Setup, access and model configuration

Configure the CRM AI system prompt

Pixie CRM guidance. Follow the route for your workspace, check your role and enabled modules, and verify the result before moving on.

Follow this guide to set the parent CRM AI system instruction that OpenAI uses as the developer message and fine-tuning system message.

Audience: CRM administrators and authorised staffPermission: Settings: EditModule v1.0.0

Jump to steps

Where to goAdmin Area → Setup → Settings → AI

Before you start

Use an account with Settings: Edit and confirm the intended record or setting before making a change.
Follow the exact route above. If the screen or action is absent, check module activation, ownership and permissions rather than using another person’s account.
Use controlled test data for configuration, integration, email, AI, payment, portal or automation changes before production-wide use.

What this guide helps you do

Set the parent CRM AI system instruction that OpenAI uses as the developer message and fine-tuning system message. The instructions reflect the supplied module’s registered menus, controller actions, views, settings and code-backed validation flow.

Follow these steps

Go to Admin Area → Setup → Settings → AI.
Go to the exact setting, option or record named in this guide.
Enter or select the required value; leave unrelated options unchanged.
Save the form and wait for the success response.
Reload the page or open a controlled record to confirm the stored value is being used.

Fields and options

AI system promptStored in ai_system_prompt and read by OpenAiProvider.

Runtime roleSent as a developer-role message for normal chat.

Training roleWritten as a system-role message in every fine-tuning example.

Rules and limits

Changing the prompt changes future normal requests immediately.
Existing fine-tuned model behaviour is not retrained automatically after a prompt change.

Check it worked

The supported configure the crm ai system prompt flow completes without bypassing permission or validation checks.
The resulting record, setting, status, file, delivery event or external response is visible from the relevant workspace.
Unexpected validation, provider or linked-record errors are investigated before retrying.

Security, privacy and operational checks

Restrict the API key and AI settings to authorised administrators.
Do not submit confidential, special-category or unnecessary personal data to the external AI provider.
Check generated text before using it in customer, staff, legal, financial or compliance communication.
Monitor provider billing and retention independently because this module has no built-in cost dashboard.

Was this guide useful?Your response is stored only in this browser.