Skip to main content

Additional features: Follow-up Suggestions

When to use​

Turn on Follow-up Suggestions to speed up chats and guide users toward the next best step—great for support triage, research flows, data entry, or any multi-step workflow where users often ask similar next questions.

What it does​

  • After every agent message, the UI shows 3 suggested replies.
  • Suggestions are predicted by the same LLM your agent uses and reflect the current conversation, base instruction, KBs, and available tools.
  • Clicking a suggestion auto-sends it as a user message to the agent (no edit dialog).

If a user needs to modify wording, they should type a custom message instead of clicking a suggestion.

Enable in the UI​

  1. Open your agent.
  2. In Additional features, click Add.
  3. Toggle Follow-up suggestions ON in the drawer.
  4. Close the drawer; the feature now appears under Additional features.

Best practices​

  • Keep them safe: Require confirmation for any suggestion that would trigger write/update tool calls.
  • Action-first phrasing: “Create a P1 ticket,” “Extract tables from the upload,” “Draft a reply email.”
  • Domain-aware: Mention the tools/KBs the agent should prefer; suggestions will skew more relevant.
  • Reduce noise: If suggestions feel generic, tighten the base instruction and the agent’s response format.
  • Measure & iterate: In Preview, click suggestions through common flows and refine wording for clarity.

Troubleshooting​

  • No suggestions shown: Re-open Additional features → Add and ensure the toggle is ON for this agent.
  • Irrelevant suggestions: Strengthen the base instruction (role, goals, available tools) and ensure the conversation includes the needed context.
  • Risky one-click actions: Add a guardrail in the base instruction: “Always ask for confirmation before executing actions that change external systems.”