Documentation sections

For users

Configure an AI agent

Give the agent a clear role, reliable knowledge, and safe boundaries for independent work.

Updated: August 7, 2026

01

Role and rules

A good setup explains whom the agent helps, which tasks it solves, and what it must never do. Write concrete rules and test them against real questions.

  • Specify the product, audience, and purpose of the conversation.
  • Set a calm tone and an acceptable answer length.
  • Describe when the agent must acknowledge that information is missing.
  • Forbid promises, discounts, and actions that require manual employee approval.
  • Define the conditions for operator handoff.
02

Project knowledge

Knowledge explains how your product works: pricing, terms, instructions, and common errors. The current status of a specific customer must come from verified context or read-only diagnostics, not from a general article.

03

Choose an automation mode

  • Use drafts during setup and after major knowledge changes.
  • Auto reply is suitable for common scenarios that have already been tested.
  • Pause the agent when the channel is connected but answers must not be created temporarily.
  • When using manual control, finish the operator’s work before returning the conversation to the agent.

Continue reading

Configure an AI agent — Kanyman Docs