Hi Product Hunt 👋 I'm Soroush, and I built 41prompts for anyone who keeps a long prompt alive. A prompt you depend on grows one rule at a time. Examples get pasted in, a constraint gets reworded, and when the answers get worse there's no easy way to tell which change did it, or to get back the version that worked. 41prompts gives the prompt a shape: Bloks: split it into typed parts (context, constraint, example, expects). Each blok owns an exact span of the prompt you ship. Compiled prompt: it rebuilds as you type, fills {{variables}}, counts tokens, and copies in one click as text, Markdown or JSON. Expects bloks describe a good answer and stay out of the prompt. Versions: every save is one. Name the one that works and restore any of them without overwriting anything. Your models: 20+ hosted providers, any OpenAI-compatible endpoint, or a local model through Ollama or LM Studio. Local models run straight from your browser, and hosted providers bill you directly. Keys are encrypted at rest and never logged. All of this is free, with no cap on prompts, bloks or versions and no card to sign up. You can export everything as Markdown and JSON at any time. Next up is testing. Expects bloks will become checks you can run across models, a linter will flag contradictions and rules nothing can test, and a failed check will point at the blok behind it. That will be the paid Performance plan; everything above stays free. If you maintain long prompts, I'd like to know how you track changes today and what you'd test first. I'm here all day.

41prompts - Break long prompts into parts you can name and version

Replies