Kiln-AI/Kiln ? reverse-engineered prompt

Build me a local first desktop app and Python library for working on AI products, where I can create a task, run evaluations on model outputs, optimize prompts, and manage datasets all in one place.

I want the app to let me drag in documents for RAG, generate synthetic data, try agents or subagents, and compare results across different models. It should also make it easy for other people on my team to review outputs, rate them, and add examples without needing to use code.

Please make it work on macOS, Windows, and Linux, and keep it able to run with my own API keys or fully offline with Ollama. If you need to check current docs or best practices online while building it, go ahead. The main goal is a clean, usable tool that helps me iterate on prompts, evals, and fine tuning without losing track of what changed.