git-agent

A .NET CLI tool that translates natural language instructions into git commands using AI providers (Claude, OpenAI, OpenRouter, Ollama).

Installation

Download Pre-built Binaries

Download the latest release for your platform from GitHub Releases:

Platform	Download
Windows (x64)	`git-agent-win-x64.exe`
Linux (x64)	`git-agent-linux-x64`
macOS (Intel)	`git-agent-osx-x64`
macOS (Apple Silicon)	`git-agent-osx-arm64`

Quick Install - Windows (PowerShell)

New-Item -ItemType Directory -Force -Path "$env:LOCALAPPDATA\git-agent" | Out-Null
Invoke-WebRequest -Uri "https://github.com/george-marcus/git-agent/releases/latest/download/git-agent-win-x64.exe" -OutFile "$env:LOCALAPPDATA\git-agent\git-agent.exe"

# Add to PATH (run once, then restart terminal)
$path = [Environment]::GetEnvironmentVariable("Path", "User")
if ($path -notlike "*git-agent*") {
    [Environment]::SetEnvironmentVariable("Path", "$path;$env:LOCALAPPDATA\git-agent", "User")
}

Quick Install - Linux

mkdir -p ~/.local/bin
curl -L https://github.com/george-marcus/git-agent/releases/latest/download/git-agent-linux-x64 -o ~/.local/bin/git-agent
chmod +x ~/.local/bin/git-agent

Quick Install - macOS

mkdir -p ~/.local/bin

# Apple Silicon (M1/M2/M3)
curl -L https://github.com/george-marcus/git-agent/releases/latest/download/git-agent-osx-arm64 -o ~/.local/bin/git-agent

# Intel Mac
curl -L https://github.com/george-marcus/git-agent/releases/latest/download/git-agent-osx-x64 -o ~/.local/bin/git-agent

chmod +x ~/.local/bin/git-agent

Make sure ~/.local/bin is in your PATH. Add to your shell profile if needed:

echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zshrc  # or ~/.bashrc

Build from Source

Windows (PowerShell)

git clone https://github.com/george-marcus/git-agent.git
cd git-agent
.\scripts\install.ps1

This will:

Build the project as a single executable
Install to %LOCALAPPDATA%\git-agent
Add to your user PATH

Linux / macOS

git clone https://github.com/george-marcus/git-agent.git
cd git-agent
chmod +x ./scripts/install.sh
./scripts/install.sh

This will:

Build for your platform (linux-x64, linux-arm64, osx-x64, osx-arm64)
Install to ~/.local/bin
Add to your PATH via shell rc file

Manual Build

dotnet publish -c Release -r win-x64 --self-contained -p:PublishSingleFile=true
dotnet publish -c Release -r linux-x64 --self-contained -p:PublishSingleFile=true
dotnet publish -c Release -r osx-arm64 --self-contained -p:PublishSingleFile=true

Install as .NET Global Tool

dotnet pack -c Release
dotnet tool install --global --add-source ./bin/Release GitAgent

Uninstall

.\scripts\uninstall.ps1

./scripts/uninstall.sh

Quick Start

git-agent config set claude.apiKey sk-ant-your-key-here
git-agent config use claude

git-agent run "commit all my changes with a descriptive message"

git-agent run "push to origin" --exec

git-agent run "merge feature branch into main" --exec --interactive

Commands

`run <instruction>`

Translate a natural language instruction into git commands.

git-agent run <instruction> [options]

Arguments:

<instruction> - Natural language instruction to translate

Options:

Option	Short	Description
`--exec`	`-x`	Execute the generated commands
`--interactive`	`-i`	Confirm each command before execution
`--provider <name>`	`-p`	Override the active provider for this run
`--no-cache`		Skip cache and force a fresh API call

Examples:

git-agent run "stage all modified files and commit with message 'fix: resolve login bug'"

git-agent run "create a new branch called feature/auth" -x

git-agent run "rebase onto main" -xi

git-agent run "show recent commits" -p openai

`config`

Manage git-agent configuration.

`config show`

Display current configuration.

git-agent config show

`config set <key> <value>`

Set a configuration value.

git-agent config set <key> <value>

Available keys:

Key	Description
`activeProvider`	Active provider (claude, openai, openrouter, ollama, stub)
`claude.apiKey`	Claude API key
`claude.model`	Claude model name (default: claude-sonnet-4-20250514)
`claude.baseUrl`	Claude API base URL
`openai.apiKey`	OpenAI API key
`openai.model`	OpenAI model name (default: gpt-4o)
`openai.baseUrl`	OpenAI API base URL
`openrouter.apiKey`	OpenRouter API key
`openrouter.model`	OpenRouter model name (default: openai/gpt-4o)
`openrouter.baseUrl`	OpenRouter API base URL
`openrouter.siteName`	App name for OpenRouter attribution (default: GitAgent)
`openrouter.siteUrl`	Site URL for OpenRouter attribution
`ollama.model`	Ollama model name (default: llama3.2)
`ollama.baseUrl`	Ollama API base URL (default: http://localhost:11434)

Examples:

git-agent config set claude.apiKey sk-ant-xxxxx
git-agent config set openai.model gpt-4-turbo
git-agent config set openrouter.apiKey sk-or-xxxxx
git-agent config set openrouter.model anthropic/claude-3-opus
git-agent config set ollama.baseUrl http://192.168.1.100:11434

`config get <key>`

Get a configuration value.

git-agent config get claude.model

`config use <provider>`

Set the active provider.

git-agent config use <provider>

Available providers: claude, openai, openrouter, ollama, stub

git-agent config use claude
git-agent config use openai
git-agent config use openrouter
git-agent config use ollama

`config path`

Show the configuration file path.

git-agent config path

`config reset`

Reset configuration to defaults.

git-agent config reset

`providers`

List available AI providers.

git-agent providers

Output:

Available providers:
  - claude (active)
  - openai
  - openrouter
  - ollama
  - stub

Use 'git-agent config use <provider>' to switch providers.

`conflicts`

Analyze and resolve merge conflicts with AI assistance.

git-agent conflicts [options] [file]

Options:

Option	Short	Description
`--suggest`	`-s`	Show AI-suggested resolutions
`--resolve`	`-r`	Interactively resolve conflicts

Examples:

# Analyze all conflicts
git-agent conflicts

# Get AI suggestions for resolution
git-agent conflicts -s

# Interactively resolve conflicts
git-agent conflicts -r

# Analyze specific file
git-agent conflicts src/app.ts

`completions`

Generate shell completion scripts.

git-agent completions <shell>

Supported shells: bash, zsh, powershell, fish

Installation:

# Bash
git-agent completions bash >> ~/.bashrc

# Zsh
git-agent completions zsh >> ~/.zshrc

# PowerShell
git-agent completions powershell >> $PROFILE

# Fish
git-agent completions fish > ~/.config/fish/completions/git-agent.fish

`cache`

Manage HTTP response cache.

`cache clear`

Clear all cached HTTP responses.

git-agent cache clear

`cache path`

Show cache directory path.

git-agent cache path

IDE Integration

VS Code Extension

A full-featured VS Code extension is included in the vscode-extension/ directory.

Installation

Build the extension:

cd vscode-extension
npm install
npm run compile
npx vsce package

Install in VS Code:
- Press Ctrl+Shift+P → "Extensions: Install from VSIX..."
- Select the generated .vsix file

Usage

Command	Keyboard Shortcut	Description
Git Agent: Run Instruction	`Ctrl+Shift+G Ctrl+Shift+R`	Enter natural language instruction
Git Agent: Run & Execute	`Ctrl+Shift+G Ctrl+Shift+E`	Run and execute immediately
Git Agent: Analyze Conflicts	-	Analyze merge conflicts
Git Agent: Set Provider	-	Switch AI provider

Settings

Configure in VS Code settings (Ctrl+,):

{
  "git-agent.provider": "claude",
  "git-agent.executablePath": "git-agent",
  "git-agent.showOutputPanel": true,
  "git-agent.confirmBeforeExecute": true
}

See vscode-extension/README.md for full documentation.

Visual Studio (Windows)

Git-agent can be integrated with Visual Studio using External Tools:

Setup External Tool

Open Visual Studio
Go to Tools → External Tools...
Click Add and configure:

Field	Value
Title	`Git Agent: Run`
Command	`cmd.exe`
Arguments	`/c git-agent run "$(PromptVariable)" & pause`
Initial directory	`$(SolutionDir)`

Add another for quick execution:

Field	Value
Title	`Git Agent: Run & Execute`
Command	`cmd.exe`
Arguments	`/c git-agent run "$(PromptVariable)" --exec & pause`
Initial directory	`$(SolutionDir)`

Assign Keyboard Shortcuts

Go to Tools → Options → Environment → Keyboard
Search for Tools.ExternalCommand1 (or the number of your git-agent tool)
Assign a shortcut like Ctrl+Shift+G, Ctrl+Shift+R

Using Terminal

Alternatively, use the integrated terminal:

Open View → Terminal (`Ctrl+``)
Run git-agent commands directly

Configuration

Configuration is stored in ~/.git-agent/config.json:

{
  "activeProvider": "claude",
  "providers": {
    "claude": {
      "apiKey": "sk-ant-...",
      "model": "claude-sonnet-4-5",
      "baseUrl": "https://api.anthropic.com"
    },
    "openai": {
      "apiKey": "sk-...",
      "model": "gpt-4o",
      "baseUrl": "https://api.openai.com"
    },
    "openrouter": {
      "apiKey": "sk-or-...",
      "model": "openai/gpt-4o",
      "baseUrl": "https://openrouter.ai",
      "siteName": "GitAgent",
      "siteUrl": ""
    },
    "ollama": {
      "model": "llama3.2",
      "baseUrl": "http://localhost:11434"
    }
  }
}

Using OpenRouter

OpenRouter is a unified API that provides access to 500+ AI models from multiple providers (OpenAI, Anthropic, Google, Meta, Mistral, and more) with a single API key.

Benefits

Access to many models: Use GPT-4, Claude, Gemini, Llama, Mistral, and hundreds more with one API key
Cost optimization: OpenRouter automatically routes to the cheapest provider for each model
Fallback support: Automatic failover if a provider is unavailable
OpenAI-compatible: Uses the standard OpenAI API format

Setup

Get an API key from openrouter.ai/keys

Configure git-agent:

git-agent config set openrouter.apiKey sk-or-xxxxx
git-agent config use openrouter

Choose a model (optional, default is openai/gpt-4o):

# Use Claude via OpenRouter
git-agent config set openrouter.model anthropic/claude-3-opus

# Use Llama 3.1 405B
git-agent config set openrouter.model meta-llama/llama-3.1-405b-instruct

Popular Models

Model	ID
GPT-4o	`openai/gpt-4o`
Claude 3 Opus	`anthropic/claude-3-opus`
Llama 3.1 405B	`meta-llama/llama-3.1-405b-instruct`

See the full model list at openrouter.ai/models.

Using Ollama for Local LLM

Ollama allows you to run LLMs locally without any API keys or cloud dependencies.

1. Install Ollama

Windows: Download and run the installer from ollama.ai/download

Linux:

curl -fsSL https://ollama.ai/install.sh | sh

macOS:

brew install ollama

2. Pull a Model

ollama pull llama3.2

Other recommended models for git commands:

llama3.2 (default, 3B parameters, fast)
llama3.1 (8B parameters, better quality)
codellama (optimized for code tasks)
mistral (7B parameters, good balance)

3. Start Ollama Server

Ollama runs as a background service. On most systems it starts automatically after installation. If not:

ollama serve

The server runs on http://localhost:11434 by default.

4. Configure git-agent

git-agent config use ollama
git-agent config set ollama.model llama3.2

Optional: If Ollama runs on a different host or port:

git-agent config set ollama.baseUrl http://192.168.1.100:11434

5. Use git-agent

git-agent run "show me the last 5 commits"
git-agent run "commit all changes with a good message" -x

Troubleshooting

"Connection refused" error:

Ensure Ollama is running: ollama serve
Check the URL: curl http://localhost:11434/api/tags

Slow responses:

Use a smaller model like llama3.2 (3B) instead of larger ones
Ensure you have enough RAM (8GB+ recommended for 7B models)

Poor quality responses:

Try a larger model: ollama pull llama3.1
Some models work better for code tasks: ollama pull codellama

Safety Features

Commands are validated against an allowlist and categorized by risk level:

Safe (green): status, add, commit, push, pull, branch, checkout, switch, merge, fetch, log, diff, stash, tag, remote, show, rebase, reset --soft
Destructive (red): Commands with --force, reset --hard, clean, push --delete, branch -D
Unknown (yellow): Git commands not in the allowlist

Destructive commands require explicit confirmation before execution.

Prompt Caching

The tool uses API-level prompt caching to reduce costs:

Claude: Uses Anthropic's prompt caching beta (anthropic-beta: prompt-caching-2024-07-31)
OpenAI: Automatic prompt caching for supported models

Cache status is displayed when available:

(prompt cache hit: 150 tokens from cache)
(prompt cache created: 200 tokens cached)

License

MIT

Name		Name	Last commit message	Last commit date
Latest commit History 21 Commits
.github/workflows		.github/workflows
Commands		Commands
Configuration		Configuration
Models		Models
Properties/PublishProfiles		Properties/PublishProfiles
Providers		Providers
Services		Services
scripts		scripts
tests/GitAgentCli.Tests		tests/GitAgentCli.Tests
vscode-extension		vscode-extension
.gitignore		.gitignore
GitAgentCli.csproj		GitAgentCli.csproj
GitAgentCli.sln		GitAgentCli.sln
JsonContext.cs		JsonContext.cs
LICENSE		LICENSE
Program.cs		Program.cs
PromptTemplate.txt		PromptTemplate.txt
README.md		README.md
nuget.config		nuget.config

Folders and files

Latest commit

History

Repository files navigation

git-agent

Installation

Download Pre-built Binaries

Quick Install - Windows (PowerShell)

Quick Install - Linux

Quick Install - macOS

Build from Source

Windows (PowerShell)

Linux / macOS

Manual Build

Install as .NET Global Tool

Uninstall

Quick Start

Commands

run <instruction>

config

config show

config set <key> <value>

config get <key>

config use <provider>

config path

config reset

providers

conflicts

completions

cache

cache clear

cache path

IDE Integration

VS Code Extension

Installation

Usage

Settings

Visual Studio (Windows)

Setup External Tool

Assign Keyboard Shortcuts

Using Terminal

Configuration

Using OpenRouter

Benefits

Setup

Popular Models

Using Ollama for Local LLM

1. Install Ollama

2. Pull a Model

3. Start Ollama Server

4. Configure git-agent

5. Use git-agent

Troubleshooting

Safety Features

Prompt Caching

License

About

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases 1

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

`run <instruction>`

`config`

`config show`

`config set <key> <value>`

`config get <key>`

`config use <provider>`

`config path`

`config reset`

`providers`

`conflicts`

`completions`

`cache`

`cache clear`

`cache path`

Packages