New open-source tools aimed at local AI deployment and Windows search functionality have been released by independent developers, according to recent reports on Reddit and GitHub. The releases include a benchmarking platform designed to match hardware specifications with compatible large language models (LLMs) and a specialized search utility for Windows users.
Hardware-Specific AI Benchmarking Platform
A new open-source project, hosted at whatmodel.sh, has been launched to assist users in identifying which AI models can run locally on their specific hardware. The tool allows users to input their available RAM and GPU capacity to receive a ranked list of compatible models, including the specific commands required to execute them.
According to the developer, the platform aggregates data from 19,731 published benchmark results to assign a single score to each model. The project is open-source, with the scoring code and benchmarking scripts available on GitHub. The developer noted that the tool can demonstrate that high-performance models, such as the Qwen 3.8 27B at 4-bit quantization, can run on consumer-grade hardware like a 32GB Mac.
While the tool has received positive feedback for its utility, some users on Reddit have raised technical queries regarding its interface and performance reporting. Specifically, users noted difficulties in inputting custom RAM/VRAM configurations and requested the inclusion of tokens per second (t/s) metrics. There were also reports from users regarding discrepancies between the site’s predicted performance and actual real-world speeds when using optimization tools like Strata.
LSearch: Localized Search for Windows
In a separate development, a developer has released LSearch, a search utility for Windows designed to function similarly to macOS’s Spotlight. The application utilizes embedding models to facilitate advanced search capabilities.
The tool employs the embeddinggemma 2 (270M) model for its embedding functionality, though the developer noted that image support is currently unavailable. LSearch also features a Mixture of Experts (MoE) architecture, which combines two fine-tuned Qwen 3.5 4B models—one optimized for coding tasks and another for agentic workflows.
The application is designed to automate model management; for users with 16GB of RAM or more, the app prompts an automatic download of the necessary models. The source code and executable files are hosted on GitHub. Early user feedback has expressed interest in the tool, with some requesting expanded multimodal support for audio and video, as well as the adoption of the safetensor format.