> ## Documentation Index
> Fetch the complete documentation index at: https://docs.firstquadrant.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Create fine-tuning

> Fine-tuning rules in FirstQuadrant allow you to customize the AI's behavior by defining how it should respond in specific scenarios. Rather than hard-coded if-statements, these rules act as flexible, natural-language instructions—like training an assistant. You can create fine-tuning rules for different input types, limit their scope to specific segments, and optionally enable Autopilot to automate actions without human approval.

## Types of fine-tuning rules

There are four types of fine-tuning rules:

### 1. Email rules

Used when the AI receives an email and needs to respond accordingly.

* Requires a **label** (e.g. "Time zone conflict") and a **color tag**.
* A **description** is automatically generated by the AI once you type the label. You can freely edit this.
  * This description helps the AI recognize when the rule applies. For example:
    * Label: "Postponed"
    * Description: "Any request to continue conversation at a later time. Includes both specific dates and indefinite delays."
* You must then provide **instructions** telling the AI what actions to take. These can include:
  * Reply content (e.g. "Send confirmation of postponement")
  * Workflow instructions (e.g. "Set reminder for follow-up at specified time")
  * Contact management steps (e.g. “mark as lost")

### 2. Note rules

Triggered when a note is added—either manually by a user or automatically via API.

* You only write **instructions**—no label or description needed.
* Example:
  * Instruction: "When meeting notes are added, always send an email summarizing key points."
  * API use-case: A signup note is added, and the AI sends a thank-you message with demo link.

### 3. General rules

Apply globally across FirstQuadrant.

* Only **instructions** are needed.
* Example use-cases:
  * "Always reply in Spanish when the recipient writes in Spanish."
  * "Use friendly tone-of-voice and concise language."
  * "Send confirmation email with Zoom link when a meeting is scheduled."

### 4. Knowledge rules

Enrich the AI’s contextual understanding of your sales environment.

* Define a **topic** (e.g. "Pricing")
* Add **knowledge** entries (e.g. "Starts at \$1,000/month; 50% off for YC startups for 3 months")
* These are referenced whenever the AI crafts replies to related inquiries.

## Defining scope

Each rule can be applied globally or scoped to a specific segment:

* **[Views](/product-manual/records/views)** (e.g. "All contacts", "Active deals")
* **[Campaigns](/product-manual/campaigns/campaign-overview)** (e.g. "CH Insurance CTOs - Outbound")
* **[Imports](/product-manual/imports/imports-overview)** (e.g. "All imports")

Use this setting to ensure that a rule only applies to a relevant subset of contacts.

## Creating a rule

To create a new rule:

1. Navigate to **Fine-tuning** in the sidebar.
2. Click **New rule** (top-right).
3. Select the rule type: Email, Note, General rule, or Knowledge.
4. Define the scope via the **Applies to** dropdown.
5. For Email rules:
   * Add a label and select a color.
   * Review and adjust the auto-generated description.
   * Write clear instructions (what should happen when the email is received).
6. For Note/General/Knowledge rules:
   * Skip the label and description.
   * Provide specific instructions (for Note and General) or contextual entries (for Knowledge).
7. Enable **Autopilot** (optional).
8. Click **Create fine-tuning rule**.

## Writing effective instructions

* Always write in clear, plain English.
* Instructions can define:
  * Email reply behavior (e.g. "Reply with apology and offer to reschedule")
  * CRM or deal changes (e.g. "Move deal to lost")
  * Actions (e.g. "Create task to follow up in 2 weeks")
  * Templates (e.g. full draft email with follow-ups)

## Autopilot mode

Email and Note rules support **[Autopilot](/product-manual/autopilot)**, which skips the Action List and is immediately executed:

* If only Autopilot-enabled rules apply to a given situation, the AI will execute them immediately.
* If multiple rules apply but one or more do **not** have Autopilot enabled, the action will appear in your list for review.

Use Autopilot carefully to automate trusted scenarios (e.g. out-of-office replies, time zone conflicts).