Skip to main content

​
Overview

Followup suggestions automatically generate contextual prompt suggestions after each agent response, helping users discover what to ask next.

​
Quick Start

import { Agent, openai } from "@radaros/core";

const agent = new Agent({
  name: "assistant",
  model: openai("gpt-4o"),
  generateFollowups: true,
});

const output = await agent.run("Tell me about TypeScript");
console.log(output.followupSuggestions);
// ["How does TypeScript compare to JavaScript?",
//  "What are TypeScript generics?",
//  "How do I set up a TypeScript project?"]

​
Configuration

const agent = new Agent({
  name: "assistant",
  model: openai("gpt-4o"),
  generateFollowups: {
    count: 5,                    // number of suggestions (default: 3)
    model: openai("gpt-4o-mini"), // cheap model for generation
  },
});
OptionTypeDefaultDescription
countnumber3Number of followup suggestions to generate
modelModelProviderAgent’s modelModel used for generating suggestions

​
Output Format

Followup suggestions appear on the RunOutput object: 
interface RunOutput {
  text: string;
  followupSuggestions?: string[];  // new field
  // ... other fields
}

​
Best Practices

  • Use a cheap model (e.g. gpt-4o-mini) for followup generation to minimize cost
  • Followup generation is best-effort — if it fails, followupSuggestions will be an empty array
  • Suggestions are generated after output guardrails, so they won’t trigger re-validation