Deep Research

Overview

DeepResearch performs multi-round web searches. It generates queries, searches, evaluates if it has enough information, refines queries, and repeats until satisfied. Then it synthesizes a report.

Question -> Generate queries -> Search -> Sufficient?
                                            |
                                    No -> Refine queries -> Search -> ...
                                    Yes -> Synthesize report

Usage

import { DeepResearch } from "@schift-io/sdk";

const research = new DeepResearch(
  {
    maxIterations: 5,
    resultsPerSearch: 10,
    queriesPerIteration: 3,
    synthesisModel: "gpt-4o",
    webSearch: {
      provider: "tavily",
      providerApiKey: "tvly-xxx",
    },
  },
  llmFn,
);

const report = await research.run("AI agent framework trends 2026");

console.log(report.answer);           // Synthesized report
console.log(report.sources.length);   // Number of sources
console.log(report.iterations);       // Iterations used
console.log(report.totalQueries);     // Total queries executed

The llmFn parameter is a function that calls any chat completions API:

const llmFn = async (messages: Array<{ role: string; content: string }>) => {
  const resp = await fetch("https://api.openai.com/v1/chat/completions", {
    method: "POST",
    headers: {
      Authorization: `Bearer ${process.env.OPENAI_API_KEY}`,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({ model: "gpt-4o-mini", messages }),
  });
  const data = await resp.json();
  return data.choices[0].message.content;
};

As an Agent Tool

const agent = new Agent({
  tools: [research.asTool()],
  ...
});

Registered as deep_research. The agent calls it for questions that need thorough, multi-source research.

Configuration

Option	Type	Default	Description
maxIterations	number	3	Maximum research rounds
resultsPerSearch	number	5	Results per query
queriesPerIteration	number	2	Queries generated per round
queryModel	ModelId | string	"gpt-4o-mini"	Model for query generation
synthesisModel	ModelId | string	"gpt-4o-mini"	Model for final report
webSearch	WebSearchConfig	—	Web search provider config