Rules

What are Rules?

Rules are behavioral instructions for your AI agent. They tell the agent how to handle specific situations — when a customer asks about returns, the agent should explain your return policy. When someone mentions a competitor, it should redirect to your strengths. When a question is outside scope, it should respond accordingly.

Rules do not change what the agent can do — they shape how it responds. Think of them as a briefing document the agent reads before every conversation, or on-demand when a relevant feature is triggered.

Create a Rule

Navigate to Rules in the left sidebar. Click New Rule in the top-right corner.

You will be prompted to enter a name. After confirming, the rule is created and you are taken to its detail page where you can configure it fully.

Set a Name

The name is internal and helps you identify the rule in the list. Use something descriptive that reflects the rule's purpose, for example Handle Return Requests or Competitor Mentions – Redirect.

When to Use

The When to use field tells the agent in plain language when this rule applies. This is the most important field — it defines the trigger condition for the agent.

Write it as a natural instruction:

  • When a customer asks about shipping costs or delivery times
  • When the conversation topic is product compatibility
  • When a user expresses frustration or dissatisfaction

The agent reads this during a conversation and decides whether the rule is relevant to what is happening.

Rule Content

Below the trigger, you define the actual instructions. There are two ways to do this:

Conditions define structured if-then logic. Each condition has a trigger and a response instruction. Use conditions when you have multiple distinct cases that each need a different response.

Simple Rules are free-form text instructions — a list of guidelines the agent should follow when this rule is active. Use simple rules when the behavior is nuanced or conversational, rather than a hard if-then case.

You can combine both within the same rule.

Loading Types

Every rule has a loading type that controls when the agent loads it into its context.

Always

The rule is loaded at the start of every conversation. Use this for global instructions that should always apply — your tone guidelines, prohibited topics, escalation procedures, or general behavior expectations.

Keep the number of always-loaded rules small. Loading too many rules for every conversation can affect response quality.

On Demand

The rule is only loaded when a linked tool or plugin is used during the conversation. Use this for instructions that are only relevant in specific situations.

For example: a rule about how to present product recommendations only needs to be active when the product search tool runs. Linking this rule to that tool and setting it to On Demand keeps it out of unrelated conversations.

Rules can be connected to other features in your account. These links tell the agent that a rule is relevant in the context of a specific tool, plugin, or product.

  • Tool links — the rule is loaded when a particular tool is invoked
  • Plugin links — the rule is loaded when a plugin is active or triggered
  • Product links — the rule is associated with specific products, guiding how the agent discusses them

Links are especially useful for On Demand rules. An always-loaded rule does not need links to fire — it is always active. But a rule set to On Demand only activates when something it is linked to is used.

To link a rule, open the rule's detail page and use the link management section to add connections to tools, plugins, or products.

Settings

| Setting | Description | |---|---| | Name | Internal label for the rule | | When to use | Plain-language trigger: when should the agent apply this rule? | | Loading type | Always (every conversation) or On Demand (linked trigger only) | | Conditions | Structured if-then behavior definitions | | Simple Rules | Free-form text instructions for nuanced behavior | | Tool links | Tools that trigger this rule when On Demand | | Plugin links | Plugins that trigger this rule when On Demand | | Product links | Products this rule is associated with |

Manage Rules

The Rules list shows all rules in your account. Use the search field at the top to filter by name. This is useful when you have many rules covering different topics or use cases.

Edit a Rule

Click on any rule in the list to open its detail page. From there you can update the name, the when-to-use instruction, the loading type, conditions, and simple rules. You can also manage the links to tools, plugins, and products.

Save your changes with the Save button. Changes apply immediately to the next conversation the agent starts.

Delete a Rule

Open the rule's detail page and use the delete option. Deletion is permanent. Any links this rule had to tools, plugins, or products are also removed.

Before deleting a rule that handles important behavior, consider editing it instead — you can clear the conditions and simple rules and repurpose it for a different use case.

FAQ

Can I have multiple rules active at the same time? Yes. The agent can load and apply multiple rules in a single conversation. Always-loaded rules are all active simultaneously. On Demand rules stack with any others that get triggered.

What is the difference between conditions and simple rules? Conditions are structured: a trigger and a specific response. Simple rules are free-form instructions. Use conditions for clear if-then logic, simple rules for tone, style, or nuanced guidance that does not fit a binary trigger.

How many rules should be set to Always? As few as possible. A handful of always-loaded rules for global behavior (tone, escalation, prohibited topics) is a good baseline. Move everything topic-specific to On Demand to keep the agent's context lean.

Do rules affect all widgets? Rules are configured at the account level and apply to the AI agent across your widgets. If you need different behavior for different widgets, create rules that are specific enough in their when-to-use instructions to only apply in the right context.