controls configured for your AI workflow. The exact validation behavior depends on the selected guardrail configuration.

Configure the LLM Provider and Model

In Configure LLM Provider & Model, select how the system determines which provider and model to use.

Select the Model Selection Type

​

​

​

The following model selection types are available:

To select a specific model:

1. Under Model Selection Type, select Classic.
2. In Provider, select the required LLM provider.
3. In Model, select an available model.

The configuration screen can display provider options such as Sprinklr Azure OpenAI. The available providers and models depend on the models enabled for your environment.

Configure Advanced Model Settings

Turn on Additional Settings under the model selection fields to configure advanced generation controls.

Reasoning Effort

Use Reasoning Effort to control the amount of reasoning that a supported model applies before generating a response.

For example, you can select medium when the selected model supports that value.

Higher reasoning effort can be useful for complex tasks. It can also affect response latency and token consumption. Available values depend on the selected provider and model.

Max Completion Tokens

Use Max Completion Tokens to set the maximum number of tokens that the model can generate in a response.

Set the value high enough for the expected response, but avoid using an unnecessarily large limit. The maximum supported value depends on the selected model.

If the generated response reaches the configured limit, the response might end before the model completes the requested content.

Stop Sequences

Use Stop Sequences to specify text sequences that cause the model to stop generating content.

Stop sequences are useful when:

• The response must end at a known delimiter.
• The prompt includes multiple sections.
• The output must exclude content that follows a specific marker.

Configure stop sequences only when the expected output has a consistent structure.

​

Configure Response Publishing

Use Publish Response to determine whether the Dialogue Tree publishes the generated Copilot response.

Publish the Response

Turn on Publish Response when the generated output must be sent as the Dialogue Tree response. Depending on the workflow configuration, you can publish the response by using supported templates such as:

• Proactive Summary
• Case Summary on Demand
• Copilot Renderable Card

Make sure that the generated output matches the structure required by the selected card template.

​

Store the Response Without Publishing It

Turn off Publish Response when the generated content must only be stored in the configured output variable.

You can then use another Dialogue Tree node to:

• Inspect or transform the output.
• Apply conditional logic.
• Add the output to a different response.
• Publish the response later in the same Dialogue Tree flow.

The output variable remains available for subsequent processing whether the generated response uses JSON or non-JSON format, provided that the prompt defines the expected format correctly.

Select the Response Format

Use Response Format to choose the format of the generated response.

​

​

JSON Response

When you select json, instruct the model to return valid JSON. Define the required properties and prohibit additional explanatory text.

For example:

Return valid JSON in the following structure:

{
"summary": "A concise case summary",
"nextAction": "The recommended next action"
}

Do not include Markdown or text outside the JSON object.

Use the exact JSON structure expected by subsequent nodes or card templates.

Non-JSON Response

When you select a non-JSON response format, tell the model what type of text to produce.

For example: Return a concise case summary in plain text. Limit the response to three sentences.

Important: Align the prompt with the selected response format. Selecting JSON does not replace the need to define the expected JSON structure in the prompt.

Save the Configuration

After configuring the node:

• Review the system prompt, user prompt, and variable references.
• Verify the output variable.
• Verify the PII masking template and guardrails.
• Confirm the provider, model, and advanced model settings.
• Configure Publish Response or Select the required Response Format.
• Select Save.

Select Cancel to close the configuration without saving the changes.

​

Proactive Summary Behavior

For a Proactive Summary card, the system considers the last applicable message as the response.

If the Dialogue Tree contains a bot reply, the system uses that bot reply as the proactive response. Review the final nodes and message order in the Dialogue Tree to make sure the intended content is selected.

Troubleshooting

The Copilot Response Is Empty

• Verify that the user prompt provides enough input.
• Confirm that all referenced variables contain values.
• Check whether a guardrail blocked the input or response.
• Verify that the selected provider and model are available.
• Confirm that the Dialogue Tree reaches the Copilot Processing node.

A Variable Value Does Not Appear in the Response

• Confirm that the variable syntax is correct.
• Verify that the variable is populated before the Copilot Processing node runs.
• Check whether the variable is available in the node's execution scope.
• Include an explicit instruction in the prompt that explains how to use the variable.
• Review PII masking behavior if the value contains sensitive information.

The JSON Response Is Invalid

• Select json as the response format.
• Define the complete JSON structure in the prompt.
• Instruct the model not to include Markdown code fences.
• Instruct the model not to include explanatory text outside the JSON object.
• Verify that the output does not exceed the maximum completion token limit.
• Check whether a stop sequence ends the response before the JSON object is complete.

The Response is Truncated

• Increase Max Completion Tokens, if the selected model permits a higher value.
• Shorten the system and user prompts.
• Request a more concise response.
• Check whether a configured stop sequence appears unexpectedly in the output.

The Response is Not Published

• Turn on Publish Response.
• Verify that the response matches the selected card template.
• Confirm that the output format is appropriate for the publishing configuration.
• Review the subsequent Dialogue Tree nodes for conditions that prevent publishing.
• For a proactive summary, verify which message appears last in the relevant Dialogue Tree path.

The Wrong Message Appears in a Proactive Summary

• Review bot reply nodes in the Dialogue Tree.
• Check the order in which the messages are generated.
• Confirm which message is the last applicable response.
• Remove or reorder unintended bot replies before publishing the Dialogue Tree.

Configuration Recommendations

• Give each Copilot Processing node a descriptive and unique name.
• Keep system instructions focused and unambiguous.
• Use separate system and user prompts for behavior and runtime input.
• Specify an exact schema for JSON responses.
• Use PII masking when prompts can contain sensitive personal information.
• Apply guardrails that match the workflow's input and output requirements.
• Set a completion token limit appropriate for the expected response length.
• Test each Dialogue Tree path with Publish Response both on and off.
• Test variable substitution with empty, short, and large values.
• Verify the generated response against every configured card template.
• Review proactive flows that contain bot replies.