Skip to content

c0ffee0wl/llm-windows-setup

BranchesTags

Repository files navigation

LLM Tools Installation Script for Windows

GitHub Repository: https://github.com/c0ffee0wl/llm-windows-setup

Automated installation script for Simon Willison's llm CLI tool and related AI/LLM command-line utilities for Windows environments.

Based on the llm-linux-setup project.

📖 Looking for the Linux version? This is the Windows adaptation. For the original Linux version with additional features like AIChat (RAG), sandboxed shell execution, and more extensive documentation, see: llm-linux-setup - Linux/Debian/Ubuntu/Kali version

Features

  • ✅ One-command installation - Run once to install everything
  • ✅ Self-updating - Re-run to update all tools automatically
  • ✅ Multi-PowerShell support - Works with both PowerShell 5.1 and PowerShell 7+
  • ✅ Azure OpenAI integration - Configured for Azure Foundry
  • ✅ AI command completion - Press Ctrl+N for intelligent command suggestions
  • ✅ Automatic session recording - Terminal history captured for AI context
  • ✅ AI-powered context retrieval - Query your command history with context or llm --tool context
  • ✅ Smart admin handling - Only requires admin for Chocolatey installation

What Gets Installed

Core Tools

  • llm - LLM CLI tool (fork with markdown markup enhancements, originally by Simon Willison - Documentation)
  • Claude Code - Anthropic's agentic coding CLI
  • OpenCode - AI coding agent for terminal
  • uv - Modern Python package installer
  • Python 3 - Via Chocolatey
  • Node.js 22.x - Via Chocolatey (latest in 22 branch)
  • Git - Via Chocolatey (if not already installed)
  • jq - JSON processor

LLM Plugins

  • llm-cmd - Command execution and management
  • llm-cmd-comp - AI-powered command completion (powers Ctrl+N)
  • llm-tools-sqlite - SQLite database tool
  • llm-tools-context - Terminal history integration (exposes context tool to AI)
  • llm-fragments-site-text - Web page content extraction
  • llm-fragments-pdf - PDF content extraction
  • llm-fragments-github - GitHub repository integration
  • llm-fragments-dir - Load all text files from a local directory recursively
  • llm-jq - JSON processing tool
  • llm-templates-fabric - Fabric prompt templates
  • llm-gemini - Google Gemini models integration
  • llm-vertex - Google Vertex AI integration (enterprise API)
  • llm-openrouter - OpenRouter API integration
  • llm-anthropic - Anthropic Claude models integration

LLM Templates

  • assistant.yaml - Custom assistant template with security/IT expertise configuration
  • code.yaml - Code-only generation template (outputs clean, executable code without markdown)

Additional Tools

  • gitingest - Convert Git repositories to LLM-friendly text
  • files-to-prompt - File content formatter for LLM prompts
  • context - PowerShell history extraction for AI context retrieval

PowerShell Integration

  • AI-powered command completion (Ctrl+N)
  • Custom llm wrapper with default assistant template
  • Automatic PowerShell transcript logging for AI context
  • Clipboard aliases (pbcopy, pbpaste) for macOS compatibility
  • PATH configuration for all installed tools

System Requirements

  • OS: Windows 10, Windows 11, or Windows Server 2016+
  • PowerShell: 5.1 or higher (PowerShell 7+ supported)
  • Internet: Required for installation and API access
  • Disk Space: ~1GB for all tools and dependencies
  • Admin Rights: Only required for initial Chocolatey installation

Installation

Quick Start

  1. Clone the repository:

    git clone https://github.com/c0ffee0wl/llm-windows-setup.git
cd llm-windows-setup

  2. Run the installation script:

    • If Chocolatey is NOT installed (first-time installation):

      # Run as Administrator (required for Chocolatey installation)
.\Install-LlmTools.ps1

    • If Chocolatey IS already installed:

      # Can run as regular user
.\Install-LlmTools.ps1

  3. Follow the prompts:

    • You'll be asked if you want to configure Azure OpenAI (optional)
    • If yes, provide your Azure API key and resource URL

Execution Policy

If you get an execution policy error, run:

Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser

Updating

Simply re-run the installation script:

cd llm-windows-setup
.\Install-LlmTools.ps1

The script will automatically:

  1. Check for script updates from the git repository
  2. Pull latest changes and re-execute if updates are found
  3. Update llm and all plugins
  4. Update custom templates (assistant.yaml)
  5. Update gitingest, files-to-prompt, Claude Code, and OpenCode
  6. Refresh PowerShell integration files

Note: Updates do not require Administrator privileges (unless you need to update Chocolatey packages). The git pull happens automatically during Phase 0.

Usage

Basic LLM Usage

# Ask a question (uses assistant template by default)
llm "What is the capital of France?"

# Start an interactive chat session
llm chat "Let's discuss PowerShell"

# Use a specific model
llm -m azure/gpt-5 "Explain quantum computing"

# List available models
llm models list

# View installed plugins
llm plugins

AI Command Completion

Type a partial command or describe what you want in natural language, then press Ctrl+N:

# Type: list all json files recursively
# Press Ctrl+N
# Result: Get-ChildItem -Recurse -Filter *.json

The AI will suggest and insert the command automatically.

Azure OpenAI Models

The following models are configured (if you set up Azure OpenAI):

  • azure/gpt-4.1-mini - GPT-4.1 Mini (default, recommended for most tasks)
  • azure/gpt-4.1 - GPT-4.1 (recommended for complex tasks)
  • azure/gpt-4.1-nano - GPT-4.1 Nano (lightweight)
  • azure/gpt-5 - GPT-5 (most capable)
  • azure/gpt-5-mini - GPT-5 Mini (advanced reasoning)
  • azure/gpt-5-nano - GPT-5 Nano (advanced lightweight)
  • azure/o4-mini - O4 Mini

To switch models for more complex tasks:

llm models default azure/gpt-4.1

Clipboard Aliases (macOS Compatibility)

# Copy to clipboard (like macOS pbcopy)
"Hello World" | pbcopy

# Paste from clipboard (like macOS pbpaste)
pbpaste

Additional Tools

# Convert Git repositories to LLM-friendly text
gitingest https://github.com/user/repo
gitingest C:\path\to\local\repo

# Convert files to LLM-friendly format
files-to-prompt src\*.py

# Use OpenCode
opencode

# Use Claude Code
claude

Context System (PowerShell History for AI)

PowerShell sessions are automatically logged via transcript recording. The AI can retrieve your terminal history for better context:

# Show last command
context

# Show last 5 commands
context 5

# Show entire session history
context all

# Check transcript file location
$env:TRANSCRIPT_LOG_FILE

# Check transcript directory
$env:TRANSCRIPT_LOG_DIR

How it works:

  • Each PowerShell session automatically starts transcript logging
  • Transcripts are stored in $env:TRANSCRIPT_LOG_DIR (configurable during installation)
  • The context command parses transcripts and extracts command history
  • The llm-tools-context plugin exposes this to AI models for contextual assistance
  • AI can call context(N) to retrieve last N commands when helping with your tasks

Storage options:

  • Temporary (default): %TEMP%\PowerShell_Transcripts - cleared on logout/reboot
  • Permanent: %USERPROFILE%\PowerShell_Transcripts - survives reboots

Configuration

Configuration Files

  • %APPDATA%\io.datasette.llm\ - LLM configuration directory
    • extra-openai-models.yaml - Azure OpenAI model definitions
    • templates\assistant.yaml - Custom assistant template (auto-installed)
    • API keys stored securely via llm's key management

PowerShell Integration Files

Located in the integration\ subdirectory:

  • integration\llm-integration.ps1 - Unified integration for PS5 & PS7

This file is automatically sourced from your PowerShell profile.

PowerShell Profile Locations

The installation script adds integration to:

  • PowerShell 5: %USERPROFILE%\Documents\WindowsPowerShell\Microsoft.PowerShell_profile.ps1
  • PowerShell 7: %USERPROFILE%\Documents\PowerShell\Microsoft.PowerShell_profile.ps1

Changing Default Model

llm models default azure/gpt-4.1

Managing API Keys

# Set Azure key
llm keys set azure

# View key storage path
llm keys path

# List all configured keys
llm keys

Troubleshooting

Command completion not working

  1. Restart your PowerShell session or reload your profile:

    . $PROFILE

  2. Verify llm is in PATH:

    Get-Command llm

  3. Test llm command completion manually:

    llm cmdcomp "list files"

  4. Check if PSReadLine is loaded:

    Get-Module PSReadLine

Azure API errors

  1. Verify API key is set:

    llm keys get azure

  2. Check model configuration:

    Get-Content $env:APPDATA\io.datasette.llm\extra-openai-models.yaml

  3. Update the API base URL in the YAML file if needed

"llm" command not found

  1. Check if the installation completed successfully
  2. Verify PATH includes: %USERPROFILE%\.local\bin
  3. Restart PowerShell
  4. Re-run the installation script

Chocolatey installation fails

  1. Ensure you're running PowerShell as Administrator
  2. Check your internet connection
  3. Verify your execution policy allows scripts: 
    Get-ExecutionPolicy
  4. Install Chocolatey manually: https://chocolatey.org/install

npm permissions errors

The script configures npm to use user-level global installs. If you still get permission errors:

npm config set prefix "$env:USERPROFILE\.npm-global"

Then add %USERPROFILE%\.npm-global to your PATH.

Supported PowerShell Versions

  • PowerShell 5.1 (Windows PowerShell - included with Windows 10/11)
  • PowerShell 7.x (PowerShell Core - cross-platform)

Documentation

Differences from Linux Version

This Windows version differs from the Linux version in the following ways:

📚 Note: The Linux version contains more extensive documentation, detailed usage examples, and additional features. Refer to its README for comprehensive guides on fragments, templates, RAG, and advanced workflows.

Platform & Package Management

Windows:

  • Uses Chocolatey for system packages (Python, Node.js, Git, jq)
  • Smart admin privilege model (admin only required for Chocolatey on first run)
  • User-scoped installations for everything else (pipx, uv, npm global)

Linux:

  • Uses apt/dpkg for system packages
  • rustup for Rust toolchain management (required for aichat/argc)
  • nvm for Node.js version management (if repository version < 20)
  • Root/sudo access required for system package installation

Shell Integration

Windows:

  • PowerShell 5.1 and 7+ support
  • Single integration file (llm-integration.ps1) works for both PS versions
  • Uses PSReadLine for Ctrl+N command completion
  • Clipboard aliases: pbcopy/pbpaste functions

Linux:

  • Bash and Zsh support
  • Three-file integration pattern:
    • llm-common.sh - Shared configuration
    • llm-integration.bash - Bash-specific widgets
    • llm-integration.zsh - Zsh-specific widgets
  • llm-zsh-plugin - Tab completion for llm commands (Zsh only)
  • Clipboard aliases via xsel (macOS compatibility layer)

Session Recording & Context System

Windows:

  • Uses native PowerShell Start-Transcript cmdlet
  • Creates .txt transcript files (UTF-16-LE or UTF-8 encoding)
  • Stores in %TEMP%\PowerShell_Transcripts (temporary) or %USERPROFILE%\PowerShell_Transcripts (permanent)
  • One transcript per PowerShell window/tab
  • Parses transcript files with context.py (indentation-based parsing)
  • Limitation: Native commands (ping, git, etc.) output not always captured by PowerShell transcription

Linux:

  • Uses asciinema for terminal recording (built from source for latest features)
  • Creates .cast files (asciinema JSON format)
  • Stores in /tmp/session_logs/asciinema (temporary) or ~/session_logs/asciinema (permanent)
  • Tmux/screen support: Each pane/window gets independent recording with unique session files
  • Parses .cast files with context script (regex-based prompt detection)
  • Full command output capture including native commands

LLM Assistant Template Differences

Windows (assistant.yaml):

  • Tools: context only

Linux (assistant.yaml):

  • Tools: context + sandboxed_shell (bubblewrap-based safe command execution)

Tools & Features Exclusive to Linux Version

The Linux version includes several additional tools not present in the Windows version:

AIChat (RAG Functionality)

  • AIChat - All-in-one LLM CLI with built-in RAG (Retrieval-Augmented Generation)
  • Built-in vector database for document querying
  • Query codebases, documents, and knowledge bases: llm rag mydocs
  • Supports multiple document sources (Git repos, URLs, PDFs, DOCX, directories)
  • Auto-configured with Azure OpenAI settings

Additional LLM Plugins

The Linux version includes several additional LLM plugins not available in the Windows version:

Fragment Plugins:

  • llm-fragments-youtube-transcript - YouTube video transcript extraction with metadata

Tool Plugins:

  • llm-tools-sandboxed-shell - Safe shell command execution (covered in Sandboxed Command Execution section below)
  • llm-tools-patch - File manipulation tools (read, write, edit, multi_edit, info) with approval workflow
  • llm-tools-llm-functions - Bridge for llm-functions framework (covered in Developer Tools section below)
  • llm-tools-quickjs - JavaScript execution environment for AI

Sandboxed Command Execution

  • llm-tools-sandboxed-shell - Execute shell commands safely using bubblewrap
  • Isolated environment (read-only root, no network access, Linux namespaces)
  • Built into the Linux assistant template by default
  • Requires bubblewrap (automatically installed)

Developer Tools & Frameworks

  • Rust/Cargo toolchain - Required for building aichat and argc
  • argc - Bash CLI framework and command runner (enables llm-functions integration)
  • yek - Fast Rust-based repository to LLM-friendly text converter (230x faster than gitingest)

Document Processing

  • poppler-utils - PDF text extraction (pdftotext)
  • pandoc - Document converter (DOCX support for RAG)

What's the Same

Both versions share these core features:

  • ✅ llm CLI tool - Simon Willison's LLM CLI
  • ✅ Self-updating installation scripts - git pull before execution
  • ✅ Azure OpenAI integration - Configured for Azure Foundry
  • ✅ AI command completion - Ctrl+N intelligent command suggestions via llm cmdcomp
  • ✅ Context system - Query terminal history with AI
  • ✅ Custom templates - assistant.yaml (security/IT expertise) and code.yaml (clean code output)
  • ✅ Core LLM plugins - gemini, openrouter, anthropic, vertex, jq, sqlite, cmd, cmd-comp, fabric templates
  • ✅ Shared fragment plugins - site-text, pdf, github, dir (Linux has 1 additional: youtube-transcript)
  • ✅ Shared tool plugins - sqlite, context (Linux has 3 additional: quickjs, patch, sandboxed-shell)
  • ✅ Additional tools - gitingest, files-to-prompt, Claude Code, OpenCode
  • ✅ Smart template application - Shell wrapper auto-applies assistant template

License

This installation script is provided as-is under the MIT License. Individual tools have their own licenses:

  • llm: Apache 2.0
  • See individual tool repositories for details

Contributing

To modify or extend this installation:

  1. Fork the repository
  2. Make your changes
  3. Test on Windows 10/11
  4. Submit a pull request

Credits

Support

For issues, questions, or suggestions:

Related Projects

About

No description, website, or topics provided.

Resources

Readme

License

GPL-3.0 license
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •  

Languages