Hey Product Hunt! I built Pith because I was tired of watching my OpenAI bills climb while knowing that most prompts are full of unnecessary tokens -- filler words, redundant instructions, verbose formatting that LLMs don't need. Pith is a proxy that sits between your app and OpenAI/Anthropic. It automatically optimizes every prompt before forwarding it -- stripping what's unnecessary while keeping meaning intact. Same output quality, fewer tokens, lower costs. Here's what makes it different: - 2-line integration: just swap your API key and base URL - Zero risk: if Pith can't reduce tokens safely, it sends the original prompt as-is - Free tier saves ~18% on average (rule-based algorithm) - Pro tier saves ~35% (algorithm + AI analysis combined) - Free tier actually beats GPT-4o-mini compression on verbose system prompts (43.8% vs 36.5%) Tech stack: FastAPI, Docker, DuckDB. Adds ~50ms latency per request. The optimization database grows with every API call, so savings improve over time without you changing anything. Would love your feedback! What features would matter most to you?

Pith

Cut LLM API costs ~30% with a single line change

Cut LLM API costs ~30% with a single line change