Introduction

Finding a truly private way to conduct deep AI-assisted research without relying on cloud-based APIs or subscription fees is a constant challenge for developers and privacy-conscious users. Crazy-Deep-Research-GPT solves this by providing a hyper-functional, sleek, and fully local web-based AI interface built with Node.js and powered by Ollama. With over 5 GitHub stars, this open-source tool allows users to run powerful local LLMs like DeepSeek-R1 and Qwen2.5-VL, enabling deep exploration and quick answers with zero tracking and no API keys required.

What Is Crazy-Deep-Research-GPT?

Crazy-Deep-Research-GPT is a local AI research interface that provides a structured web UI for interacting with Ollama-hosted models. It is written in Node.js and utilizes a privacy-first architecture that ensures all AI processing happens on the user’s own hardware. The project is licensed under the MIT License, allowing for wide adoption and modification.

The tool is designed to move beyond simple chat interfaces by offering specialized research modes that control how the AI behaves during complex queries. By integrating HTML-to-Markdown conversion, it transforms web content into a format that is easier for LLMs to process and for users to export, making it a comprehensive tool for local AI research.

Why Crazy-Deep-Research-GPT Matters

Most AI research tools are locked behind monthly subscriptions or require the use of cloud APIs that compromise user privacy. Crazy-Deep-Research-GPT fills this gap by bringing the “Deep Research” experience to the local machine. It allows users to maintain complete control over their data, avoiding the risk of data leaks or vendor lock-in associated with cloud AI services.

The project’s focus on local execution is particularly valuable for those working with sensitive information or in environments where internet connectivity is limited. By leveraging Ollama, it provides a seamless way to run state-of-the-art local models like DeepSeek-R1, which are increasingly capable of complex reasoning and research tasks.

As the local LLM ecosystem grows, having a dedicated interface that is optimized for research rather than just chatting is critical. Crazy-Deep-Research-GPT provides the structured approach needed to turn a local LLM into a professional research assistant.

Key Features

• HTML-to-Markdown Conversion: Uses Turndown (via Cheerio) to extract and convert web content to Markdown, ensuring that AI outputs are readable, structured, and easily exportable.
• Ollama Model Integration: Seamlessly integrates with Ollama to run local LLMs such as qwen2.5vl:3b and deepseek-r1:1.5b, removing the need for external API keys.
• Intelligent Research Modes: Offers multiple research depths and breadths settings, including Normal, Deep Think, Search, Deep Research Lite, and Heavy Duty Deep Research, for fine-grained control over response quality.
• Customizable AI Settings: Provides a dedicated settings modal for adjusting system prompts, selecting primary and auxiliary models, and tweaking temperature sliders for creativity.
• WebSocket Concurrency Limits: Allows users to manage WebSocket concurrency limits to optimize performance based on their hardware capabilities.
• Privacy-First Architecture: Built with zero tracking and no API keys, ensuring that all data remains on the local machine.
• Responsive UI: Developed with Tailwind CSS and Alpine.js, the interface is polished, mobile-friendly, and provides a smooth user experience.
• Location-Based Support: Enables the AI to handle queries involving local context, such as nearby restaurants or weather updates, when configured correctly.

How Crazy-Deep-Research-GPT Compares

When comparing Crazy-Deep-Research-GPT to alternatives like Open WebUI, the primary differentiator is the specialized focus on research workflows. While Open WebUI is a general-purpose chat interface, Crazy-Deep-Research-GPT is built specifically for deep exploration. Its integrated research modes (like “Heavy Duty Deep Research”) allow for more structured and comprehensive responses than a standard chat prompt.

Compared to cloud-based solutions like GPT-4o, the tradeoff is primarily in raw model power. While cloud models are generally larger and more capable, Crazy-Deep-Research-GPT allows you to run models like DeepSeek-R1 locally, which are designed for reasoning. The absolute privacy and lack of subscription fees make it a superior choice for those who prioritize data sovereignty over absolute peak performance.

Getting Started: Installation

To get Crazy-Deep-Research-GPT running on your machine, you will need Node.js and Ollama installed. Follow these steps to set up the environment.

Prerequisites

Ensure you have Node.js installed on your system. You must also have Ollama installed and running in the background.

Clone and Install

git clone https://github.com/crazystuffxyz/Crazy-Deep-Research-GPT.git
cd Crazy-Deep-Research-GPT
npm install

Configure Ollama Models

The project recommends using specific models for the best experience. Pull these models using the Ollama CLI:

ollama pull qwen2.5vl:3b
ollama pull deepseek-r1:1.5b

Once the models are pulled, you can verify their availability in the Advanced Settings modal within the web interface.

How to Use Crazy-Deep-Research-GPT

Using the interface is straightforward. Once you have installed the dependencies and pulled the models, start the local server:

npm start

Open your web browser and navigate to http://localhost:3000. You will see a sleek, dark-themed interface with a query input area at the bottom.

Select a research mode from the dropdown menu. For a quick answer, use Normal. For a complex topic that requires reasoning, use Deep Think or Heavy Duty Deep Research. Enter your query and send it to receive a response generated by your local LLM.

Code Examples

The project is built on a Node.js backend. Below is a conceptual example of how the application handles a query to the Ollama API, reflecting the project’s architecture.

// Example of the internal query handling logic
const axios = require('axios');

async function runOllamaModel(query, model = 'deepseek-r1:1.5b') {
  try {
    const response = await axios.post('http://localhost:11434/api/generate', {
      model: model,
      prompt: query,
      stream: false
    });
    return response.data.response;
  } catch (error) {
    console.error('Ollama API Error:', error);
    }
}

// Usage in the application
runOllamaModel('What is the Kepler Equation?').then(console.log);

This example demonstrates the basic communication between the Node.js server and the Ollama local API, which is the core of how Crazy-Deep-Research-GPT provides its research capabilities.

Real-World Use Cases

Crazy-Deep-Research-GPT is ideal for several specific scenarios where local AI is a requirement.

• Privacy-Sensitive Research: A security researcher analyzing malware or analyzing proprietary codebases without uploading data to a cloud provider.
• Offline Knowledge Base: A developer working in a secure environment or on a travel, using a local LLM to research technical documentation and solve coding problems.
• Comparative Model Testing: An AI enthusiast testing how different local models (like Qwen vs. DeepSeek) handle the same complex research query across different research modes.
• Local Content Synthesis: A writer using the HTML-to-Markdown conversion to synthesize information from various local web pages into a structured research paper.

Contributing to Crazy-Deep-Research-GPT

Contributions are welcome and encouraged. Since the project is licensed under the MIT License, users can fork the repository and submit pull requests to improve the AI integration, UI/UX polishing, or add new research modes.

To contribute, start by opening an issue to discuss a feature request or report a bug. The project maintainer encourages a privacy-first approach to any new features, ensuring that no tracking or cloud-based dependencies are introduced.

Community and Support

Support for Crazy-Deep-Research-GPT is primarily managed through the GitHub repository. Users can report bugs, request features, and discuss improvements in the Issues section of the repo.

The project is a small, focused open-source project. While there is no dedicated Discord or Slack, the GitHub Discussions and Issues sections are the best place to find help and the community of users running local AI research tools.

Conclusion

Crazy-Deep-Research-GPT is a powerful tool for anyone who wants to turn their local machine into a private, professional research station. By combining the power of Ollama and a research-optimized UI, it removes the cost and privacy risks associated with cloud AI.

It is the right choice when you need a dedicated research interface that prioritizes privacy and data sovereignty. While it is a smaller project compared to general-purpose AI interfaces, its specialized focus on deep research makes it a standout choice for the local LLM community.

Star the repo, try the quickstart, and join the community of local AI enthusiasts.

Resources

• GitHub Repository: Crazy-Deep-Research-GPT
• Ollama Official Website