Show HN: Vibe zsh, turn natural language into shell commands

Developer Skymoore released Vibe Zsh, an open-source tool that converts natural language descriptions into shell commands using AI. The plugin integrates with Oh-My-Zsh and supports multiple AI providers including OpenAI, Anthropic, and local models via Ollama, offering features like OS-aware generation and command previews.

Transform natural language into shell commands using AI. Works natively with OpenAI, Anthropic, Groq, OpenRouter, Ollama, LM Studio, and more. Type your intent in natural language list all docker containers Press Ctrl+G Get the command with explanations docker ps -a docker: Docker command-line tool ps: List containers -a: Show all containers not just running - 🧠 Natural language to commands - Just describe what you want - 🖥️ OS-aware generation - Commands that work on YOUR system macOS/Linux/Windows - ⚡ Lightning fast - Cached responses are 100-400x faster - 🎬 Streaming output - Typewriter effect with progress indicators - 🔌 Multi-provider - Native support for OpenAI, Anthropic, Groq, OpenRouter, Ollama, LM Studio, and more - 🛡️ Safe by default - Preview commands before execution - 📚 Learn while you work - Inline explanations for every command - 📜 Query history - Interactive menu to browse and re-run previous queries - 🎯 Single binary - One compiled binary, no runtime dependencies to install brew tap skymoore/tap brew install --cask vibe-zsh The Homebrew installation automatically: - Installs vibe to ~/.oh-my-zsh/custom/plugins/vibe - Adds vibe to your plugins list in ~/.zshrc - Creates a global vibe-zsh command for CLI usage After installation, reload your shell: source ~/.zshrc Note: Requires Oh-My-Zsh. If not installed, the installer will provide instructions. For oh-my-zsh integration: curl -fsSL https://raw.githubusercontent.com/skymoore/vibe-zsh/main/install.sh | bash Or using wget: wget -qO- https://raw.githubusercontent.com/skymoore/vibe-zsh/main/install.sh | bash This script downloads the latest release and installs it to ~/.oh-my-zsh/custom/plugins/vibe . You'll need to manually add vibe to your plugins list in ~/.zshrc . - Clone the repository: git clone https://github.com/skymoore/vibe-zsh.git ~/.oh-my-zsh/custom/plugins/vibe - Build the binary: cd ~/.oh-my-zsh/custom/plugins/vibe make build - Add to your .zshrc : plugins= ... vibe - Reload your shell: source ~/.zshrc Add these to your ~/.zshrc all optional . vibe-zsh uses gollm https://github.com/teilomillet/gollm to talk to each provider natively. There are two kinds of provider: Hosted providers openai , anthropic , groq , openrouter , deepseek , google-openai , mistral , cohere have a fixed endpoint built in. You only need to choose the provider, set VIBE API KEY , and pick a VIBE MODEL — you do not set VIBE API URL . Local providers ollama , lmstudio , vllm run on your machine. Set VIBE API URL to point at the local server; no API key is required. Select a provider with VIBE PROVIDER . If you don't set it, vibe infers the provider from VIBE API URL e.g. an openrouter.ai or :11434 URL , which is mainly there to keep older configs working. Setting VIBE PROVIDER explicitly is the recommended approach. The default VIBE MODEL is llama3:8b chosen for the default Ollama setup . Always set VIBE MODEL when you use a hosted provider, or requests will ask for a model that doesn't exist there. export VIBE PROVIDER="openai" export VIBE API KEY="sk-..." export VIBE MODEL="gpt-4o" No VIBE API URL needed — the openai provider always targets api.openai.com. export VIBE PROVIDER="anthropic" export VIBE API KEY="sk-ant-..." export VIBE MODEL="claude-3-5-sonnet-20241022" export VIBE PROVIDER="groq" export VIBE API KEY="gsk ..." export VIBE MODEL="llama-3.1-70b-versatile" export VIBE PROVIDER="openrouter" export VIBE API KEY="sk-or-..." export VIBE MODEL="anthropic/claude-3.5-sonnet" export VIBE PROVIDER="ollama" export VIBE API URL="http://localhost:11434/v1" export VIBE MODEL="llama3:8b" Ollama must be running; no API key required. export VIBE PROVIDER="lmstudio" export VIBE API URL="http://localhost:1234/v1" export VIBE MODEL="local-model" LM Studio must be running and reachable when vibe starts. Note on validation:gollm checks your configuration when vibe starts. Hosted providers validate the API key format up front for example, Anthropic keys must start with sk-ant- , and local providers must already be running and reachable. If a provider is misconfigured, vibe reports an actionable error on the first query rather than silently failing. Custom OpenAI-compatible gateways:the openai provider always points at api.openai.com , so it can't be redirected to a self-hosted gateway via VIBE API URL . To use an OpenAI-compatible endpoint that isn't OpenAI itself, set VIBE PROVIDER to lmstudio or vllm and point VIBE API URL at your gateway. - Type a natural language description in your terminal - Press Ctrl+G - Review the generated command with explanations - Press Enter to execute, or edit first History Menu: Access your query history in two ways: Keybinding : Press Ctrl+X then H Command : Type vh and press Enter Both methods open an interactive menu where you can: - Browse previous queries with arrow keys - Search with / filter mode - Press Enter to insert the generated command into your buffer - Press G to regenerate a new command from the original query - Press V to edit the original query in your buffer - Press A or Home to jump to the first entry - Press E or End to jump to the last entry - Press Esc or Q to cancel The selected command appears on your command line, ready to execute Quick Regenerate: Press Ctrl+X then G to instantly regenerate a new command from your most recent query without opening the menu. Note : Don't run vibe-zsh history or ./vibe history directly - use vh or the keybinding instead. You can also use vibe directly from the command line: Homebrew installation vibe-zsh "list all docker containers" Manual installation Oh-My-Zsh plugin ~/.oh-my-zsh/custom/plugins/vibe/vibe "list all docker containers" CLI Flags: vibe-zsh --help Show all available flags vibe-zsh --debug "query" Enable debug logging vibe-zsh --temperature 0.1 "query" Override temperature vibe-zsh --interactive "query" Confirm before execution History Commands: vh Interactive menu recommended vibe-zsh history list List history in plain text vibe-zsh history clear Clear all history Note: Use 'vh' or Ctrl+X H for interactive menu Don't use 'vibe-zsh history' directly it just outputs to stdout Query History: Open history menu with keybinding Ctrl+X H Or use the command vh List history in plain text vibe-zsh history list Clear all history vibe-zsh history clear File Operations: show me all hidden files including their sizes → ls -lah Docker: show logs of nginx container and follow them → docker logs -f nginx Git: show me commits from last week → git log --since="1 week ago" Find & Search: find all python files modified today → find . -name " .py" -mtime 0 Tab Completion: vibe <TAB <TAB Shows common query suggestions Hide Explanations: export VIBE SHOW EXPLANATION=false Interactive Mode confirm before inserting : export VIBE INTERACTIVE=true Disable Cache: export VIBE ENABLE CACHE=false Disable Auto-Updates: export VIBE AUTO UPDATE=false Customize History Keybinding: export VIBE HISTORY KEY="^R" Use Ctrl+R for history menu export VIBE REGENERATE KEY="^ r" Use Alt+R for quick regenerate Note: Avoid ^H Ctrl+H as it conflicts with Backspace Disable History: export VIBE ENABLE HISTORY=false | Variable | Default | Description | |---|---|---| API Configuration | || VIBE PROVIDER | inferred from VIBE API URL | LLM provider. Hosted: openai , anthropic , groq , openrouter , deepseek , google-openai , mistral , cohere . Local: ollama , lmstudio , vllm . Recommended to set explicitly. | VIBE API URL | http://localhost:11434/v1 | Endpoint URL for local providers only ollama , vllm . Hosted providers ignore this and use their fixed endpoints. | VIBE API KEY | "" | API key. Required for hosted providers; ignored by local providers. | VIBE MODEL | llama3:8b | Model to use. Set this for hosted providers — the default only suits Ollama. | VIBE TEMPERATURE | 0.2 | Generation temperature 0.0-2.0 | VIBE MAX TOKENS | 1000 | Max response tokens | VIBE TIMEOUT | 30s | Request timeout | Display Options | || VIBE SHOW EXPLANATION | true | Show command explanations | VIBE SHOW WARNINGS | true | Show warnings for dangerous commands | VIBE SHOW PROGRESS | true | Show progress spinner during generation | VIBE PROGRESS STYLE | dots | Spinner style: dots, line, circle, bounce, arrow, runes ᛜ ᛃ ᛋ | VIBE STREAM OUTPUT | true | Stream output with typewriter effect | VIBE STREAM DELAY | 20ms | Delay between streamed words | Behavior | || VIBE INTERACTIVE | false | Confirm before inserting command | VIBE USE STRUCTURED OUTPUT | true | Use JSON schema for structured responses | VIBE ENABLE CACHE | true | Enable response caching | VIBE CACHE TTL | 24h | Cache lifetime | Parsing & Retry | || VIBE MAX RETRIES | 3 | Max retry attempts for failed parsing | VIBE ENABLE JSON EXTRACTION | true | Extract JSON from corrupted responses | VIBE STRICT VALIDATION | true | Validate response structure | VIBE SHOW RETRY STATUS | true | Show retry progress during generation | History | || VIBE ENABLE HISTORY | true | Enable query history tracking | VIBE HISTORY SIZE | 100 | Maximum number of history entries | VIBE HISTORY KEY | ^Xh Ctrl+X H | Keybinding for history menu | VIBE REGENERATE KEY | ^Xg Ctrl+X G | Keybinding to regenerate last command | Updates & Debugging | || VIBE AUTO UPDATE | true | Enable auto-update checks | VIBE UPDATE CHECK INTERVAL | 7d | How often to check for updates | VIBE DEBUG LOGS | false | Enable debug logging for troubleshooting | Capture - You type natural language and press Ctrl+G Context - vibe detects your OS macOS/Linux/Windows and shell zsh/bash/etc. Generate - vibe sends your query with system context to the configured LLM Parse - Response is structured as command + explanations Cache - Response is cached for 24 hours configurable Stream - Command is streamed to your terminal with typewriter effect Insert - Command appears in your buffer for review Execute - You press Enter to run or edit first vibe automatically detects your operating system and shell, ensuring generated commands work on your system: macOS : Uses BSD utilities e.g., find without -printf , sed -i '' Linux : Uses GNU utilities e.g., find -printf , sed -i Shell-specific : Generates syntax appropriate for zsh, bash, etc. This means you get commands that actually work on your system, not generic Linux commands that fail on macOS First query: ~500ms-2s depends on LLM Cached query: ~5-10ms 100-400x faster Binary size: ~8MB Memory usage: <10MB - ✅ Commands are never executed automatically - ✅ Full preview before execution - ✅ Edit commands before running - ✅ Warnings for dangerous commands - ✅ Optional interactive confirmation mode - ✅ Local-first with Ollama your data stays private vibe automatically checks for updates once a week in the background zero impact on performance . When an update is available, you'll see a notification: ⚠️ vibe v1.2.4 available current: v1.2.3 Run: vibe --update Manual Commands: vibe --version Show current version vibe --update Download and install latest version Configuration: export VIBE AUTO UPDATE=false Disable auto-update checks export VIBE UPDATE CHECK INTERVAL=14d Check every 2 weeks instead Security: - All downloads are verified using SHA256 checksums - Checksums are published with each release - Original binary is backed up before replacement - Safe atomic file replacement The update check runs in a background process after each command, so it never slows down your workflow. Updates are never installed automatically - you always control when to update. Command not working after install: - Reload your shell: source ~/.zshrc - Verify plugin is in list: echo $plugins - Check binary exists: ls ~/.oh-my-zsh/custom/plugins/vibe/vibe Slow responses: - Enable caching: export VIBE ENABLE CACHE=true - Use a faster model or local LLM - Check your network connection Bad command suggestions: - Try a different/better model - Make your query more specific - Check model supports structured output Corrupted or garbage output: - vibe now automatically handles corrupted LLM responses with multi-layer parsing - Enable debug logs to see what's happening: export VIBE DEBUG LOGS=true - Check logs for raw responses and parsing attempts - Try increasing retries: export VIBE MAX RETRIES=5 Parsing errors: - Enable JSON extraction: export VIBE ENABLE JSON EXTRACTION=true default - Disable strict validation temporarily: export VIBE STRICT VALIDATION=false - Check debug logs to see which parsing layer succeeded/failed - Some models produce cleaner JSON with lower temperature: export VIBE TEMPERATURE=0.3 Ctrl+G does nothing: - Ensure plugin is loaded: which vibe should show a function - Check no other plugin uses Ctrl+G - Try rebinding: bindkey '^G' vibe - Zsh with Oh-My-Zsh - An LLM provider — hosted OpenAI, Anthropic, Groq, OpenRouter, etc. or local Ollama, LM Studio, vLLM - macOS or Linux brew uninstall --cask vibe-zsh brew untap skymoore/tap Optional: remove the tap Then remove vibe from your plugins list in ~/.zshrc and reload: source ~/.zshrc curl -fsSL https://raw.githubusercontent.com/skymoore/vibe-zsh/main/uninstall.sh | bash Or using wget: wget -qO- https://raw.githubusercontent.com/skymoore/vibe-zsh/main/uninstall.sh | bash Then remove vibe from your plugins list in ~/.zshrc and reload: source ~/.zshrc git clone https://github.com/skymoore/vibe-zsh.git cd vibe-zsh make build Build for current platform make build-all Build for all platforms make test Run tests make install Install to oh-my-zsh Contributions welcome This project uses: - Go 1.26+ - Standard zsh scripting gollm https://github.com/teilomillet/gollm for native multi-provider LLM access GPLv3 License - see LICENSE /skymoore/vibe-zsh/blob/main/LICENSE Inspired by LoganPederson/vibe https://github.com/LoganPederson/vibe Built for terminal productivity enthusiasts 🚀