What will your AI app actually cost to run? Not the per-token price on the pricing page, but the real number after retries, caching, and scale.
BeforeYouShip is a new web tool that estimates LLM infrastructure costs before you write a line of production code. It covers GPT-4o, Claude, Gemini, and DeepSeek, and goes beyond simple token math by modeling the architectural factors that inflate real-world bills.
The tool accounts for four things most pricing calculators skip: retry costs from failed API requests, savings from prompt caching (where providers like Anthropic charge as little as 10% of normal price for cached input tokens), batch processing discounts, and growth projections at 3x and 10x your current usage.
That last feature is the most useful one. Plenty of solo developers have launched AI features that cost $50 a month at 100 users, then panicked when the bill hit $2,000 at 2,000 users because they never modeled the curve. BeforeYouShip gives you that projection up front.
The tool is free, runs in-browser, and loads pricing data dynamically. There's no public GitHub repo yet, and the project appears to be a solo-developer MVP. It's early-stage and limited in scope compared to production cost management platforms like Helicone, but it fills a different gap: the "should I even build this" phase rather than the "optimize what I already shipped" phase.
Useful for freelancers scoping AI features for clients, or anyone building a side project who wants to know the bill before committing to an architecture.