https://github.com/dimuzzo/ai-terminal
https://github.com/dimuzzo/ai-terminal
Last synced: 29 days ago
JSON representation
- Host: GitHub
- URL: https://github.com/dimuzzo/ai-terminal
- Owner: dimuzzo
- Created: 2026-04-09T18:15:16.000Z (about 2 months ago)
- Default Branch: main
- Last Pushed: 2026-04-09T21:09:48.000Z (about 2 months ago)
- Last Synced: 2026-04-09T23:15:25.463Z (about 2 months ago)
- Language: TypeScript
- Size: 234 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# 🚀 AI-Terminal: Self-Learning Local CLI
A blazing-fast, custom terminal emulator powered by Rust (Tauri), React, and xterm.js, integrated with a local LLM (Qwen) via Ollama.
This is not just an AI chat wrapper. The ultimate goal of this project is to collect user commands and AI outputs locally to build a personalized dataset for future fine-tuning (LoRA), making the terminal natively adapted to your specific workflow.
## Features
- **Local AI Inference:** Completely offline, privacy-first AI responses using local models.
- **High Performance:** Built on Tauri and Rust for minimal overhead.
- **Native Rendering:** Uses `xterm.js` for an authentic terminal experience.
- **Data Collection:** Silently builds a `.jsonl` dataset of prompt-completion pairs.
- **Custom System Commands:** Intercepts standard OS commands (like `dir` or `ls`) for clean, privacy-focused outputs.
## Prerequisites
- [Node.js](https://nodejs.org/) & npm
- [Rust](https://rustup.rs/)
- [Ollama](https://ollama.com/)
- **For Training:** Python 3.11 & NVIDIA GPU
## Installation & Setup (Terminal Application)
1. **Clone the repository:**
```bash
git clone https://github.com/dimuzzo/ai-terminal.git
cd ai-terminal
```
2. **Install frontend dependencies:**
```bash
npm install
```
3. **Start the local AI model:**
Open a separate terminal and run:
```bash
ollama run qwen2.5:1.5b
```
4. **Launch the terminal:**
```bash
npm run tauri dev
```
## Local Fine-Tuning Pipeline (WIP)
To train the model on your collected data (`dataset.jsonl`), we use Unsloth for highly optimized, low-VRAM LoRA fine-tuning.
1. Navigate to the training directory:
```bash
cd training-pipeline
```
2. Create and activate a Python 3.11 virtual environment:
```bash
# Windows
C:\path\to\python3.11.exe -m venv venv
.\venv\Scripts\activate
```
3. Install dependencies:
```bash
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"
pip install trl peft accelerate bitsandbytes
```
4. Run the training script:
```bash
python train.py
```
5. Export the trained adapter to GGUF format:
```bash
python export.py
```
6. Load your new custom brain into Ollama:
```bash
cd ai-terminal-custom_gguf
ollama create ai-terminal-custom -f Modelfile
```
## Stack
- **Frontend**: React, TypeScript, xterm.js
- **Backend**: Rust, Tauri
- **AI Engine**: Ollama (Inference) / PyTorch & Unsloth (Training)