npm install -g @vendodev/flipswitch
flipswitch        # sign in with Vendo
flipswitch on     # enable routing

Sign in with Vendo, flip the switch, restart Claude Code. You're routed through Vendo with access to 300+ models via OpenRouter.

Why

• No Anthropic subscription required. Claude Code's CLI is free to install. Flipswitch routes API calls through Vendo — you never need a Pro, Max, or Team plan.
• Dead simple. Sign in with Vendo, run flipswitch on, done. No config files to edit, no env vars to manage, no proxy servers.
• Access 300+ models. Remap Claude Code's model slots to any model on OpenRouter — Gemini, GPT, DeepSeek, Llama, Devstral, and more.
• Profiles. Switch between curated open-source model presets — max, speed, budget, or free.
• Pay as you go. With Vendo, you get a managed OpenRouter API key with built-in credit limits. No surprise bills.

Install

# npm
npm install -g @vendodev/flipswitch

# Homebrew
brew tap vendodev/tap
brew install flipswitch

Requires Node.js 18+.

Getting Started

Sign in with Vendo

$ flipswitch

  Welcome to Flipswitch
  Use Claude Code with any model. No Anthropic subscription needed.
  Powered by Vendo (https://vendo.run)

  Opening Vendo in your browser...
  Logged in via Vendo.
  Run `flipswitch on` to start routing Claude Code through Vendo.

Enable routing

$ flipswitch on

  Flipswitch ON
  Claude Code will now route through Vendo.
  Restart any running Claude Code sessions for changes to take effect.

Restart Claude Code and you're all set.

Already have an OpenRouter key?

If you already have an OpenRouter API key, you can skip Vendo:

flipswitch key sk-or-v1-your-key
flipswitch on

Commands

| Command | Description | |---|---| | flipswitch | Sign in with Vendo (default) | | flipswitch login | Sign in with Vendo | | flipswitch logout | Sign out and disable routing | | flipswitch on | Enable OpenRouter routing | | flipswitch off | Disable routing, revert to direct Anthropic | | flipswitch status | Show current configuration | | flipswitch profile <name> | Apply a preset model profile | | flipswitch model <slot> <id> | Remap a model slot to any OpenRouter model | | flipswitch models | Browse available models | | flipswitch key <key> | Use your own OpenRouter key instead of Vendo |

Profiles

Switch all three model slots at once with a single command. All profiles use open-source models:

flipswitch profile max      # open-source frontier
flipswitch profile speed    # fast iteration
flipswitch profile budget   # cheapest that still codes well
flipswitch profile free     # zero cost, rate-limited
flipswitch profile default  # reset to standard Claude models

| Profile | Sonnet slot | Opus slot | Haiku slot | Cost | |---------|-------------|-----------|------------|------| | max | MiniMax M2.5 (80% SWE-bench) | GLM-5 (78% SWE-bench) | Qwen3 Coder (480B MoE) | $0.22–$2.30/M | | speed | DeepSeek V3.2 | Qwen 3.6 Plus (1M ctx) | Gemma 4 31B | $0.14–$1/M | | budget | Qwen3 Coder Next | DeepSeek V3.2 | Devstral Small | $0.10–$0.75/M | | free | Qwen3 Coder | Qwen 3.6 Plus (1M ctx) | Nemotron 3 Super (262K) | $0 |

Free models are subsidized by OpenRouter and community inference providers — genuinely $0, no credit card needed. Rate limits: ~20 req/min, ~200 req/day.

Profiles set all three slots at once. You can still override individual slots with flipswitch model after applying a profile.

How do these compare to Claude?

In a real-world test of 33 models, only 4 produced working code — Claude Opus, Claude Sonnet, GLM-5, and GLM-5.1. The rest hallucinated APIs that don't exist. MiniMax M2.5 scores 80% on SWE-bench (nearly tied with Opus at 81%) but still invented fake methods in practice.

Open-source models handle ~80% of routine coding — single-file edits, well-known frameworks, quick iterations. They struggle with long agentic chains, niche libraries, and multi-file refactors.

See docs/open-source-models.md for the full breakdown: tier list, what goes wrong, cost per session, and every model's real-world track record.

Model Remapping

You can also remap individual slots to any of the 300+ models on OpenRouter:

# Route Sonnet requests to Gemini 2.5 Pro
flipswitch model sonnet google/gemini-2.5-pro

# Route Haiku requests to a cheaper model
flipswitch model haiku deepseek/deepseek-v3.2

# See what's available
flipswitch models

# Reset a slot back to default
flipswitch model sonnet --reset

Slots: sonnet, opus, haiku.

How It Works

Flipswitch writes env vars to ~/.claude/settings.json, Claude Code's official config mechanism:

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://openrouter.ai/api",
    "ANTHROPIC_AUTH_TOKEN": "sk-or-v1-...",
    "ANTHROPIC_API_KEY": ""
  }
}

Both Vendo and direct key users route through OpenRouter's Anthropic-compatible API. No background process, no proxy server, no shell profile modifications. All existing settings (permissions, hooks, plugins) are preserved. flipswitch off removes only the keys it set.

FAQ

Do I need an Anthropic subscription? No. Flipswitch bypasses Anthropic entirely. You only need the free Claude Code CLI and a Vendo account (or your own OpenRouter key).

Do I need to restart Claude Code? Yes, after flipswitch on, flipswitch off, flipswitch profile, or flipswitch model. Claude Code reads settings at startup.

What does Vendo do? Vendo provisions a managed OpenRouter API key with built-in credit limits. You sign in once, buy credits on Vendo, and flipswitch handles the rest. No OpenRouter account needed.

Can I use my own OpenRouter key instead? Yes — run flipswitch key sk-or-v1-your-key. This routes directly to OpenRouter, bypassing Vendo.

What about rate limits? Anthropic's rate limits don't apply since requests don't go to Anthropic.

Is this safe? Yes. Flipswitch uses Claude Code's documented env var override feature. It makes atomic file writes to prevent corruption and tracks which keys it owns.

Development

git clone https://github.com/vendodev/flipswitch
cd flipswitch
npm install
npm run build
npm test
./bin/flipswitch --help

License

MIT