Hey Product Hunt! I'm Ryan, the solo developer behind Calcis. I built this after getting surprised by LLM API bills one too many times. A prompt tweak, a model swap, a new tool call, and suddenly the invoice looks nothing like what I expected. Calcis fixes that by telling you the cost before the request goes out. What surprised me building this is how deep the problem goes. It is not just "multiply tokens by price." Output length is unpredictable, thinking effort scales cost non-linearly, tool calls compound, and switching models mid-project can change your bill by 90%. So Calcis has a 4-layer prediction stack: an LLM predictor for precise mode, a regression model trained on 33k prompt and response pairs, a Bayesian confidence interval engine, and a heuristic fallback. You get a cost estimate with a P10 to P90 range, not just a single number. Here is what it actually does: Pre-flight cost estimation for 25+ models across OpenAI, Anthropic, and Google Session simulator that models full multi-turn conversations so you see cumulative cost before committing to an architecture Subscription quota predictor that tells you what percentage of your Claude, ChatGPT, or Gemini plan a session will consume Model comparison that suggests cheaper alternatives. Switching to the recommended model saves 60 to 90% per prompt on average Thinking effort slider that models reasoning cost for o3, Claude Opus, and similar models Agentic cost forecasting that accounts for tool call overhead and follow-up reasoning Live cost badges for READMEs, six unit converters, six workload calculators, and 20+ head-to-head model comparison pages It works across 10 distribution channels so it fits however you already build: Browser estimator at calcis.dev, no account needed CLI: npx calcis VS Code extension Browser extensions for Chrome and Firefox with live cost overlays on ChatGPT, Claude, and Gemini GitHub Action that posts cost breakdowns on every pull request npm packages for LangChain, LlamaIndex, Vercel AI SDK, and MCP 1,900+ unique npm downloads before today with zero promotion. Free to use. Happy to answer anything about how it works or what is coming next.

Forum Threads

p/calcis

•

2d ago

What's the most you've ever been surprised by an LLM API bill?

Building Calcis got me thinking about how often developers get caught off guard by unexpected API costs. What's your worst bill shock story? A prompt change that tripled costs? A model swap that wrecked your budget?

View all

Calcis

Know what your LLM calls cost before you make them.

Know what your LLM calls cost before you make them.

Forum Threads

What's the most you've ever been surprised by an LLM API bill?

Forum Threads

What's the most you've ever been surprised by an LLM API bill?