Show HN: //Beforeyouship is a pre-build tool to estimate the LLM cost(llm-architecture-cost-modeler.vercel.app)
llm-architecture-cost-modeler.vercel.app
Show HN: //Beforeyouship is a pre-build tool to estimate the LLM cost
https://llm-architecture-cost-modeler.vercel.app/
5 comments
Really interesting angle — I've been working on the post-build side of this problem (finding waste in production LLM spend) and the pre-build estimation gap is something I hadn't thought about. The two tools solve different moments in the same problem: estimate before you build, profile after you ship. Would be curious whether your estimates hold up against what people actually spend in production — that delta seems like useful data for improving the model.
Retry logic modeling is a great inclusion most cost estimates miss this completely. In my experience retries account for 15-25% of total cost in production agent systems, especially with tool calling where validation failures trigger re-prompts. Would be useful to see a "worst case" scenario that models cascading retries (retry triggers another tool call that also retries).
This is really valuable, especially the 15-25% figure from real production systems — current assumption of 8% is clearly too conservative for agent workloads with tool calling. The cascading retry scenario (retry triggers another tool call that also retries) is a great idea for a worst case column alongside the current realistic estimate. Going to look into this for v2.
I think this would be super useful as a claude code / visual studio extension, so I can see the cost impact of changes I am making in code. For example by adding a prompt pre-processing function how much am I actually going to save in LLM token cost.
That's actually a great direction I haven't thought about — bringing the cost estimate closer to the code rather than keeping it as a separate planning step. The delta view (how much does adding this preprocessing function actually save) would be super useful. Adding it to the v2 roadmap, thanks for this.
So I built an MVP to test with you guys — if any of you relate to the problem, please tell me what works and what's missing. Already included retries, prompt caching, batch discounts, and 3×/10× growth scenarios across 6 models (GPT-4o, Claude, Gemini, DeepSeek and more). Also the app models full architecture, the user just needs to pick an app type and set the usage pattern.