👋🏻 Hi everyone! GPT Unplugged🔌 is an interconnected, local server-based, fully customizable, privacy-focused research agent with consistent memory across models. Focused on giving you the ultimate control of your data and conversations.

Scenario 1

Tomorrow if there is No internet access this can be your Local personal assistant with new knowledge. Feed it your documents or books or pdfs and keep the conversation rolling. It has cross model memory so none of your previous chat or context is lost in between. Switch moel and keep the conversation rolling.

Scenario 2

In your local network just run on one machine and use in all your personal devices. All device under the same WIFI network. So basically your personal bot whom you have deep conversations with when youre sleepy or in bed...
(Watch the movie 'HER' you'll get the reference...)

✨ Features

• Complete Privacy: All data stays local on your machine
• Cross-Device Access: Use Tailscale to access the service across devices in your local network
• Multi-Platform: Full support for PC, mobile, and tablet devices
• Ollama Integration: Host models locally for complete independence (Local inference for full independence)
• Model Supports: Qwen 32B , GPT OSS 20B ,GPT OSS 120B, Kimi K2 Instruct , LLama 3.3 70B , All Ollama Models.
• Supports Multiturn, Long context memory, Rag, rerankers...

📦 Installation

Step 1️: Installing it. (One-time thing)

Create a new folder where you want everything to be stored. Open terminal inside the folder, and run these commands one after another(Inside the same terminal):

git clone https://github.com/t-majumder/GPT-Unplugged.git
cd GPT-Unplugged
python -m venv gpt
pip install -r requirements.txt

Close everything. Installation is done.

Step 2: Setting the free API key (After this technical stuff is done)

-> Go to the Groq website : https://console.groq.com/keys (Get your free API key here)
-> Paste your API key here. Go to ->

GPT-Unplugged/
├── backend/
│   ├── .env    # Paste your GROQ API key here

🎯Running the project:

1. Go to the GPT-Unplugged folder.
2. Open terminal inside the folder and run this command based on Windows or Linux.

Finally Run:

run run.ps1

or

bash run.sh

Now you can access the website from your browser. (copy-paste this link inside your browser)

http://localhost:5173/

🔧 Advanced Configuration

• You can configure the prompt (Behavior) and add models using the config files directly.

GPT-Unplugged/
├── config/
│   ├── prompt.yaml    # Add your custom prompt
│   ├── models.yaml    # Add the model name
│   ├── hyperparams.yaml    # Other hyperparams you can tune as well

🦙 Ollama Integration

For local model hosting with Ollama: Just install Ollama and download the models. (done)

ollama run gemma3:4b

This command downloads the model. You can download and add other models as well...

⚡Local Server

To access GPT Unplugged from other devices on your network:

1. Install Tailscale on all devices
2. Connect your devices to the same Tailscale network
3. Access the application using your Tailscale IP

http://Your Tailscale IP:5173/

Multi Device Support

📄 License

MIT Liscense

⭐ Support

If you find this project useful, please consider giving it a star on GitHub!

Will be adding tools in version 3.

---

Note: This project prioritises your privacy. All data processing happens locally on your machine, and you maintain complete control over your information.