What problem does it solve? Sending every LLM request to a premium model wastes money, while manually choosing models per request is impractical. This Skill runs a self-hosted OpenAI-compatible proxy that classifies each request, routes it to the cheapest capable model across 12+ OpenRouter models, and records every routing decision in a ZK Merkle journal for auditability. ## Core Features & Use Cases - Tier-Based Smart Routing: A 14-dimension weighted classifier assigns requests to SIMPLE, MEDIUM, COMPLEX, or REASONING tiers in under 1ms with zero LLM cost, cutting inference spend by roughly 60-80%. - Mode Overrides: Force a tier with prefixes like /max, /complex, or [simple] without the LLM ever seeing the override token. - ZK-Stamped Audit Trail: Every routing decision is appended to a Merkle tree, with the root and index returned in response headers for verifiable replay. - Zero-Config Birth Provisioning: On first boot, the Birth bot mints a sponsored OpenRouter key via the x402.wtf control plane, so no signup or API key is required. - Use Case: Point your existing OpenAI-compatible agent or chat client at http://127.0.0.1:18800 and let simple queries hit Gemini Flash while complex coding tasks route to Claude Sonnet automatically. ## Quick Start Ask the AI to start the zk-router proxy and send a chat completion request to http://127.0.0.1:18800/v1/chat/completions using the model zkrouter/auto.