Open Gemini Deep Research

A powerful open-source research assistant powered by Google's Gemini AI that performs deep, multi-layered research on any topic.

Features

• Automated deep research with adjustable breadth and depth
• Follow-up question generation for better context
• Concurrent processing of multiple research queries
• Comprehensive final report generation with citations
• Three research modes: fast, balanced, and comprehensive
• Progress tracking and detailed logging
• Source tracking and citation management

Prerequisites

• Python 3.9+
• Google Gemini API key
• Docker (if using dev container)
• VS Code with Dev Containers extension (if using dev container)

Installation

You can set up this project in one of two ways:

Option 1: Using Dev Container (Recommended)

1. Open the project in VS Code
2. When prompted, click "Reopen in Container" or run the "Dev Containers: Reopen in Container" command
3. Create a .env file in the root directory and add your Gemini API key:

   GEMINI_KEY=your_api_key_here
   

Option 2: Local Installation

1. Clone the repository:

   git clone <repository-url>
   cd open-gemini-deep-research

2. Create and activate a virtual environment (recommended):

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Create a .env file in the root directory and add your Gemini API key:

   GEMINI_KEY=your_api_key_here
   

Usage

Run the main script with your research query:

python main.py "your research query here"

Optional Arguments

• --mode: Research mode (choices: fast, balanced, comprehensive) [default: balanced]
• --num-queries: Number of queries to generate [default: 3]
• --learnings: List of previous learnings [optional]

Example:

python main.py "Impact of artificial intelligence on healthcare" --mode comprehensive --num-queries 5

Output

The script will:

1. Analyze your query for optimal research parameters
2. Ask follow-up questions for clarification
3. Conduct multi-layered research
4. Generate a comprehensive report saved as final_report.md
5. Show progress updates throughout the process

Project Structure

open-gemini-deep-research/
├── .devcontainer/
│   └── devcontainer.json
├── src/
│   ├── __init__.py
│   └── deep_research.py
├── .env
├── .gitignore
├── dockerfile
├── main.py
├── README.md
└── requirements.txt

How It Works

Research Modes

The application offers three research modes that affect how deeply and broadly the research is conducted:

1. Fast Mode

   • Performs quick, surface-level research
   • Maximum of 3 concurrent queries
   • No recursive deep diving
   • Typically generates 2-3 follow-up questions per query
   • Best for time-sensitive queries or initial exploration
   • Processing time: ~1-3 minutes

2. Balanced Mode (Default)

   • Provides moderate depth and breadth
   • Maximum of 7 concurrent queries
   • No recursive deep diving
   • Generates 3-5 follow-up questions per query
   • Explores main concepts and their immediate relationships
   • Processing time: ~3-6 minutes
   • Recommended for most research needs

3. Comprehensive Mode

   • Conducts exhaustive, in-depth research
   • Maximum of 5 initial queries, but includes recursive deep diving
   • Each query can spawn sub-queries that go deeper into the topic
   • Generates 5-7 follow-up questions with recursive exploration
   • Explores primary, secondary, and tertiary relationships
   • Includes counter-arguments and alternative viewpoints
   • Processing time: ~5-12 minutes
   • Best for academic or detailed analysis

Research Process

1. Query Analysis

   • Analyzes initial query to determine optimal research parameters
   • Assigns breadth (1-10 scale) and depth (1-5 scale) values
   • Adjusts parameters based on query complexity and chosen mode

2. Query Generation

   • Creates unique, non-overlapping search queries
   • Uses semantic similarity checking to avoid redundant queries
   • Maintains query history to prevent duplicates
   • Adapts number of queries based on mode settings

3. Research Tree Building

   • Implements a tree structure to track research progress
   • Each query gets a unique UUID for tracking
   • Maintains parent-child relationships between queries
   • Tracks query order and completion status
   • Provides detailed progress visualization through JSON tree structure

4. Deep Research (Comprehensive Mode)

   • Implements recursive research strategy
   • Each query can generate one follow-up query
   • Reduces breadth at deeper levels (breadth/2)
   • Maintains visited URLs to avoid duplicates
   • Combines learnings from all levels

5. Report Generation

   • Synthesizes findings into a coherent narrative
   • Minimum 3000-word detailed report
   • Includes inline citations and source tracking
   • Organizes information by relevance and relationship
   • Adds creative elements like scenarios and analogies
   • Maintains factual accuracy while being engaging

Technical Implementation

• Uses Google's Gemini AI for:
  • Query analysis and generation
  • Content processing and synthesis
  • Semantic similarity checking
  • Report generation
• Implements concurrent processing for queries
• Uses progress tracking system with tree visualization
• Maintains research tree structure for relationship mapping

Research Tree Implementation

The research tree is implemented through the ResearchProgress class that tracks:

• Query relationships (parent-child)
• Query completion status
• Learnings per query
• Query order
• Unique IDs for each query

The complete research tree structure is automatically saved to research_tree.json when generating the final report, allowing for later analysis or visualization of the research process.

Example tree structure:

{
  "query": "root query",
  "id": "uuid-1",
  "status": "completed",
  "depth": 2,
  "learnings": ["learning 1", "learning 2"],
  "sub_queries": [
    {
      "query": "sub-query 1",
      "id": "uuid-2",
      "status": "completed",
      "depth": 1,
      "learnings": ["learning 3"],
      "sub_queries": [],
      "parent_query": "root query"
    }
  ],
  "parent_query": null
}