Show HN: Switchboard – route AI prompts instead of capping budgets (github.com)

Switchboard, a newly announced tool, aims to optimize AI prompt management by intelligently routing requests through various backend models, including local ones, instead of continuously relying on costly premium agents. In its benchmark tests, Switchboard achieved a 62% reduction in premium usage while maintaining a quality score of 4.1/5 compared to a 4.6/5 baseline for always-premium use, without leaks or missed responses. This innovative routing approach enables users to decide on-the-fly which AI model to engage based on the context and sensitivity of the prompt, effectively enhancing both budget control and privacy. The significance of Switchboard lies in its determination-first routing strategy, which prioritizes local processing for privacy-sensitive tasks. By default, it implements a strict privacy floor that prevents sensitive prompts from being sent to external servers. Additionally, it allows seamless transitions between models within a single session without losing context, using an embedded semantic memory for quick access to previous interactions. This not only optimizes resource use but also offers a practical framework for organizations facing spending caps on AI services, illustrating a shift towards more granular and efficient AI service management.