GARGI

Generally A Really Good Interface

Terminal-native AI companion with animated ASCII face, voice output, and personality that shifts with time of day.

What it does

• Animated 60×32 ASCII portrait (blinks, talks, thinks)
• Streaming chat with NVIDIA NIM API (free, no credit card)
• Voice output via TTS (Windows SAPI / macOS say / espeak)
• Time-aware mood system
• Slash commands (/mood, /voice, /model, /setup, etc.)
• Full keyboard shortcuts

Quick start

Direct USAGE

Download the latest release and open gargi.exe

If Windows blocks the executable:

1. Open Windows Security.
2. Go to App & browser control.
3. Open Smart App Control settings.
4. Turn it off.
5. Launch the application again.
6. It will open in CMD/Powershell.
7. Use /help to get context, /setup for adding the API key.
8. Get the API key (free, no credit card) from build.nvidia.com.
9. And, get your agent running.

Use by Source Code

git clone https://github.com/lakshyabuilds/gargi.git
cd gargi
pip install textual rich openai
python gargi.py

First run asks for NVIDIA NIM API key. Get one free at build.nvidia.com.

Build standalone exe

python build.py pyinstaller     # onedir mode (fast, ~30s)
python build.py nuitka          # compile to C (slower, fewer AV flags)

Windows: double-click build_exe.bat.

Output in dist_exe/<method>/gargi/. No Python needed on target machine.

Structure

gargi/
├── gargi.py              # entry point
├── gargi_app/
│   ├── tui.py            # Textual TUI
│   ├── agent.py          # NIM client
│   ├── persona.py        # personality
│   ├── ascii_art.py      # animated face
│   └── voice.py          # TTS engine
├── build.py              # exe builder
├── build_exe.bat         # Windows one-click
└── requirements.txt

Commands

/setup      connect/replace API key
/models     list free models
/model <n>  switch model
/voice      toggle TTS
/mood       show current vibe
/face <m>   lock expression
/name <you> set your name
/reset      clear memory
/clear      clear screen
/help       show all
/quit       exit

Keyboard: Ctrl+Q quit · Ctrl+R reset · Ctrl+T voice · Ctrl+S stop · Ctrl+F cycle face · Esc focus input.

Screenshots

Tech Stack

• UI: Textual (terminal UI framework)
• AI: NVIDIA NIM (free LLM API, OpenAI-compatible)
• Voice: Windows SAPI / macOS say / espeak
• Pure Python, no Rust, no Docker, no native deps

Default model: nvidia/llama-3.3-nemotron-super-49b-v1.5. Switch with /model.

Terminal requirements

• ~90 cols × ~36 rows minimum
• Windows Terminal, PowerShell 7, or chcp 65001 on legacy cmd.exe
• Python 3.10+

License

MIT