Show HN: Shlexball, describe a file and a local 1.5B model finds it on your Mac (shlexball.com)

Shlexball has launched a local AI file search tool that allows users to find files on their Macs simply by describing them. This innovative application, which operates entirely offline using a 1.5 billion parameter model, has shown impressive performance by passing 308 out of 312 unseen file requests in its benchmarking tests. In comparison, Google's Gemini 3.8 Flash model passed 295 tests. Shlexball's approach includes pre-processing search terms to capture both singular and plural forms, ensuring that the search is comprehensive and fair across all models tested. This tool is significant for the AI/ML community as it showcases the capability of on-device AI to perform complex tasks, such as searching through the contents of files without uploading any data to the cloud. Shlexball can efficiently locate files based on partial descriptions, accessing names and text within PDFs and Word documents, which traditional search systems might miss. With an average response time of just 0.4 seconds on an M4 Max MacBook Pro and the promise of keeping user information confidential, Shlexball represents a notable advancement in AI-driven local file search technology, setting a high benchmark for future applications in privacy-focused AI solutions.