DOOM

Your terminal understands English. You stop Googling commands.

DOOM is a local AI terminal agent that translates natural-language requests into zsh commands. Type what you want in plain English, preview the command, and run it. One request → one command. Private, fast, runs on your machine.

$ doom "how much ram does this mac have"

╭───── command ─────╮
│ sysctl hw.memsize │
╰───────────────────╯
[Displays the total physical memory installed in the machine.]
 run    cancel    edit (n):

Installation

From PyPI

pip install doomcli
doom "your request here"

On first run, DOOM will ask permission to download the model (~850MB) from Hugging Face. The model is cached in ~/.doom/model/ for future use.

From source

git clone https://github.com/Ahmad170412/DoomCLI.git
cd DoomCLI
pip install -e .
doom "list all running processes"

Standalone (no pip)

# Clone with the model
git clone --recurse-submodules https://github.com/Ahmad170412/DoomCLI.git
cd DoomCLI

# Create venv
python3 -m venv .venv
.venv/bin/pip install mlx mlx-lm rich prompt_toolkit

# Run
./doom "your request here"

Usage

One-shot mode

doom "create a new react project called my-app"
doom "what's my public ip"
doom "add OPENROUTER_API_KEY to my zshrc"
doom "find all files larger than 100mb"

Interactive REPL

doom

Drops you into a loop where you can type request after request. Type exit or Ctrl-C to quit.

Flags

Edit flow

When you see a command you want to tweak:

 run    cancel    edit (n): e
edit command: echo 'export API_KEY="sk-123"' >> ~/.zshrc

The command is pre-filled — arrow keys, backspace, type your changes, Enter.

How It Works

DOOM runs a fine-tuned Qwen2.5-Coder-1.5B model locally on your machine. No API calls. No data leaves your computer.

1. You type a natural-language request
2. The model generates exactly one zsh command + a short explanation
3. DOOM previews the command in a clean panel
4. You confirm: [y] run, [n] cancel, or [e] edit the command
5. DOOM executes the command and shows the output

Model

The fine-tuned model is available on Hugging Face: Ahmad170412/Doom-V1

Safety

DOOM has a 30-pattern destructive classifier that flags dangerous commands before execution:

• File deletion (rm, rmdir, find -delete)
• Process control (kill, pkill, dd)
• Privilege escalation (sudo, eval, curl | sh)

Destructive commands get a red warning panel and always require explicit [y] confirmation.

One command contract: DOOM will never generate &&, ;, pipes, or loops. One request → one command.

Requirements

• macOS (Apple Silicon — M1/M2/M3/M4)
• Python 3.11+ (3.13 recommended)

License

MIT