FAUXCLAW

Real Claude. Fake Bill.
A free AI proxy and standalone chat client that scratches your AI itch without scratching your wallet.

Fauxclaw routes requests through multiple free providers – AWS CodeWhisperer (Kiro), OpenRouter free tier, iFlow, NVIDIA NIM, Groq, Gemini, DeepSeek, Mistral, and Ollama – with automatic parallel racing (try all at once, take the fastest), circuit breaker failover, token refresh, and zero configuration beyond the initial setup.

No credit card. No Claude Code required. Just chat.

Quick Start

# Install globally
npm install -g fauxclaw

# Setup providers (Kiro OAuth, API keys)
fxc setup

# Start chatting immediately (no Claude Code needed)
fxc chat

Or use the proxy with any client:

fxc start
# Then set:
#   OPENAI_BASE_URL=http://localhost:8083/v1/chat/completions
#   ANTHROPIC_BASE_URL=http://localhost:8083
# Any API key works

That's it. Free AI, no bill.

What You Get

| Provider | Models | Cost | Auto Refresh | |---|---|---|---| | Kiro (AWS CodeWhisperer) | Claude Sonnet 4.5, Claude Haiku | Free (AWS Builder ID) | Yes | | OpenRouter | GPT-4o mini, Claude Opus 4 | Free tier | No | | iFlow | DeepSeek R1, Qwen, Llama | Free (token) | No | | NVIDIA NIM | Nemotron, Llama, Qwen | Free (40 req/min) | No | | Groq | Llama 3.3 70B, Mixtral | Free tier | No | | Gemini | Gemini 2.0 Flash, 1.5 Pro | Free tier | No | | DeepSeek | DeepSeek Chat, DeepSeek R1 | Free tier | No | | Mistral | Mistral Small, Large, Embed | Free tier | No | | Ollama | Any local model (Llama, Mistral, etc.) | Free (local) | No |

Parallel racing mode: Instead of trying providers one-by-one, Fauxclaw fires them all at once and returns the fastest response. Set FXC_RACE=false to revert to sequential fallback.

Commands

| Command | Description | |---|---| | fxc setup | Interactive provider setup wizard | | fxc chat / fxc c | Standalone interactive chat (no Claude Code required) | | fxc start | Start proxy server (for external clients) | | fxc status | Show provider health and token status | | fxc metrics | Request latency, success rates, provider breakdown | | fxc doctor | Diagnose system issues and configuration | | fxc purge | Delete expired conversation sessions | | fxc help | Display this help information |

Interactive Chat (standalone)

No Claude Code. No external client. Just run:

fxc chat

Then type messages directly. Commands:

| Command | Description | |---|---| | /exit or /quit | Exit chat | | /clear | Clear conversation history | | /status | Show current model, history count, providers | | /models | List all available models | | /model <name> | Switch to a different model | | /providers | Show configured provider chain | | /save | Save current model as default |

Web Dashboard

Start the proxy, then open: http://localhost:8083/dashboard

Real-time metrics, provider status, and an in-browser chat interface.

Configuration

Environment Variables

| Variable | Default | Description | |---|---|---| | FXC_PORT | 8083 | Port to listen on | | FXC_HOST | 127.0.0.1 | Bind address (use 0.0.0.0 for external) | | FXC_API_KEY | (none) | Require this in X-Proxy-Key header | | FXC_RATE_LIMIT | 100 | Requests per minute per IP | | FXC_RACE | true | Parallel provider racing (set false for sequential) | | FXC_LOG | info | debug | info | warn | error | | FXC_TIMEOUT | 30000 | Request timeout in milliseconds | | FXC_MAX_BUFFER | 5242880 | Max buffer size for streaming (5MB) |

Config File

All tokens and API keys are stored in ~/.fauxclaw/config.json with 0600 permissions.

How It Works

1. You type a message in fxc chat (or send it via the proxy)
2. Fauxclaw races all configured providers simultaneously
3. The first provider to respond wins
4. Failed providers are tracked by a circuit breaker (60s cooldown)
5. If all fail, falls back to sequential retry

Provider Routing

• Model prefix routing: use groq/llama-3.3-70b or ollama/llama3.1 to pin a provider
• No prefix = parallel race across all configured providers

Streaming Formats

Fauxclaw translates all provider formats to standard SSE:

• Kiro: AWS EventStream (binary) → Anthropic SSE
• OpenAI-compatible (OpenRouter, Groq, NVIDIA, Gemini, Mistral) → Anthropic SSE
• Ollama: JSON streaming → Anthropic SSE
• DeepSeek: Native Anthropic SSE passthrough

FAQ

Is this legal? Yes. You're using free tiers exactly as intended. No ToS violation, no credit card required.

Do I need an AWS account? A free one for Kiro (the best Claude provider). No billing info required.

Can I use this in production? Probably not a great idea, but you can enable FXC_API_KEY, bind to localhost, and add HTTPS.

Why "Fauxclaw"? Fake (faux) Claw (Claude). Lobsters have claws. Bad naming, great AI.

License

Mozilla Public License 2.0 (MPL-2.0)

npm install -g fauxclaw && fxc setup && fxc chat

"Fake it till you make it, but today you made it."