hey-cli

Your terminal buddy that turns plain English into shell scripts — and runs them for you.

hey is a terminal-native AI assistant that translates plain English into executable shell commands using a locally-hosted LLM via Ollama. Your data never leaves your machine.

Features

• 100% Local & Private — All reasoning happens on your machine via Ollama. No API keys, no cloud, no telemetry.
• Cross-Platform Intelligence — Detects your OS (macOS/Linux/Windows) and generates the correct flags. Won't suggest xargs -d on BSD or apt on Arch.
• Agentic Context Gathering — Ask "is Docker running?" and hey silently runs diagnostics, reads the output, and answers in plain English.
• Security Governance — Dangerous commands (rm -rf, mkfs, DROP TABLE) are intercepted and require explicit confirmation before execution.
• Pipe-Friendly — Pipe error logs directly: npm run build 2>&1 | hey what is causing this error?
• Conversational Memory — Remembers your recent interactions for follow-up questions.

---

Installation

hey requires Ollama running locally. Install it first, then choose your platform:

macOS (Homebrew)

brew tap sinsniwal/hey-cli
brew install hey-cli

Tip

Apple Silicon (M1/M2/M3) Note: If you see an error about Rosetta 2 while installing via Homebrew, ensure your terminal is running natively (not under Rosetta emulation). You can force a native installation by running: arch -arm64 brew install hey-cli

macOS & Linux (curl)

curl -sL https://raw.githubusercontent.com/sinsniwal/hey-cli/main/install.sh | bash

Windows (Scoop)

scoop install https://raw.githubusercontent.com/sinsniwal/hey-cli/main/scoop/hey-cli.json

Windows (PowerShell Installer)

Invoke-WebRequest -Uri "https://raw.githubusercontent.com/sinsniwal/hey-cli/main/install.ps1" -OutFile "$env:TEMP\hey_install.ps1"; & "$env:TEMP\hey_install.ps1"

Windows (Standalone Binary)

Download hey.exe from the latest release. No Python required.

uv (Recommended)

uv tool install hey-cli-python

Note: After installation, authenticate with ollama login and pull the default model: ollama pull gpt-oss:20b-cloud

Uninstallation

macOS & Linux:

curl -sL https://raw.githubusercontent.com/sinsniwal/hey-cli/main/uninstall.sh | bash

Windows (PowerShell):

Invoke-WebRequest -Uri "https://raw.githubusercontent.com/sinsniwal/hey-cli/main/uninstall.ps1" -OutFile "$env:TEMP\hey_uninstall.ps1"; & "$env:TEMP\hey_uninstall.ps1"

---

Usage

hey <your objective in plain English>

Examples

Command	What happens
hey list all running docker containers	Generates and runs docker ps
hey is port 8080 in use?	Silently runs lsof -i :8080, reads output, answers in English
hey forcefully delete all .pyc files	Generates find . -name "*.pyc" -delete, pauses for confirmation
hey compress this folder into a tar.gz	Generates the correct tar command for your OS
npm run build 2>&1 | hey what broke?	Reads piped stderr and explains the error
hey --clear	Wipes conversational memory

---

Shell Integration (Recommended)

By default, CLI tools cannot change your terminal's directory because they run in a subshell. To enable hey to change your directory (e.g., hey go to desktop), add the following to your shell configuration (~/.zshrc or ~/.bashrc):

eval "$(hey --shell-init)"

For Windows (PowerShell):

hey --shell-init | Out-String | iex

---

Execution Levels

Level	Flag	Behavior
0	--level 0	Dry-run — shows the command but never executes
1	(default)	Supervised — safe commands auto-run, risky ones ask for confirmation
2	--level 2	Unrestricted — executes everything without confirmation
3	--level 3	Troubleshooter — iteratively debugs until the objective is resolved

---

---

Authentication & Custom Endpoints

hey works locally by default, but it also supports authenticated Ollama instances and custom hosts:

• Standard Login: Most users should run ollama login once to authenticate their terminal with the local or cloud instance.
• Auth Key: If you are in a CI/CD or server environment, you can set the OLLAMA_API_KEY environment variable.
• Custom Host: If Ollama is running on a different port or machine, set OLLAMA_HOST (e.g., export OLLAMA_HOST="http://192.168.1.10:11434").
• Custom Model: You can provide a custom model via --model:

  hey "summarize this file" --model llama3

---

Security

Safety is enforced at runtime via a local governance engine (~/.hey-rules.json):

• Blocked — rm -rf /, mkfs, :(){ :|:& };: are permanently rejected.
• Explicit Confirm — High-risk operations (rm, truncate, DROP) require typing a keyword to authorize.
• Y/N Confirm — Moderate-risk operations require a quick y.
• Auto-Run — Safe diagnostics (ls, cat, grep, git status) execute immediately.

Initialize or customize your rules:

hey --init

---

OS Skills

hey ships with built-in knowledge for macOS, Ubuntu/Debian, Arch Linux, Fedora/RHEL, Windows PowerShell, FreeBSD, and ChromeOS.

Want to improve it for your OS? Add a Markdown file to hey_cli/skills/ with plain-English rules (e.g., "On Alpine, use apk add instead of apt install"). The engine loads them dynamically at runtime.

Pull requests for new OS skills are welcome!

---

Architecture

hey "your question"
    │
    ▼
┌──────────────┐     ┌──────────────────┐
│  CLI Parser  │────▶│ Governance Check │
└──────────────┘     └────────┬─────────┘
                              │
                    ┌─────────▼──────────┐
                    │  Ollama (local LLM)│
                    │  localhost:11434   │
                    └─────────┬──────────┘
                              │
                    ┌─────────▼──────────┐
                    │  Command Runner    │
                    │  (execute / confirm)│
                    └────────────────────┘

• Zero external API calls — communicates with Ollama via localhost:11434 using Python's built-in urllib.
• Zero compiled dependencies — the core logic is pure Python (only rich is used for formatting).
• Pure Python 3.9–3.14 — works with uv, pipx, or standard pip. No build tools required.
• Instant Speed — Installation with uv or brew binaries takes < 2 seconds.

---

Contributing

1. Fork the repository
2. Create a feature branch: git checkout -b feature/my-feature
3. Commit changes: git commit -m "feat: add my feature"
4. Push and open a Pull Request

See RELEASING.md for maintainer release instructions.

---

License

MIT — Mohit Singh Sinsniwal