A lightweight deep research framework based on progressive search and cross-evaluation.

English | 简体中文

Introduction

DeepResearch focuses on solving complex information analysis problems and supports local deployment for individual developers. Through modular context assembly (covering knowledge bases, tool descriptions, and interaction history) and progressive search optimization, it builds an intelligent research workflow of "Task Planning → Tool Calling → Evaluation & Iteration". This workflow effectively alleviates the issues of attention dispersion and information loss when large models process long contexts. Meanwhile, it allows users to introduce custom research workflows, ensuring the output content has thematic focus, comprehensive argumentation, and logical hierarchy.

Features:

• Delivers high-quality results without requiring model customization.
• Enables collaboration between small and large models to boost research efficiency and control usage costs.
• Reduces large model hallucinations through knowledge extraction and cross-evaluation verification.
• Supports lightweight deployment and flexible configuration.

Framework:

Samples:

Deep Research Products Global and Domestic Landscape Analysis

Global AI Agent Products Panoramic Analysis: Core Capabilities and Application Scenarios

Quick Start

This section will explain how to configure the local runtime environment for DeepResearch, or you can visit SparkDesk and go to the "Analysis & Research" tab for an online experience.

1. Environment Setup

• Recommended Python version: 3.10.0 (using other versions may cause dependency issues).
• Clone the repository.

  git clone [email protected]:iflytek/DeepResearch.git 

• Ensure you have Poetry installed(Recommended version:2.2.1).

  poetry --version
  # If Poetry is not installed yet, you can try installing it via the following methods
  # Install Poetry on Bash
  curl -sSL https://install.python-poetry.org | python3 -
  # Install Poetry on PowerShell
  (Invoke-WebRequest -Uri https://install.python-poetry.org -UseBasicParsing).Content | python -

• Set up your runtime environment

  cd DeepResearch
  poetry install
  poetry env activate

2. Environment Configuration

According to DeepResearch's workflow, you need to fill in LLM configuration parameters for each module (for the Planner, it is recommended to use a reasoning LLM, such as DeepSeekR1).

Edit config/llms.toml and config/search.toml files provide your actual API keys and configuration values:

• api_base/api_key/model: OpenAI-compatible API, from Iflytek MaaS or other platforms.

• jina_api_key or tavily_api_key: Get your key from Jina or Tavily for web page reading.

3. Running DeepResearch

Enjoy the moment.

poetry run python -m src.run

Contributing

We welcome contributions of all kinds! Please see our Contributing Guide

Support

• Community Discussion: GitHub Discussions
• Bug Reports: Issues

License

This project is licensed under the Apache 2.0 License.