See what Pura would do — without changing anything.
The shadow sidecar sits next to your current setup and logs what cascade routing would have done for each request. No traffic redirect. No config changes. Just data.
── INSTALL
npx @pura/shadow
The sidecar watches outbound LLM requests on your machine, runs them through the Pura confidence heuristic, and writes a comparison log. Takes about two minutes.
── WHAT YOU GET
Request #1 Your routing: GPT-4o $0.034 Pura cascade: Groq (tier 1) $0.0003 Savings: 99.1% Confidence: 0.94 Request #2 Your routing: GPT-4o $0.034 Pura cascade: Gemini (tier 2) $0.0008 Savings: 97.6% Confidence: 0.87 Request #3 Your routing: GPT-4o $0.034 Pura cascade: GPT-4o (tier 3) $0.034 Savings: 0% Confidence: 0.42 → escalated to tier 3 Summary: 73% average savings across 3 requests
── READY TO SWITCH?
When you like what you see, change one line:
const openai = new OpenAI({ baseURL: "https://api.pura.xyz/v1" });