Skip to main content

Suggested actions

Suggested actions are clickable buttons that help people work with your AI agent. There are two kinds:

  • Initial suggested actions — Starter prompts when someone opens a new conversation.
  • Follow-up suggested actions — Contextual buttons after a model reply with natural next steps.

Initial suggested actions​

Initial suggested actions appear when you open a new chat with an agent that has them configured. They show as buttons before the first message — no typing required.

How they work​

With initial suggested actions configured, you see buttons before any message is sent. Each button has a label, an icon, and a color.

Clicking a button sends that text as your first message. The model responds as if you had typed it. After the first message (button or typed), initial suggested actions hide and the chat continues as usual.

Example: A project-management agent might offer:

  • Create a task — "Help me create a new task for my project."
  • Check my schedule — "What's on my calendar for today?"
  • Summarize updates — "Give me a summary of recent project updates."

Configuring initial suggested actions​

Agent builders set these in the agent editor:

  1. Open the agent.
  2. Open the Character tab.
  3. Scroll to Initial Suggested Actions.
  4. Click Add Initial Suggested Action.
  5. For each action, set:
    • Label — Text on the button (for example, "Summarize this topic"). Keep it short.
    • Message — Full text sent when the button is clicked.
    • Icon — Options such as sparkle, chat, search, book, edit, tools, message, user, settings, or file.
    • Color — Button background.
  6. Save the agent.

If you skip icon or color, defaults are inferred from the label and message.

Tips for initial suggested actions​

  • Be specific — Prefer "Get a summary of last week's sales report" over "Ask me anything."
  • Highlight what the agent does best — Pick actions that show core value.
  • Keep labels short — A few clear words.
  • Write solid messages — That text is what the model receives.
  • Use two to four actions — Too many buttons feel noisy.

Limitations​

  • They only show at the start of a new conversation.
  • They are configured per agent.
  • Labels are limited to 60 characters.

Follow-up suggested actions​

Follow-up suggested actions show as buttons after a model reply so you can continue without typing.

How they work​

After the model responds, the system looks for offers, questions, or next steps. If it finds any, it shows them as colored buttons under the reply. Clicking sends that text for you.

Typical triggers:

  • Questions to you — "Would you like me to explain further?"
  • Offers — "I can also look into pricing if you'd like."
  • Option lists — "Here are a few things I can do next: 1. Summarize the report 2. Draft an email"
  • Prompts to continue — "Let me know if you'd like more detail on any of these."

When they do not show:

  • The model asked you to sign in, confirm something, or use a form — those flows take priority.
  • The reply is very short or has no clear follow-ups.
  • The feature is off for your organization.

Up to four follow-up actions can appear. They animate in shortly after the reply completes.

Enabling follow-up suggested actions​

An organization admin must turn the feature on:

  1. Open Organization Settings.
  2. Go to Features.
  3. Enable Suggested Actions.

When on, they appear in organization chats whenever the latest reply includes follow-up options. Turn the toggle off to disable.

Tips for follow-up suggested actions​

  • Use them as hints — If you are not sure what to ask next, the buttons often match good next steps.
  • Save typing — Click instead of retyping "Yes, explain more."
  • They follow the topic — Analysis, writing, technical help, and other domains are all supported.

Limitations​

  • They work best when the model’s reply is in English.
  • They reflect only the latest model message, not the full thread.
  • Detection can miss an offer or suggest something slightly off; you can always type your own message.
⌘J