GLM CLI - AI Code Generator

AI-powered code generation with streaming output. Generate code directly from your terminal with real-time feedback.

Installation Model

GLM CLI is installed globally, but Claude prompts can be installed globally or locally:

┌─────────────────────────────────────────────────────────┐
│  NPM Global Install (once)                              │
│  npm install -g glm-coding                              │
│  → glm command available system-wide                    │
└─────────────────────────────────────────────────────────┘
                         ↓
        ┌────────────────┴────────────────┐
        ↓                                 ↓
┌──────────────────┐            ┌─────────────────────┐
│ Global Setup     │            │ Local Setup         │
│ glm init -g      │            │ glm init            │
│                  │            │                     │
│ ~/.glm/          │            │ {project}/.glm/     │
│ ~/.claude/       │            │ {project}/.claude/  │
└──────────────────┘            └─────────────────────┘

Quick Start

1. Install GLM CLI Globally

npm install -g glm-coding

This installs the glm command system-wide.

2. Choose Your Setup Style

Option A: Global Setup (Shared Across All Projects)

glm init -g

This creates:

• ~/.glm/config.json - Your API key and settings
• ~/.glm/instructions/ - Code quality guidelines
• ~/.glm/profiles/ - Specialized agent profiles
• ~/.glm/logs/ - Usage statistics
• ~/.claude/CLAUDE.md - Instructions for Claude Code

Use when: You want one configuration for all projects.

Option B: Local Setup (Per-Project Configuration)

cd your-project
glm init

This creates:

• {project}/.glm/config.json - Project-specific API key/settings
• {project}/.glm/instructions/ - Project-specific quality guidelines
• {project}/.glm/profiles/ - Project-specific profiles
• {project}/.claude/CLAUDE.md - Project-specific Claude instructions

Use when: Different projects need different API keys, profiles, or quality standards.

Hybrid Approach (Global + Local Override)

# Set up global defaults
glm init -g

# Override in specific projects
cd special-project
glm init

Local settings override global ones. Logs always go to ~/.glm/logs/.

Usage

Generate Code

# Basic usage
glm -q "Create a function to validate emails"

# With profile
glm -q "React user profile component" -p frontend-design

# Save to file
glm -q "REST API client for GitHub" -o client.py -p api-integration

# Pipe input
echo "Parse JSON with error handling" | glm -o parser.py

Execute Mode (NEW in v0.6.0)

Generate and immediately execute code with automatic security validation:

# Execute generated code (default: Python)
glm -x -q "Print sum of 1 to 100"
# Output: 5050

# Generate content directly (poem, data, etc.)
glm -x -q "Write a haiku about coding"
# Output: [actual haiku printed]

# Save and execute
glm -x -q "Hello world" -o hello.py
# Output: /path/to/hello.py
#         Hello, World!

# Specify language
glm -x -l node -q "console.log('Hello')"

# Security check only (dry-run)
glm --dry-run -q "System information script"

# Force execution (bypass security - not recommended)
glm -x --force -q "some dangerous code"

Security Features:

• Automatic validation before execution
• Blocks dangerous imports: os, subprocess, shutil, etc.
• Blocks dangerous operations: file deletion, system commands
• Blocks dynamic code execution functions
• Detailed violation reports with line numbers
• Blocked code saved to temp directory for inspection

Available Profiles

Options

-q, --query <prompt>      Query prompt (required if no pipe)
-o, --output <file>       Save output to file (enables token tracking)
-p, --profile <name>      Use specific profile
-l, --language <lang>     Language for execution (python, node, bash)
-m, --max-tokens <num>    Maximum tokens (default: 20000, GLM max: 20K)
-x, --exec                Execute generated code after security check
--force                   Bypass security check (not recommended)
--dry-run                 Security check only, no execution
--no-quality              Disable quality instructions

Commands

glm init                  Initialize local configuration
glm init -g               Initialize global configuration
glm stats                 Show usage statistics (daily)
glm stats --monthly       Show monthly statistics
glm usage                 Alias for stats command
glm usage --monthly       Monthly statistics (alias)
glm version               Show version
glm help                  Show help

Claude Code Integration

NEW in v0.5.0: glm init now automatically installs Claude Code integration!

When you run glm init or glm init -g, the installer automatically sets up:

1. Hook System (Auto-Trigger)

Detect keywords in your prompts:

@glm Create a REST API client → Automatically activates GLM mode
-glm Parse JSON with validation → Activates GLM mode
--glm React button component → Activates GLM mode

The hook injects GLM-specific instructions when keywords are detected.

2. Slash Command

Explicit GLM invocation:

/glm REST API client → src/api.ts
/glm fibonacci function → utils/math.py
/glm 병렬로 여러 파일 생성

3. CLAUDE.md Instructions

Automatically adds GLM usage guidelines to ~/.claude/CLAUDE.md or .claude/CLAUDE.md

What Gets Installed

~/.claude/
├── hooks/
│   └── glm-detector.sh          # Keyword detection hook
├── commands/
│   └── glm.md                   # /glm slash command
├── settings.json                # Hook configuration (auto-updated)
└── CLAUDE.md                    # GLM usage guidelines

Manual Installation (Optional)

If you prefer manual setup, the hook script is available at:

~/.claude/hooks/glm-detector.sh

To verify hook installation:

# Check if hook is configured
cat ~/.claude/settings.json | grep -A 5 "UserPromptSubmit"

Configuration

Config File Structure

Global: ~/.glm/config.json Local: {project}/.glm/config.json

{
  "apiKey": "your-glm-api-key",
  "apiModel": "glm-4.6",
  "apiBaseUrl": "https://api.z.ai/api/coding/paas/v4/chat/completions",
  "apiPlan": "max",
  "maxRetries": 5,
  "timeout": 120000,
  "debug": false,
  "useQuality": true,
  "verboseLog": true,
  "enableLogging": true
}

Environment Variables (Fallback)

If config.json doesn't have a value, GLM checks environment variables:

export GLM_API_KEY="your-api-key"
export GLM_API_MODEL="glm-4.6"
export GLM_DEBUG="true"

Configuration Priority

1. Local config: {project}/.glm/config.json
2. Global config: ~/.glm/config.json
3. Environment variables: GLM_API_KEY, etc.
4. Default values

Usage Limits and Plans

GLM CLI supports three API plan tiers:

Note: Each prompt ≈ 15-20 model calls internally.

Setting Your Plan

Edit ~/.glm/config.json:

{
  "apiPlan": "pro"
}

Or set environment variable:

export GLM_API_PLAN="pro"

Monitoring Usage

View 5-hour usage window:

glm stats

Output:

5-Hour Usage Window
────────────────────────────────────────────────────────────────────────────────
Plan: PRO
Limit: 600 prompts per 5 hours
Used:  145 prompts (24.2%)

[█████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░] 24.2%

Claude Code Integration

When you run glm init (global or local), it automatically:

1. Updates CLAUDE.md with GLM usage guidelines
2. Preserves existing content (only adds/updates GLM section)

Using in Claude Code

The CLAUDE.md section includes:

• Complexity guidelines (Simple < 50 lines, Medium < 200 lines, Complex > 500 lines)
• Profile list (default, frontend-design, api-integration, database-ops, web-crawler)
• Usage example

This helps Claude understand when to suggest using GLM for appropriate tasks.

Usage Statistics

Track API usage with detailed statistics:

# View 5-hour usage + daily stats (both commands work)
glm stats
glm usage

# View monthly stats
glm stats --monthly
glm usage --monthly

Statistics include:

• 5-Hour Usage Window: Current usage vs plan limit with progress bar
• Historical Stats: Daily/monthly aggregated data (in your local timezone)
• Profile breakdown: Top 5 profiles used
• Request counts (total, successful, failed)
• Token usage (prompt, completion, total) - NEW in v0.7.0

Log locations:

• Token stats: ~/.glm/usage.jsonl
• Chat logs: ~/.glm/logs/chat-*.jsonl

Token Usage Tracking (NEW in v0.7.0)

Automatic token tracking when using -o flag:

# This captures token usage statistics
glm -q "create API client" -o api.py

# Check usage
glm stats

Output example:

Date          Requests  Success  Failed  Total Tokens    Prompt  Completion
────────────────────────────────────────────────────────────────────────────────
2024-12-10          10        10       0         45,231    32,148      13,083

How it works:

• Interactive mode (glm -q "prompt"): Streaming output, no token tracking
• File output mode (glm -q "prompt" -o file.py): Non-streaming, captures token usage
• No configuration needed - automatically optimizes for each use case

Controlling Output Length (NEW in v0.7.0)

Use --max-tokens to control output size and costs:

# Limit to 1000 tokens (good for simple functions)
glm -q "simple validation function" -o validator.py -m 1000

# Use full capacity for complex files (default: 20000)
glm -q "comprehensive REST API" -o api.py -m 20000

# Default is 20K, so no flag needed for complex tasks
glm -q "large application" -o app.py

Benefits:

• Cost control: Limit tokens for simple tasks
• No truncation: Default 20K handles complex files (increased from 4K)
• Flexible: Override per-request based on complexity

Project Structure

{user-home}/
├── .glm/                    # Global config (if using glm init -g)
│   ├── config.json
│   ├── usage.jsonl          # Token usage statistics (centralized)
│   ├── logs/
│   │   ├── chat-2025-12-01T10-30-45-123Z.jsonl
│   │   ├── chat-2025-12-01T11-15-22-456Z.jsonl
│   │   └── ...              # Per-run chat logs (detailed)
│   ├── instructions/
│   │   └── quality.txt
│   └── profiles/
│       ├── default/
│       ├── frontend-design/
│       ├── api-integration/
│       ├── database-ops/
│       └── web-crawler/
└── .claude/
    └── CLAUDE.md            # GLM section added here

{project}/
├── .glm/                    # Local config (if using glm init)
│   ├── config.json
│   ├── instructions/
│   └── profiles/
└── .claude/
    └── CLAUDE.md

Development

Prerequisites

• Node.js 18+
• npm

Setup

git clone https://github.com/your-org/glm-coding.git
cd glm-coding
npm install

Building

npm run build          # Build CLI
npm run dev            # Watch mode
npm run typecheck      # Type check without build
npm run clean          # Clean build output

Testing Locally

# Link package globally
npm link

# Test in a directory
cd /tmp/test
glm init
glm -q "hello world function"

# Unlink when done
npm unlink -g glm-coding

Repository Structure

glm-coding/
├── src/
│   └── cli/              # CLI source code
│       ├── cli.ts        # Entry point & router
│       ├── commands/     # Command implementations
│       │   ├── generate.ts   # Code generation
│       │   ├── init.ts       # Installation
│       │   ├── stats.ts      # Statistics
│       │   ├── help.ts       # Help
│       │   └── version.ts    # Version
│       └── core/         # Core modules
│           ├── glmClient.ts     # API client
│           ├── config.ts        # Config loader
│           ├── instructions.ts  # Quality loader
│           ├── profiles.ts      # Profile loader
│           ├── usageLogger.ts   # Usage logging
│           └── ...
├── templates/            # Installed to ~/.glm/ or {project}/.glm/
│   ├── instructions/
│   │   └── quality.txt
│   └── profiles/
│       ├── default/
│       ├── frontend-design/
│       ├── api-integration/
│       ├── database-ops/
│       └── web-crawler/
├── dist/                 # Build output (gitignored)
├── package.json
├── tsconfig.json
└── tsconfig.cli.json

Making Changes

1. Edit src/cli/ files
2. Run npm run build
3. Test with npm link
4. Commit and push

Publishing

npm version patch  # or minor, major
npm publish

Migration from MCP Server

If you're migrating from the old MCP server version:

# Old MCP installation will be detected
glm init -g

# Remove old MCP server manually
rm -rf ~/.claude/mcp-servers/glm-coding
# Remove "glm-coding" from ~/.claude/mcp.json

Examples

Simple Function

glm -q "Python function to calculate fibonacci"

React Component

glm -q "React card component with image, title, description" \
  -p frontend-design \
  -o components/Card.tsx

API Client

glm -q "REST client for GitHub API with auth, pagination, error handling" \
  -p api-integration \
  -o github_client.py

Database Schema

glm -q "PostgreSQL schema for e-commerce with users, products, orders" \
  -p database-ops \
  -o schema.sql

License

MIT

Support

For issues and questions, visit GitHub Issues.