# zsh-opencode-ai 🤖 - AI-Powered Shell Command Generation

Convert natural language comments to shell commands with a single Tab press
Lightning-fast AI completion powered by OpenCode and GitHub Copilot

Quick Start • Features • Configuration • Usage • Development

Stop googling shell commands. Just describe what you want in plain English, press Tab, and get the exact command you need.

zsh-opencode-ai gives you:

• ⚡ Instant AI completion: Type # list all files → press Tab → get ls -la
• 🎯 Context-aware suggestions: Understands your intent and generates precise commands
• 🔄 Seamless integration: Works alongside existing completions (fzf-tab, zsh-autosuggestions)
• 🎨 Clean UX: Animated spinner feedback, no prompt clutter
• 🔧 Configurable: Switch AI models, adjust behavior via environment variables

• Installation
• Quick Start
• Features
• Configuration
• Usage
• Development
• Troubleshooting
• Contributing
• License

• Zsh (shell)
• OpenCode installed and configured (installation guide)
• Authenticated AI provider (GitHub Copilot recommended, OpenAI supported)
• Oh My Zsh (optional, recommended)

# Clone this repository to Oh My Zsh custom plugins directory
git clone https://github.com/USER/zsh-opencode-plugin ${ZSH_CUSTOM:-~/.oh-my-zsh/custom}/plugins/zsh-opencode-ai

# Add to plugins in ~/.zshrc
plugins=(... zsh-opencode-ai)

# Reload shell
source ~/.zshrc

# Clone the repository
git clone https://github.com/USER/zsh-opencode-plugin ~/zsh-opencode-ai

# Add to ~/.zshrc
source ~/zsh-opencode-ai/zsh-opencode-ai.plugin.zsh

# Reload shell
source ~/.zshrc

# Type a comment and press Tab
# remove all log files
<Tab>

# Should see spinner, then command appears

Ensure you have an AI provider authenticated:

# Check authentication status
opencode auth list

# If GitHub Copilot not authenticated
opencode config
# Select GitHub Copilot and follow prompts

# Type a natural language comment (starts with #)
$ # list all files

# Press Tab key
⠋ AI generating command...

# Command appears clean
$ ls -la

# Press Enter to execute

# Find files
# find all PDF files modified in last week
<Tab>
→ find . -type f -name "*.pdf" -mtime -7

# System operations
# show disk usage sorted by size
<Tab>
→ du -sh * | sort -h

# Git operations
# show git commits from last month
<Tab>
→ git log --since="1 month ago" --oneline

# Process management
# kill all node processes
<Tab>
→ pkill -f node

Press ? for examples, Ctrl+C to cancel generation.

• Natural language → command conversion: Describe intent, get executable command
• Tab key trigger: Seamless integration with existing Tab completion
• Animated spinner: Visual feedback during AI generation (~8s response time)
• Clean prompt handling: Original comment removed, only command visible
• Multiple AI models: GitHub Copilot, OpenAI GPT-4, Anthropic Claude supported

• fzf-tab compatibility: Falls back to standard completion for non-# lines
• Multi-format response parsing: Handles markdown blocks, inline code, raw text
• Prompt echo removal: Strips duplicated prompt from AI responses
• Configurable model selection: Switch models via environment variable
• Spinner customization: Adjust animation speed and style

• OpenCode integration: Uses local OpenCode server for AI access
• GitHub Copilot support: Authenticated provider, ~8s response time
• OpenAI support: Requires re-authentication, ~2-4s response time
• Anthropic support: Alternative AI provider option

Configuration is merged from multiple sources (in order of precedence):

1. Defaults (hardcoded in plugin)
2. Environment variables (highest priority)

# Set AI model (default: github-copilot/gpt-4o)
export OPENCODE_MODEL="github-copilot/gpt-4o"

# Alternative models (requires authentication)
export OPENCODE_MODEL="openai/gpt-4"
export OPENCODE_MODEL="anthropic/claude-3-5-sonnet-20241022"

# Adjust spinner animation speed (default: 0.1 seconds)
export OPENCODE_SPINNER_DELAY=0.1

# Set in ~/.zshrc for persistence
echo 'export OPENCODE_MODEL="github-copilot/gpt-4o"' >> ~/.zshrc

No config file needed. All settings via environment variables.

# File operations
# copy all images to backup folder
<Tab>
→ cp *.{jpg,png,gif} backup/

# compress all logs into archive
<Tab>
→ tar -czf logs.tar.gz *.log

# find files larger than 100MB
<Tab>
→ find . -type f -size +100M

# System monitoring
# show top 10 memory consuming processes
<Tab>
→ ps aux --sort=-%mem | head -11

# check listening ports
<Tab>
→ lsof -iTCP -sTCP:LISTEN -n -P

# Git workflows
# undo last commit but keep changes
<Tab>
→ git reset --soft HEAD~1

# show files changed in last commit
<Tab>
→ git diff --name-only HEAD~1

# Network operations
# download file from URL
<Tab>
→ curl -O <URL>

# test network connectivity to domain
<Tab>
→ ping -c 4 google.com

1. Type comment: Start line with # followed by natural language
2. Press Tab: Triggers AI generation
3. See spinner: Visual feedback during processing (~8s)
4. Review command: AI-generated command appears in buffer
5. Execute or edit: Press Enter to run, or edit before running

• Be specific: More detail = better commands

  • ❌ # list files
  • ✅ # list all files including hidden sorted by modification time

• Include context: Mention file types, paths, constraints

  • ❌ # find files
  • ✅ # find all .log files in /var/log modified today

• Use natural language: Write like you're asking a human

  • ✅ # show me the 10 largest files in current directory
  • ✅ # kill all processes using port 8080

• Zsh 5.0+ (required for ZLE widgets)
• Python 3.x (for response parsing)
• OpenCode (installed and running)

zsh-opencode-plugin/
├── zsh-opencode-ai.plugin.zsh    # Main plugin file (ZLE widget)
├── README.md                      # This file
├── AGENTS.md                      # AI agent development guidelines
├── LICENSE                        # MIT License
├── .gitignore                    # Git ignore rules
├── banner.svg                     # Project banner (generated)
├── banner.png                     # Project banner (generated)
└── docs/                          # Local documentation (gitignored)
    └── INSTALL.md                 # Detailed installation guide

1. ZLE Widget Binding: ^I (Tab) bound to _opencode_ai_suggest_or_fzf
2. Trigger Detection: Checks if BUFFER starts with #
3. Prompt Extraction: Strips # and whitespace from buffer
4. Line Clearing: zle kill-whole-line removes original prompt
5. Spinner Start: Background process with Braille animation (⠋⠙⠹⠸⠼⠴⠦⠧⠇⠏)
6. OpenCode Call: opencode run --model <MODEL> "<SYSTEM_PROMPT>"
7. Response Parsing: Python script handles markdown/inline/raw formats
8. Spinner Kill: Clean shutdown of background process
9. Command Insertion: Only generated command placed in BUFFER
10. Fallback: Non-# lines fall through to fzf-tab or default completion

• OpenCode: AI model orchestration
• GitHub Copilot: Default AI provider
• Python: Response parsing
• Zsh: Shell and ZLE framework

• Formatter: shfmt -w zsh-opencode-ai.plugin.zsh before commit
• Linter: shellcheck zsh-opencode-ai.plugin.zsh must pass
• Keep Zsh widget logic minimal and readable
• Test with multiple AI models (GitHub Copilot, OpenAI, Anthropic)
• Ensure spinner cleanup on all exit paths
• Handle edge cases (empty prompts, network failures, timeout)

Cause: AI model not authenticated or unavailable

Fix:

# Check authentication
opencode auth list

# Re-authenticate if needed
opencode config

Cause: ZLE widget not properly bound or conflicting plugin

Fix:

# Check binding
bindkey | grep '\^I'

# Should show: "^I" _opencode_ai_suggest_or_fzf

# Disable conflicting plugins temporarily
# Edit ~/.zshrc, comment out other Tab completion plugins

Cause: /dev/tty write issue or terminal doesn't support \r\033[K

Fix:

# Test terminal capabilities
printf "\r\033[K" > /dev/tty

# Try different terminal emulator (iTerm2, Kitty, Alacritty)

Cause: Using OpenAI without authentication or network latency

Fix:

# Switch to GitHub Copilot (faster)
export OPENCODE_MODEL="github-copilot/gpt-4o"

# Or re-authenticate OpenAI for faster API access
opencode config

Cause: Ambiguous prompt or model misunderstanding

Fix:

• Be more specific in your natural language description
• Include file types, paths, constraints explicitly
• Test different phrasings
• Check if model supports your use case

Cause: OpenAI OAuth token expired

Fix:

# Re-authenticate
opencode config

# Select OpenAI provider
# Follow OAuth flow in browser

# Verify authentication
opencode auth list

Contributions are welcome! Please follow these guidelines:

1. Fork the repository
2. Create a feature branch (git checkout -b feat/amazing-feature)
3. Make your changes
4. Test thoroughly (multiple shells, AI models, edge cases)
5. Lint code (shellcheck zsh-opencode-ai.plugin.zsh)
6. Commit your changes (git commit -m 'feat(completion): add amazing feature')
7. Push to the branch (git push origin feat/amazing-feature)
8. Open a Pull Request

• All PRs require review before merging
• Code must pass shellcheck linting
• Must test with at least 2 AI models
• Update documentation as needed
• Follow conventional commit format

This project is licensed under the MIT License - see the LICENSE file for details.

Built using:

• OpenCode - AI model orchestration and session management
• GitHub Copilot - Default AI provider for command generation
• Zsh - Shell and ZLE (Zsh Line Editor) framework

Inspired by:

• zsh-autosuggestions - Fish-like autosuggestions
• fzf-tab - Fuzzy completion framework

Special thanks to early testers and contributors.

• Documentation: See docs/INSTALL.md for detailed installation
• Issues: GitHub Issues
• Discussions: GitHub Discussions
• OpenCode: OpenCode Documentation

Stop googling. Start describing. 🤖