Skip to main content
To configure the agent’s AI settings, click the Settings button in the sidebar and select AI. Agent AI settings

​
Available settings

​
AI Model

The AI model that the agent will use to think, plan, and generate answers to the user queries. Every model has different capabilities and costs. We recommend testing with different models in the Playground to see which one works best for your agent.

​
Temperature

Controls randomness/creativity in the Agent’s writing and decision-making. Lower = more deterministic; Higher = more varied.
  • 0.0-0.3 (Precise): Best for support, policy-bound replies, data extraction, or when strict adherence to facts and formats is required.
  • 0.4-0.6 (Balanced): Good general setting for helpful responses with light creativity.
  • 0.7-1.0 (Creative): Use for brainstorming, marketing copy, or when variety is desirable. Expect less consistency.
Tips
  • If your Agent must follow exact steps (e.g., collecting parameters for an HTTP Request), keep temperature low.
  • Raise temperature only where tone/creativity matters and accuracy isn’t compromised.

​
Limit answers to knowledge base

When enabled, the Agent only answers using information from its knowledge base. When disabled, it may use its general knowledge when the knowledge base does not contain a relevant answer.

​
Attachments

Allow users to attach images and otehr files for the AI Agent to analyze. You can choose the type of files that are allowed to be attached and how many attachments are allowed per message. Learn more about file attachments.

​
AI reply exclusion rules

Use exclusion rules to control how the Agent handles messages containing specific words or phrases. Rules can apply to any channel or a selected channel. When a message matches a rule, the Agent can:
  • Skip that reply only.
  • Skip the reply and pause AI for the conversation.
  • Reply, then pause AI for the conversation.
Paused conversations can be resumed manually from the conversation page or automatically after a set time. Rules are checked from top to bottom, and the first matching rule is applied.

​
Instructions

Define the Agent’s role, goals, guardrails, and style (often called the “system prompt”). Here’s some of the things you can include in the instructions:
  • Role & purpose: What the Agent is for and what success looks like.
  • Scope & boundaries: What it should/shouldn’t answer.
  • Tone & language: Brand voice, formality level, and multilingual behavior (auto-detect language; reply in user’s language).
  • Compliance & safety: Any legal disclaimers, restricted topics, PII handling, and masking sensitive values.
  • Formatting: Preferred reply structure (short summaries, bullet points, tables).