CCS Logo - Claude Code Switch multi-account AI tool for seamless model switching

CCS

Claude Code Switch

One command
Zero downtime
Multiple accounts

Run work, personal, and team Claude accounts simultaneously.
Plus cost-optimized GLM 4.6 and Kimi for Coding fallbacks.

$ bun add -g @kaitranntt/ccs
ccs
$
ccs glm
$

CCS v3.0 Terminal Demo

Use ccs for Sonnet 4.5 (complex tasks) • ccs glm for GLM 4.6 (routine work)

Why Developers Love CCS

Parallel workflow orchestration without session limit anxiety.
Zero config hell. Unlimited flow state.

🔄

Parallel Workflow Execution

Run Claude and GLM simultaneously with ccs in parallel.
Never hit session limits mid-work. Flow state preserved.

Unlimited flow state
Deep focus immersion
📈 3.4x productivity multiplier when interruptions eliminated
🔒

Work/Personal Separation

Keep work and personal contexts completely isolated with ccs work and ccs personal.
No logout dance. No cross-contamination.

Privacy, organization
Mental boundary peace
📈 40% faster context switching between projects

Zero-Config Philosophy

Install in 30 seconds.
Run ccs immediately with Claude.
Add GLM in just 5 minutes.

Time saved, simplicity
Instant gratification
📈 Setup in 30 seconds vs 2 hours for alternatives
Plus Model Optimization
💰

Unlimited Flow State

Combine your existing Claude with GLM for parallel workflow orchestration.
Strategic planning with Claude, execution with GLM. No more session limits.

Resource efficiency, sustainability
Smart resource optimization
📈 Maximize your existing subscriptions without hitting limits
🔍

Zero Telemetry, Zero Magic

Inspect the code yourself. No telemetry. No API proxying.
Just a thin routing layer over official Claude CLI. Open source, MIT license.

Trust, transparency
Complete peace of mind
📈 Full visibility into your AI workflows
🌐

Works Everywhere

Runs seamlessly on Linux, macOS, and Windows.
Works with bash, zsh, fish, and PowerShell. Install via npm install -g @kaitranntt/ccs.

Portability, reliability
Workflow consistency
📈 Zero retraining on different machines

One command. Zero downtime.
Right account, right model, right task.

Get Started
CCS is 100% FREE and Open Source

Choose Your Path

CCS costs $0. You only pay for the AI providers you choose.

Most developers use a hybrid approach - subscriptions for quality, APIs for volume.

Most Popular
💳

Subscription Path

Zero-config simplicity

Use your existing AI subscriptions. No API keys, no usage tracking, no surprise bills.

Providers

Claude Pro/Team

Strategic planning, complex reasoning

$20-30/mo
Gemini (OAuth)

Quick iterations, code generation

Free tier
Codex (OAuth)

GPT-4o, o1 models

ChatGPT Plus

Benefits

  • Predictable monthly cost
  • No API key management
  • Generous rate limits
  • Premium model access

Trade-offs

  • Session limits on heavy days
  • Less granular model control

Best for: Individual developers, small teams, those who prefer subscription simplicity

🔑

API Path

Pay-per-token flexibility

Bring your own API keys. Pay only for what you use with full model control.

Providers

GLM 4.6

Cost-optimized bulk tasks

~$0.50/1M tokens
Kimi

1M context, thinking mode

~$0.25/1M tokens
Custom

Any OpenAI-compatible endpoint

Varies

Benefits

  • Pay only for usage
  • No session limits
  • Full model selection
  • Cost optimization

Trade-offs

  • API key management
  • Variable monthly costs

Best for: Heavy users, budget-conscious developers, enterprises with existing API credits

Pro Tip: Most power users combine both paths - Claude for strategy, GLM for bulk tasks.
New in v5.0

Zero-Config OAuth Models

Access premium AI models with browser authentication. No API keys, no config files. Just ccs gemini and start coding.

How OAuth authentication works (click to expand)

OAuth providers use browser-based authentication instead of API keys. Authenticate once per provider, tokens are cached and refresh automatically.

Available OAuth Providers
gemini → Gemini 2.5 Pro, 2.0 Flash, Flash Lite
codex → o1, GPT-4o, GPT-4o-mini
agy → Antigravity Pro, Turbo
Authentication Flow
1. Run ccs gemini (first time)
2. Browser opens → Complete OAuth login
3. Token cached at ~/.ccs/cliproxy/auth/
4. Future runs → Instant, no re-auth needed
Tokens refresh automatically via OAuth 2.0 - more secure than static API keys
Terminal 1: Gemini via OAuth
$ ccs gemini "Explain this codebase architecture"
> Opening browser for authentication...
> Auth complete. Token cached at ~/.ccs/cliproxy/auth/gemini/
> Using model: gemini-2.5-pro
> Analyzing codebase structure...
Active session • OAuth authenticated
Models Available
gemini-2.5-pro, gemini-2.0-flash
Free tier available • No API key needed
Terminal 2: Codex via OAuth
$ ccs codex "Implement the authentication flow"
> Opening browser for authentication...
> Auth complete. Token cached at ~/.ccs/cliproxy/auth/codex/
> Using model: gpt-5.1-codex-max
> Implementing authentication endpoints...
Active session • OAuth authenticated
Models Available
o1, gpt-4o, gpt-4o-mini
ChatGPT Plus subscription • Premium models
ccs gemini | ccs codex | ccs agy
🔐

Zero API Key Management

No API key signup, no environment variables, no config files. Browser-based OAuth authentication takes 30 seconds - authenticate once, use forever.

🌐

Premium Model Access

Access Gemini 2.5 Pro, GPT-4o, o1, and Antigravity models through your existing subscriptions. Same models, simpler authentication.

🔄

Automatic Token Refresh

OAuth 2.0 tokens refresh automatically. No manual key rotation, no expiration surprises. More secure than static API keys.

💻

Headless Mode Available

Working on a server without browser? Use --headless flag to get a URL, complete auth on another device. Full flexibility.

Quick Reference

Basic Usage
ccs gemini "your prompt"
ccs codex "your prompt"
ccs agy "your prompt"
Auth Management
ccs gemini --auth # Auth only, no session
ccs gemini --logout # Clear tokens
ccs gemini --headless # No browser mode

Concurrent Account Workflows

v3.0's login-per-profile architecture enables true multi-account switching. Run work and personal Claude subscriptions simultaneously—zero credential sharing.

One-time setup (click to expand)
$ ccs auth create work
> Opening Claude CLI for login... > Profile 'work' created at ~/.ccs/instances/work
$ ccs auth create personal
> Opening Claude CLI for login... > Profile 'personal' created at ~/.ccs/instances/personal
💡 Only need to run this once per profile. Each profile gets its own Claude CLI instance.
Terminal 1: Work Account
$ ccs work
Using profile: work
$ ccs --version
Claude Code v3.0.0
> Implement user authentication for enterprise dashboard
Active session • Isolated instance
Terminal 2: Personal Account
$ ccs personal
Using profile: personal
$ ccs --version
Claude Code v3.0.0
> Write blog post about v3.0 multi-account workflows
Active session • Isolated instance
Both running simultaneously
🔒

Complete Isolation

Work context never mixes with personal. Zero credential sharing between profiles.

Zero Downtime

Hit rate limit on work account? Switch to personal instantly, keep coding.

💳

Separate Billing

Different subscriptions, separate usage tracking. Clear cost allocation.

📝

Independent State

Each profile has own todos, sessions, and logs. True isolation.

Intelligent Model Delegation

Strategic model selection based on task complexity. Use Sonnet 4.5 for planning and architecture, GLM 4.6 for execution and fixes. Automatic routing optimizes quality and cost simultaneously.

Understanding model delegation (click to expand)

CCS enables intelligent model delegation—matching the right AI model to each task based on complexity. Premium models for premium problems, cost-effective models for routine work.

The Delegation Strategy
1. Planning Phase → Sonnet 4.5
Complex reasoning, architecture, ambiguous problems
Best for: /plan, /brainstorm, /debug
2. Implementation Phase → GLM 4.6
Straightforward execution, routine fixes, git ops
Best for: /code, /fix:fast, /git:cm
3. Automatic Routing → ccs-delegation skill
Analyzes complexity, chooses optimal model
💡 Real developers save $500-1000/month with strategic delegation
Terminal 1: Planning with Sonnet 4.5
// Complex architectural planning
$ /plan "Build multi-tenant authentication with role-based access control"
// Deep debugging and root cause analysis
$ /debug "Race condition causing duplicate order creation"
// Exploring advanced design patterns
$ /brainstorm "Database sharding strategy for 100M+ users"
> Analyzing system architecture...
Active routing • Optimal for planning
Why Sonnet?
Sonnet 4.5 excels at ambiguous problems requiring creativity, complex system architecture, and deep debugging with multiple trade-offs.
Premium tier • Best for complex work
Terminal 2: Execution with GLM 4.6
// Straightforward implementation from plan
$ /code "Implement JWT authentication endpoints following plan.md"
// Routine bug fixes and corrections
$ /fix:fast "TypeScript errors in UserService after interface change"
// Stage and commit completed work
$ /git:cm
> Committed 3 files • Ready for next task
Active routing • Optimal for execution
Why GLM?
GLM 4.6 handles straightforward implementation efficiently when given clear specifications or well-defined patterns.
80% cheaper • Handles 70% of daily tasks
Strategic model routing • 70% cost savings
🎯

Premium for Premium Problems

Use Sonnet 4.5 for complex reasoning, system architecture, and ambiguous problems requiring creativity. Save premium model usage for work that truly needs it.

💰

Cost-Effective Execution

Delegate routine implementation, quick fixes, and git operations to GLM 4.6. Handle 70% of daily tasks at 20% of the cost without sacrificing quality.

Automatic Routing

Use ccs-delegation skill for intelligent task routing. Analyzes complexity and automatically selects optimal model—no manual switching required.

📊

Measurable Savings

Real developers report $500-1000/month savings using delegation strategy. GLM handles 70% of tasks at 80% lower cost while maintaining quality.

Different Tools, Different Needs

CCS and Claude Code Router (CCR) both solve real problems for developers.
Here's how to know which fits your situation.

Claude Code Router

You Might Need Claude Code Router If...

  • You're an extremely powerful user managing multiple paid API endpoints
  • You need advanced routing across different API services (OpenAI, OpenRouter, etc.)
  • You require custom endpoint configuration and routing control
  • You want flexibility to switch between multiple API providers
Learn about CCR

CCS

You Might Need CCS If...

  • You're a normal developer who prefers subscription-based simplicity
  • You want parallel workflow orchestration without complexity
  • You need work/personal account isolation without configuration
  • You want zero-config setup (install → ship in minutes)
Get Started with CCS

Both tools serve different developer needs.
CCR enables advanced routing for power users managing multiple paid API services.
CCS provides subscription-based simplicity for normal developers.
Use the right tool for your workflow.

We deeply respect the CCR team's work enabling flexible API routing for advanced users.

Frequently Asked Questions

Clear answers to common misconceptions about CCS.
Understand how parallel workflow orchestration really works.

Do I need to run ccs auth create to start using CCS?

Common misconception Many users think they need complex setup to start, but CCS works out-of-box with existing Claude access.

No! You only need ccs auth create if you have MULTIPLE Claude accounts. If you have one Claude subscription, just install and run ccs - it works immediately. The ccs auth create command is an advanced feature for power users who want work/personal separation.

setup Still have questions? Open an issue

Should I use OAuth providers (gemini, codex, agy) or API key models (glm, kimi)?

Common misconception Users think OAuth is less secure, but it uses standard OAuth 2.0 with token refresh - more secure than static API keys.

OAuth providers are fastest for quick testing - browser auth takes 30 seconds, no API key signup needed. Models like Gemini (gemini-2.5-pro) and Codex (gpt-5.1-codex-max) work instantly after authentication.

API key models are better for production - explicit key control, no browser dependency, works in headless environments. GLM-4.6 and Kimi require editing ~/.ccs/*.settings.json but offer more control over credentials.

Best practice: Use OAuth for exploration, API keys for production workflows.

setup Still have questions? Open an issue

Is CCS just for switching when I hit session limits?

Common misconception CCS is for reactive switching, not proactive parallel workflows.

No! CCS is designed for parallel workflow orchestration from the start. Run ccs for strategic planning and ccs glm for execution simultaneously in different terminals. This proactive approach prevents session limits entirely, rather than reacting when you get blocked.

workflow Still have questions? Open an issue

How is CCS different from Claude Code Router?

Common misconception Users think CCS is a 'sucker version' of CCR, but they optimize for completely different user needs and technical requirements.

They solve different problems for different users. CCR is designed for extremely powerful users who need advanced routing across multiple paid API endpoints and custom configuration. CCS is optimized for normal developers who prefer subscription-based simplicity and parallel workflow orchestration. If you need flexible endpoint configuration across multiple paid API services, use CCR. If you want zero-config setup and continuous productivity, use CCS.

comparison Still have questions? Open an issue

Is CCS a complex multi-account manager?

Common misconception All users need complex configuration, when most just want simple parallel workflows.

No! CCS is a simple wrapper that enables parallel workflows. For one Claude account, it's zero-config: install → run → ship. The multi-account features are optional for power users who want work/personal separation. We follow anti-complexity principles - inspect the code yourself!
complexity Still have questions? Open an issue

What's the real cost and value proposition?

Common misconception Users don't understand the economic value of parallel workflows.

$13/month (1 Claude Pro + 1 GLM Lite) = unlimited flow state. Strategic planning with Claude, execution with GLM. Never hit session limits. Zero config anxiety. Compared to Claude Max at $100-$200/month with same session limits, CCS saves you money while eliminating productivity interruptions.
economics Still have questions? Open an issue

Does CCS work with my existing Claude setup?

Common misconception Users think they need to change their existing setup, but CCS integrates seamlessly.

Yes! CCS uses your existing Claude CLI configuration and API keys. It's a thin wrapper that doesn't modify your settings. Just install CCS and it immediately works with your current setup. Zero migration required.
integration Still have questions? Open an issue

Ready to eliminate session anxiety?

Get started with CCS in 30 seconds. Parallel workflows, zero config, unlimited flow state.

Installation & Removal

Install via npm or one command. Zero downtime, multiple accounts.

bash
bun add -g @kaitranntt/ccs

Ultra-fast JavaScript runtime with blazingly quick installs.

Quick Start

  • Cross-platform: Works on macOS, Linux, and Windows
  • Easy updates: bun add -g @kaitranntt/ccs
  • Version pinning: Install specific versions with @kaitranntt/ccs@version

After Installation

  • 1. npm users: ccs is available immediately. No terminal restart needed.
    One-liner users: Restart terminal or run source ~/.bashrc
  • 2. Verify installation: ccs --version
  • 3. Start using:
    ccs "prompt with Claude"
    ccs gemini "prompt here" (OAuth - browser opens)
    ccs glm "prompt here" (requires API key setup)
  • 4. API key models (glm, kimi): Configure your API key in settings:
    ~/.ccs/glm.settings.json
    Update the apiKey field. Same pattern for kimi, gemini, etc.

Requirements:

  • npm installation: Node.js 14+ (Download Node.js →)
  • All methods: Claude CLI must be installed (Get Claude CLI →)
  • OAuth providers (gemini, codex, agy): Browser for authentication (or use --headless)