Skip to content

Documentation Index

Fetch the complete documentation index at: /llms.txt

Use this file to discover all available pages before exploring further.

Define your agent's task and response style

Write clear instructions for what your agent helps with, how it responds and what it does when information is missing.

Open copy menu
View as Markdown

Instructions define your AI agent’s task and behavior. Data sources contain the information it uses to answer. Keep these separate: put your returns policy in a source and use instructions to explain that the agent helps with returns questions.

Open the correct workspace and agent in Oma Aihio. You need permission to change the agent’s settings. Keep a copy of the current instructions before replacing them.

  1. Open the Instructions field

    Open the agent’s AI settings and find Instructions. If you use an instruction template, read it before editing: choosing a different template can replace the current field content.

  2. Describe the task, boundaries and style

    Explain who the agent helps, which questions it answers and when it directs the customer elsewhere. Include what to do when information is missing. Avoid conflicting rules.

  3. Save and check

    Select Save and wait for confirmation. Then test a common question, an out-of-scope question and a question that the sources do not answer. Review the responses yourself.

The following text is an editable example, not a guarantee of agent behavior:

Help customers find information about our products and deliveries.
Keep responses brief and use clear language.
Base product and delivery answers on the data sources.
If information is missing, say so. Do not invent prices or delivery times.
Ask for one clarification at a time when needed.

Include a contact method only if customers can actually use it. An instruction to hand over to a human does not enable human support.

  • Check that saving succeeded and you are testing the correct agent.
  • Remove duplicate or conflicting instructions and try the same question in a new conversation.
  • If the facts are wrong, correct the data source and update the knowledge base.
  • If the change made responses worse, restore your saved copy of the instructions, save and retest.

Instructions do not guarantee correct answers. Keep a few important test questions and repeat them after each material change.