A lightweight Gemini CLI written in Go
A love letter to Google's Gemini CLI

Why gmn? • Installation • Quick Start • Chat Mode • Tools • MCP • Benchmarks

The official Gemini CLI is an amazing tool with excellent MCP support and seamless Google authentication. However, for scripting and automation, its Node.js runtime adds startup overhead.

gmn reimplements the core functionality in Go, achieving ~40x faster startup while maintaining full compatibility with the official CLI's authentication.

$ time gmn "hi" > /dev/null
0.02s user 0.01s system

$ time gemini -p "hi" > /dev/null
0.94s user 0.20s system

• Fast startup — Native Go binary, no runtime overhead
• Interactive chat mode — Rich TUI with multi-turn conversations
• Built-in tools — File operations, web search, shell commands
• YOLO mode — Skip confirmations for automated workflows (--yolo)
• Session stats — Token usage tracking with Ctrl+C graceful exit
• Gemini 3 Pro support — Full compatibility with gemini-3-pro-preview
• MCP support — Connect to Model Context Protocol servers
• Credential reuse — Uses existing Gemini CLI authentication

gmn does not have its own authentication. You must authenticate once using the official Gemini CLI first:

npm install -g @google/gemini-cli
gemini  # Choose "Login with Google"

gmn reuses these credentials automatically from ~/.gemini/. Your free tier quota or Workspace Code Assist quota applies.

go install github.com/linkalls/gmn@latest

Download from Releases

# Simple prompt (one-shot)
gmn "Explain quantum computing"

# Interactive chat mode
gmn chat

# Chat with specific model (tab completion available)
gmn chat -m gemini-3-pro-preview

# Chat with initial prompt
gmn chat -p "Review this codebase"

# With file context
gmn "Review this code" -f main.go

# Pipe input
cat error.log | gmn "What's wrong?"

# JSON output
gmn "List 3 colors" -o json

Start an interactive session with a rich TUI and tool execution support:

gmn chat                              # Default model (gemini-2.5-flash)
gmn chat -m gemini-3-pro-preview      # Use Gemini 3 Pro
gmn chat -p "explain this codebase"   # Start with a prompt
gmn chat -r last                      # Resume the last session
gmn chat -r my-project                # Resume a named session
gmn chat --yolo                       # Skip all confirmations (dangerous!)
gmn chat --shell /bin/zsh             # Use custom shell

╭──────────────────────────────────────╮
│  ✨ gmn   gemini-3-pro-preview       │
│  📁 /path/to/your/project            │
╰──────────────────────────────────────╯
Type /help for commands, /exit to quit

❯ _

• Rich header — Model badge, working directory, YOLO indicator
• Thinking indicator — Spinner while waiting for response
• Tool notifications — Visual feedback for tool calls
• Session persistence — Auto-save conversations, resume anytime
• Session stats — Token usage on exit (including Ctrl+C)
• Tab completion — Auto-complete models and commands
• Command history — Navigate with Up/Down arrows

In chat mode, Gemini can automatically call these tools:

For dangerous operations, gmn shows a rich confirmation dialog:

╭─ Allow Shell Command? ────────────────────╮
│  Command: rm -rf ./build                  │
│                                           │
│  [y] Yes  [n] No  [a] Always allow shell  │
╰───────────────────────────────────────────╯

Use --yolo to skip all confirmations (be careful!).

gmn [prompt] [flags]
gmn chat [flags]
gmn mcp <command>

Commands:
  chat                         Start interactive chat session
  mcp list                     List MCP servers and tools
  mcp call <server> <tool>     Call an MCP tool

Global Flags:
  -p, --prompt string          Prompt (alternative to positional arg)
  -m, --model string           Model (default "gemini-2.5-flash")
  -f, --file strings           Files to include
  -o, --output-format string   text, json, stream-json (default "text")
  -t, --timeout duration       Timeout (default 5m)
      --debug                  Debug output
  -v, --version                Version

Chat Flags:
  -p, --prompt string          Initial prompt to send
  -m, --model string           Model (default based on tier)
  -f, --file strings           Files to include in context
  -r, --resume string          Resume a session (ID, name, or 'last')
      --yolo                   Skip all confirmation prompts
      --shell string           Custom shell path (default: auto-detect)

gmn supports Model Context Protocol servers.

Configure in ~/.gemini/settings.json:

{
  "mcpServers": {
    "my-server": {
      "command": "/path/to/mcp-server"
    }
  }
}

# List available tools
gmn mcp list

# Call a tool
gmn mcp call my-server tool-name arg=value

Measured on macOS/Linux. Windows startup may vary.

git clone https://github.com/linkalls/gmn.git
cd gmn
make build          # Current platform
make cross-compile  # All platforms

• OAuth flow → authenticate with official CLI first
• API Key / Vertex AI auth
• Some advanced official CLI features

Apache License 2.0 — See LICENSE

This project is a derivative work based on Gemini CLI by Google LLC.

• Google Gemini CLI — The incredible original
• OpenAI Codex CLI — TUI design inspiration
• Google Gemini API — The underlying API