Hi Product Hunt, I'm the maker of JevStation. The problem: most "AI decisions" in production are a chat model asked to reply in JSON, then parsed, retried and hoped for. You count tokens, handle malformed output, and get no real confidence number. JevStation is the other approach. Decision models (Jev, Cloudflare Clef, Perplexity Decider, OpenAI Luna Decisions, Microsoft-Decision-1) take a state and typed questions (Choice, Score, Noul) and return the answer with its full probability distribution and a confidence value. No prose, nothing to parse, so your code can threshold and branch directly. What you can do today: - Try it with no sign-up in the playground, with presets for sentiment, intent, ticket triage, moderation and more - Call it from the REST API or the MCP server (agents can use it to guard tool calls, route models or moderate text) - Run spreadsheets through the batch engine, for example review sentiment, lead scoring or a site SEO audit - Compare the decision models side by side on one key Pricing is simple: a standard evaluation costs 1 credit, a failed one costs 0 (refunded automatically), and new accounts get 200 free credits. I'd love to hear which decisions you would put behind a threshold, and which question types are missing. Happy to answer anything.

JevStation

Typed decisions for your code: probabilities, not prose

Typed decisions for your code: probabilities, not prose