Clily

English | Türkçe

clily turns natural language into a single shell command, checks it against local safety rules, and lets you run or cancel it.

NPM package: @emiran/clily CLI command: clily

Why Clily

• Turn plain English into shell commands
• Keep a local safety layer between the model and your terminal
• Review the generated command before execution
• Reuse local context such as recent shell history and the last command result
• Work across Windows, macOS, and Linux

Features

• Gemini, Groq, OpenAI, and OpenRouter provider support
• Interactive setup wizard
• Safety modes: safe, balanced, auto
• Local allowlist, warnlist, and denylist
• Secure local API key storage
• Config management from the CLI
• Config health check with clily config doctor
• Terminal UI with readable previews and confirmations

Platform Support

• Windows: PowerShell and CMD
• macOS: bash and zsh
• Linux: bash and zsh

Clily detects your current shell and asks the model for shell-appropriate commands.

Install

Requirements: Node.js >=20.11.0

npm install -g @emiran/clily

Clily may occasionally check whether a newer npm version is available. Disable this with CLILY_DISABLE_UPDATE_CHECK=1.

Quick Start

Run setup first:

clily --setup

You can also run setup again later with:

clily setup

Then try a few prompts:

clily "show git status"
clily "check whether Ruby is installed"
clily "list running Docker containers"

If local rules allow it, you can also run directly:

clily "show Node.js version" --run

--run tells Clily to execute the generated command when your safety mode allows it.

How It Works

1. You write a natural-language request.
2. Clily asks the selected model for one shell command.
3. The command is checked against local safety rules.
4. Clily shows a preview with command, risk, and reason.
5. Clily asks for confirmation or runs it directly, depending on your mode.

Safety Model

Clily uses three local rule groups:

• allowlist: trusted commands
• warnlist: commands that should require confirmation
• denylist: commands that should be blocked

Safety modes:

• safe: always ask before running
• balanced: allow trusted commands to auto-run, confirm others
• auto: run commands without confirmation unless a local rule blocks them

Manage rules from the CLI:

clily safety allow list
clily safety allow add "git status"
clily safety warn add "docker rm *"
clily safety deny add "rm -rf *"

Configuration

Useful commands:

clily config show
clily config path
clily config doctor
clily config set mode auto
clily config set provider.name groq
clily config set provider.model openai/gpt-oss-20b
clily config set provider.apiKey YOUR_KEY

clily config doctor checks for common setup, config, and API key problems.

Privacy And Secret Storage

• secret-like values can be masked before context is sent to a provider
• shell history can be enabled, limited, or disabled
• the last command result can be reused as local context

Providers

Gemini

• Model selection is available during setup

Groq

• Works best today with openai/gpt-oss-20b and openai/gpt-oss-120b

OpenAI

• Defaults to gpt-4o-mini

OpenRouter

• Model selection is available during setup

Documentation

• Turkish README
• Contributing Guide

Development

For local development:

npm install
npm run check
npm run build
npm run test
npm run dev -- --setup

Package preflight:

npm pack

Contributing

Issues and pull requests are welcome.

• Bug reports and feature requests: use GitHub Issues
• Code contributions: open a pull request
• Before opening a PR, run:

npm run check
npm run build
npm run test

More details: CONTRIBUTING.md

Support

If something looks wrong or confusing:

• open an issue for bugs or unexpected behavior
• open an issue for docs or setup problems
• include your platform, shell, provider, and the command you ran when possible

License

MIT