Hey Product Hunt 👋 I'm Frank, and Onisin OS is what happens when you spend a few years building case-analysis software for insurance and police and keep running into the same wall: every project rebuilds the same things — a data model, a CRUD UI, a search box, some reports. So I stopped rebuilding and started describing. In Onisin OS, you model your domain in a small text language — persons, cases, tickets, whatever you work with. From that one source of truth, the system derives the GraphQL schema, the forms users see, the semantic index, and the event streams. A ReAct agent turns natural-language questions into actual queries against your data. Pipelines combine structured data, documents and LLM steps into reports. Everything runs locally. Your data stays on your network. Bring your own LLM — OpenAI, Anthropic, Ollama, vLLM, anything that speaks the OpenAI API. What I'd love to hear: where does this break for your use case? Which domain would you try to model first? And — because this is the question that keeps me up at night — does the chat-first interaction actually work for your users, or do they still want the buttons? Happy to answer anything in the comments.

Onisin OS

Natural language meets your domain — locally

Natural language meets your domain — locally