Show HN: AptSelect – A local LLM client for parallel testing and evaluation (aptselect.com)

A new tool called AptSelect has been introduced, enabling parallel testing and evaluation of various large language models (LLMs) like GPT-4, Claude 3, and Gemini. This local-first client allows users to send a single prompt to multiple models simultaneously, providing a side-by-side comparison of raw JSON outputs, latency metrics, and token consumption—eliminating the hassle of managing multiple browser tabs. AptSelect emphasizes a hands-on approach to prompt engineering, enabling users to iterate and benchmark effectively without relying on cloud services. Significantly, AptSelect enhances the ability of AI/ML practitioners to perform rigorous model comparisons by offering features such as local storage of API keys in an encrypted SQLite database, automatic saving of iterations, and the ability to inject test data for baseline comparisons. Users can tweak parameters like temperature and frequency penalties to observe their effect on output quality while maintaining a clear history of their modifications. Built on a straightforward stack using Electron for cross-platform compatibility, AptSelect prioritizes user security and data ownership, making it a valuable addition to the toolkit of anyone involved in AI model evaluation and testing.