TokenGuard

TokenGuard is an intelligent hybrid LLM routing proxy that dramatically cuts API costs without sacrificing response quality. Instead of sending every prompt to an expensive paid model, TokenGuard automatically classifies each request by complexity — simple queries (summarization, translation, classification) are routed to Groq's free, lightning-fast local models, while complex tasks (code generation, reasoning, creative writing) are forwarded to Fireworks AI's powerful paid models. The application features a production-grade React dashboard with real-time analytics including total requests, cost savings, cache hit rates, and latency tracking. Users can test prompts live, compare model responses side-by-side, analyze cost trends over time, and monitor per-model performance metrics. The backend is powered by a deployed Supabase Edge Function that handles intelligent routing decisions, while PostgreSQL stores every request for complete auditability. Built with React 18, TypeScript, Vite, Tailwind CSS, and Framer Motion, TokenGuard demonstrates how a smart caching and routing layer can reduce LLM API spend by 40%+ with zero code changes on the client side — making advanced AI accessible and affordable.