Bantz v2

Local-first AI terminal assistant. Runs on your machine, talks to your shell, understands Turkish.

Built with Ollama + Textual + MarianMT.

What it does

You type in Turkish (or English). Bantz figures out what you need — runs a shell command, checks system stats, fetches weather, pulls news headlines — and responds in Turkish.

No cloud APIs for core functionality. LLM runs locally via Ollama. Translation runs locally via MarianMT.

Tools

Tool	What it does	Source
shell	Runs bash commands with security controls	blocked/destructive command detection
system	CPU, RAM, disk, uptime	psutil
filesystem	Read/write/list files	path-restricted to home directory
weather	Current conditions + 3-day forecast	wttr.in (no API key)
news	Headlines from Hacker News + Google News	HN API + RSS, 15min cache

Architecture

User input (Turkish)
  → MarianMT bridge (TR → EN)
  → Quick route (keyword match) or LLM router (Ollama)
  → Tool execution
  → Finalizer (responds in Turkish, time-aware)
  → Textual TUI output

Key modules:

src/bantz/
├── __main__.py          # CLI entry point (--doctor, --once)
├── app.py               # Textual TUI
├── config.py            # Pydantic settings (.env)
├── core/
│   ├── brain.py         # Orchestrator — routing, LLM calls, tool dispatch
│   ├── location.py      # IP geolocation (ipinfo.io) with session cache
│   └── time_context.py  # Time-of-day awareness (greeting, prompt hints)
├── i18n/
│   └── bridge.py        # TR↔EN translation (MarianMT, local inference)
├── llm/
│   └── ollama.py        # Ollama HTTP client
└── tools/
    ├── __init__.py      # ToolResult, BaseTool, ToolRegistry
    ├── shell.py         # Bash execution
    ├── system.py        # System metrics
    ├── filesystem.py    # File operations
    ├── weather.py       # wttr.in weather
    └── news.py          # HN + Google News

Requirements

• Python 3.11+
• Ollama running locally with a model (e.g. qwen2.5-coder:7b)
• Linux (tested on Linux Mint 21.3)

Setup

git clone git@github.com:miclaldogan/bantzv2.git
cd bantzv2
bash setup.sh

setup.sh creates a venv, installs dependencies, and copies .env.example to .env.

To install translation support (MarianMT):

source .venv/bin/activate
pip install transformers torch sentencepiece

Usage

# Full TUI
python -m bantz

# Single command mode
python -m bantz --once "diskimi kontrol et"

# Health check
python -m bantz --doctor

Configuration

Copy .env.example to .env and edit as needed:

BANTZ_OLLAMA_MODEL=qwen2.5-coder:7b
BANTZ_OLLAMA_BASE_URL=http://localhost:11434
BANTZ_SHELL_TIMEOUT=30
BANTZ_SHELL_CONFIRM_DESTRUCTIVE=true
BANTZ_BRIDGE_ENABLED=true

How routing works

1. Quick route — keyword matching for obvious patterns (disk, ram, weather, news, file creation). No LLM call needed.
2. LLM router — for ambiguous requests, Ollama picks the right tool and args.
3. Command generator — for write/create operations, LLM generates a bash command from both the Turkish original and English translation (prevents content loss from translation).

License

MIT