Skip to main content

Prompt Management

Carlos Camacho

Prompt Management lets administrators view and customize the AI prompts and policies that control how the Stix and Brix chatbot responds for your organization. You can review the base system prompt (read-only), edit tenant-specific policies, and monitor token usage — all changes take effect immediately for new conversations.

Admin Chat or Super Admin access required. You need admin chat or super admin permissions to access Prompt Management. If you don't see the Prompt Management tile on the Home screen, contact your administrator.

Accessing Prompt Management

  1. From the Home screen, click the Prompt Management tile
  2. The page loads with three collapsible sections: Prompt Statistics, System Prompt, and Tenant Policies

Prompt Statistics

Click the Prompt Statistics section header to expand it. This panel shows token and character usage at a glance:

Metric Description Color Coding
Estimated Tokens Approximate token count for the full prompt Green (<50k), Amber (<75k), Red (≥75k)
Total Characters Combined character count across system prompt and policies -
System Prompt Character count of the base system prompt -
Tenant Policies Character count of all tenant policy text -

Below the cards, a Policy Breakdown shows the character count for each of the three policies: Domain Context, Response Style, and Quality Assurance.

System Prompt (Read-Only)

The System Prompt section shows the base prompt that drives the AI. This prompt is managed by NXT and cannot be edited by tenant admins.

  • The header shows the prompt version and the number of sections
  • Each section is listed with its name and content
  • The {client_policies} placeholder in the system prompt is where your tenant policies get injected

Why is this read-only? The system prompt contains the core instructions that ensure the AI behaves correctly across all tenants. Your customizations happen in the Tenant Policies section below, which gets injected into the system prompt at the {client_policies} placeholder.

Tenant Policies (Editable)

This is where you customize how the AI responds for your organization. There are three policy sections, each containing editable text elements:

Domain Context

Controls what topics the AI is allowed and restricted from discussing. Typical elements include:

  • Allowed Topics — subjects the AI should confidently answer about
  • Restricted Topics — subjects the AI should decline or redirect

Response Style

Controls the tone and formatting of AI responses. Typical elements include guidelines for formality, length preferences, and communication style.

Quality Assurance

Controls accuracy and citation requirements. Typical elements include guidelines for sourcing, fact-checking, and when the AI should say "I don't know" rather than guess.

Editing a Policy

  1. Expand the Tenant Policies section
  2. Click on the policy element you want to edit (e.g., "Allowed Topics")
  3. Modify the text in the textarea — a character counter shows {current}/{max}
  4. Click Save Changes to apply (the button appears when changes are detected)
  5. Changes take effect immediately for new conversations

To restore the default text for any policy, click Reset to Default.

Troubleshooting

I don't see the Prompt Management tile

Solutions:

  • You need admin chat or super admin permissions. Contact your administrator.

Token count is in the red zone

Solutions:

  • Your combined policies are very large (≥75k tokens). This can slow down AI responses and increase costs.
  • Review each policy section and trim unnecessary text — focus on the most important guidelines
  • Remove duplicated instructions across policies

My policy changes don't seem to affect responses

Solutions:

  • Policy changes apply to new conversations only — they don't retroactively change existing chat threads
  • Start a new chat to test your updated policies
  • Make sure you clicked Save Changes — if the button is still visible, the change hasn't been saved yet

Need help? Contact our support team at support@nxtconstruction.ai

Comments

0 comments

Please sign in to leave a comment.