### **An Open-Source Alternative to OpenAI’s Deep Research: Open Deep Research**
After OpenAI introduced Deep Research, many open source versions comes up. Actually, there was already some built agents similar to deep research.
Huggingface published a blog post about that and they also run some agents for it.
Other great open implementations of Deep Research emerged from the community, specifically from
Each of these implementations use different libraries for indexing data, browsing the web and querying LLMs.
In this article, I want to mention [Nicolas Silberstein Camara](https://www.linkedin.com/in/nicolas-silberstein-camara/) for his great action against Openai Deep Research Premium. Here is the details and tutorial.
—
## **Open-Source Deep Research: Overview**
**Open Deep Research** is an **open-source clone** of OpenAI’s **Deep Research** experiment. Unlike OpenAI’s proprietary model, this project leverages **Firecrawl’s extract + search technology** combined with a **reasoning model** to conduct deep research across the web.
### **Open-Source Deep Research: Key Features**
– **Firecrawl Extract + Search**
– Feeds real-time data to the AI via **search**.
– Extracts **structured data** from multiple websites.
– **Next.js App Router**
– Uses **React Server Components (RSCs)** for efficient rendering.
– Supports **server-side rendering** for performance optimization.
– **AI SDK Integration**
– Supports multiple **LLM providers**:
– OpenAI (default: **gpt-4o**)
– **Anthropic, Cohere, DeepSeek**, and more.
– **Advanced UI Components**
– Styled with **Tailwind CSS**.
– Uses **shadcn/ui** with **Radix UI** for flexible component handling.
– **Data Persistence**
– Uses **Vercel Postgres (Neon)** for chat history and user data.
– Stores files efficiently with **Vercel Blob**.
– **Authentication System**
– Implemented using **NextAuth.js** for secure user login.
—
## **Open Deep Research: Installation Guide (Run Locally)**
### **1. Clone the Repository**
Open your terminal and run:
“`bash
git clone https://github.com/nickscamara/open-deep-research.git
cd open-deep-research
“`
### **2. Install Dependencies**
Install **pnpm** if not installed:
“`bash
npm install -g pnpm
“`
Then install all dependencies:
“`bash
pnpm install
“`
### **3. Set Up Environment Variables**
You’ll need to define environment variables in `.env` using the `.env.example` file as a reference.
To automatically configure environment variables:
“`bash
vercel env pull
“`
Or manually create a `.env` file and include:
“`bash
OPENAI_API_KEY=your_openai_api_key
FIRECRAWL_API_KEY=your_firecrawl_api_key
AUTH_SECRET=your_auth_secret
“`
⚠️ **Do not commit the `.env` file** to avoid exposing sensitive API keys.
### **4. Run Database Migrations**
“`bash
pnpm db:migrate
“`
### **5. Start the Application**
“`bash
pnpm dev
“`
Your app should now be running at:
🔗 **[http://localhost:3000](http://localhost:3000/)**
—
## **Alternative Deployment: One-Click Deploy to Vercel**
If you prefer **not to run it locally**, you can **deploy to Vercel** in one click:
[](https://vercel.com/new/clone?repository-url=https%3A%2F%2Fgithub.com%2Fnickscamara%2Fopen-deep-research&env=AUTH_SECRET,OPENAI_API_KEY,OPENROUTER_API_KEY,FIRECRAWL_API_KEY&envDescription=Learn%20more%20about%20how%20to%20get%20the%20API%20Keys%20for%20the%20application&envLink=https%3A%2F%2Fgithub.com%2Fvercel%2Fai-chatbot%2Fblob%2Fmain%2F.env.example&demo-title=AI%20Chatbot&demo-description=An%20Open-Source%20AI%20Chatbot%20Template%20Built%20With%20Next.js%20and%20the%20AI%20SDK%20by%20Vercel.&demo-url=https%3A%2F%2Fchat.vercel.ai&stores=[{%22type%22:%22postgres%22},{%22type%22:%22blob%22}])
—
## **Model Providers**
By default, the project uses **OpenAI’s GPT-4o**.
However, it **supports multiple LLM providers** via **Vercel’s AI SDK**, including:
– **Anthropic** (Claude)
– **Cohere**
– **DeepSeek**
– **TogetherAI**
– **OpenRouter**
### **Switching Models**
Modify the `.env` file:
“`bash
REASONING_MODEL=deepseek-reasoner
BYPASS_JSON_VALIDATION=true
“`
—
## **Adding Model Dependencies**
If you want to **use a model other than GPT-4o**, install the respective dependency.
### **DeepSeek AI Model**
“`bash
pnpm add @ai-sdk/deepseek
“`
### **TogetherAI Model**
“`bash
pnpm add @ai-sdk/togetherai
“`
🚨 **Check TogetherAI rate limits:**
🔗 [Rate Limit Info](https://docs.together.ai/docs/rate-limits)
—
## **Reasoning Model Configuration**
This project includes a **reasoning model** for **structured outputs** such as research analysis, data extraction, and document summarization.
| Provider | Models Supported | Notes |
|————|————————|——–|
| **OpenAI** | `gpt-4o`, `o1`, `o3-mini` | Native JSON support |
| **TogetherAI** | `deepseek-ai/DeepSeek-R1` | Requires `BYPASS_JSON_VALIDATION=true` |
| **DeepSeek** | `deepseek-reasoner` | Requires `BYPASS_JSON_VALIDATION=true` |
### **Key Notes**
– **GPT-4o, o1, o3-mini** → Natively support **structured JSON outputs**.
– **DeepSeek & TogetherAI** → Need `BYPASS_JSON_VALIDATION=true`.
– **If no model is set, it defaults to `o1-mini`**.
– **If an invalid model is chosen, it falls back to `o1-mini`**.
To use **DeepSeek as the reasoning model**, add this to `.env`:
“`bash
REASONING_MODEL=deepseek-reasoner
BYPASS_JSON_VALIDATION=true
“`
—
## **Open-Source Deep Research: Conclusion**
**Open Deep Research** by *Nicolas Silberstein Camara* is a powerful, **open-source alternative** to OpenAI’s **Deep Research**.
It allows users to **autonomously research the web**, retrieve **structured data**, and leverage **multiple AI models**.
By following the steps above, you can:
✅ **Run it locally**
✅ **Deploy it to Vercel**
✅ **Customize model providers**
For the latest updates, visit the **[GitHub Repository](https://github.com/nickscamara/open-deep-research)**.
**Developer:** *Nicolas Silberstein Camara*
**GitHub Repository:** [Github](https://github.com/nickscamara/open-deep-research)
**Demo:** [Live Demo](https://x.com/nickscamara_/status/1886459999905521912)