Show HN: Deep Research – Open-Source Customizable Reasoning Framework for Devs

6 days ago 3

Deep Research is an open source library for conducting deep, multi-hop research with reasoning capabilities. It performs focused web searches with recursive exploration to provide comprehensive, evidence-backed answers to complex questions.

Deep Research is designed to be your comprehensive solution for AI-powered research:

🧠 Advanced multi-hop reasoning for complex questions
🌐 Real-time web search with recursive exploration
🔍 Automatic subquery generation for comprehensive coverage
📊 Intelligent depth and breadth control for research thoroughness
📝 Evidence-based report generation with proper citations
📚 Automatic bibliography generation with source tracking
🔄 Iterative research cycles for deeper understanding
🤖 Multi-model support with specialized reasoning capabilities
⚡ Flexible configuration for customizing research parameters
📈 Scalable from simple inquiries to complex research problems

Concept Description

Deep Thinking	The system breaks down a question into logical parts, reasons through them independently, and synthesizes an answer.
Deep Research	The system performs multi-hop, focused web searches, compares the findings, and composes an evidence-backed answer.

npm i deep-research # or yarn add deep-research # or bun i deep-research

import { createDeepResearch } from "deep-research"; // Create instance using the factory function with default settings const deepResearch = createDeepResearch({ OPENAI_API_KEY: process.env.OPENAI_API_KEY, GEMINI_API_KEY: process.env.GEMINI_API_KEY, DEEPINFRA_API_KEY: process.env.DEEPINFRA_API_KEY, JIGSAW_API_KEY: process.env.JIGSAW_API_KEY, }); // Research prompt const prompt = "What are the recent developments in quantum computing?"; // Generate research report const result = await deepResearch.generate(prompt); console.log(result.data.text); console.log(result.data.bibliography);

import { createDeepResearch } from "open-deep-research"; import { createGoogleGenerativeAI } from "@ai-sdk/google"; import { createDeepInfra } from "@ai-sdk/deepinfra"; import { createOpenAI } from "@ai-sdk/openai"; // Initialize AI providers const gemini = createGoogleGenerativeAI({ apiKey: process.env.GEMINI_API_KEY, }); const deepinfra = createDeepInfra({ apiKey: process.env.DEEPINFRA_API_KEY, }); const openaiProvider = createOpenAI({ apiKey: process.env.OPENAI_API_KEY, }); // Get model instances const geminiModel = gemini("gemini-2.0-flash"); const deepseekModel = deepinfra("deepseek-ai/DeepSeek-R1"); const openaiModel = openaiProvider("gpt-4o"); // Create instance with custom configuration const deepResearch = createDeepResearch({ report: { maxOutputTokens: 30000, // Hard limit on report length targetOutputTokens: 10000, // Target report length }, depth: { maxDepth: 4, // How many iterations of research to perform }, breadth: { maxBreadth: 3, // How many subqueries to generate }, models: { default: openaiModel, // Custom models from AI SDK reasoning: deepseekModel, output: geminiModel, }, logging: { enabled: true, // Enable console logging }, }); // Research prompt const prompt = "What are the recent developments in quantum computing?"; // Generate research report const result = await deepResearch.generate(prompt); console.log(result.data.text); console.log(result.data.bibliography);

Configuration Options for Deep Research

Category Option Type Default Description

depth	maxDepth	Number	3	Controls how many iterations of research the system will perform. Higher values allow for more thorough, multi-hop research. The system will continue researching until it has a complete answer or reaches this limit.
breadth	maxBreadth	Number	3	Controls how many subqueries are generated for each research iteration. Higher values enable wider exploration of the topic. Determines how many parallel search paths are pursued.
report	maxOutputTokens	Number	32000	Hard upper limit on the length of the final report. Must be greater than targetOutputTokens.
	targetOutputTokens	Number	optional	The ideal length for the generated report. The system will try to produce a report of approximately this length.
models	default	LanguageModelV1	GPT-4.1	The primary model used for most operations.
	reasoning	LanguageModelV1	DeepSeek-R1	Model used for reasoning about search results.
	output	LanguageModelV1	GPT-4.1	Model used for generating the final report.
logging	enabled	Boolean	false	When set to true, enables detailed console logging. Helpful for debugging and understanding the research process.
API Keys	JIGSAW_API_KEY	String	required	For accessing the JigsawStack API for web searches.
	OPENAI_API_KEY	String	required	For OpenAI model access.
	GEMINI_API_KEY	String	required	For Google Gemini model access.
	DEEPINFRA_API_KEY	String	required	For DeepInfra model access.