Working with Rulesets in Gen AI Builder

What are Rulesets

Rulesets in Gen AI Builder are collections of natural language instructions that guide the behavior of your Assistants and Structures. They help ensure responses are aligned with your specific instructions and organizational guidelines.

Rulesets serve as a layer of explicit control over how your AI behaves — complementing the knowledge it retrieves from Knowledge Bases.

In short: Rulesets tell your Assistants how to behave, not just what to retrieve.

For a deep dive, see: Rulesets explained.

Why use Rulesets

• Guide behavior: Ensure Assistants respond in ways that match your tone, style, and policies.
• Enforce constraints: Prohibit certain types of content or behaviors (e.g., avoid legal advice, don’t mention competitors).
• Support multiple personas: Define distinct behavioral patterns for different Assistants.
• Complement RAG pipelines: Combine retrieved knowledge with consistent behavior.

Rulesets are particularly powerful when combined with:

• Knowledge Bases (what content to retrieve)
• Structures (how to process tasks)
• Tools (how to interact with external systems)

When to use Rulesets

• When building Assistants that interact with end users.
• When defining behavior for internal Agents and Pipelines.
• Whenever you want to ensure consistent tone, style, or compliance across AI applications.

Examples:

• Customer support Assistants using polite and professional tone.
• Internal policy chatbots with strict legal disclaimers.
• Marketing copy Assistants that reflect brand guidelines.

How Rulesets fit into Gen AI Builder

The typical workflow is:

Create Ruleset → Add Rules → Create Assistant → Assign Ruleset to Assistant

At runtime:

• The assigned Ruleset is applied to every message processed by the Assistant.
• The Assistant uses these Rules in combination with retrieved content and its base LLM.

Components of a Ruleset

• Name — Required, unique name.
• Description — Optional description.
• Rules — One or more Rules, each written in natural language.
• Alias — Optional unique alias (for versioning or API access).
• Metadata — Optional JSON object for advanced configuration.

Rules themselves consist of:

• Name (required)
• Rule text (required) — written as an instruction in natural language.

Getting started

See Create a Ruleset for a full step-by-step guide.

Example flow:

1. Create a Ruleset → "Polite Support Tone"
2. Create Rules:

• "Use Formal Salutations"
• "Offer Assistance Clearly"

3. Create an Assistant and assign the Ruleset.
4. The Assistant now responds with this defined tone and style.

Related topics

• Rulesets explained
• Create a Ruleset
• Structures explained
• AI Factory Concepts
• Hybrid Manager: Using Gen AI Builder