Nativ: Run frontier open models locally on your Mac

**Nativ** reached the **Hacker News Front Page** as a tool designed to run frontier open models locally on your Mac. The software features a **curated library** containing standout open models from **Google**, **Cohere**, and **Liquid AI**. The initial lineup includes **Gemma 4 E2B** from Google at 10.28 GB with a 128K context window, **North Mini Code** from Cohere at 19.38 GB with a 500K context window, and **LFM2.5-VL 1.6B** from Liquid AI at 3.20 GB with a 128K context window.

The core product mechanics center on automatic hardware matching and local model execution. **Nativ** evaluates system hardware to recommend the right partner model for your specific Mac configuration. The technical footprints span a distinct range, from the lightweight 3.20 GB file size of Liquid AI's **LFM2.5-VL 1.6B** vision-language model to the 19.38 GB footprint of Cohere's **North Mini Code**, accommodating context lengths up to 500K.

For engineers and builders, local execution removes external API latency while providing direct access to specialized model architectures. Running models like **North Mini Code** with a 500K context window or **LFM2.5-VL 1.6B** locally on a Mac allows developers to build and test applications entirely on-device. The platform's recommendation feature removes setup friction by matching model requirements directly to local machine resources.

In the market context, **Nativ** enters the local AI tooling space by offering a curated partner library rather than an open-ended model index. By featuring selected releases from **Google**, **Cohere**, and **Liquid AI**, the app provides a focused range of options with memory footprints extending from 3.20 GB to 19.38 GB. This structured curation positions **Nativ** as an accessible local runner tailored for Mac hardware.

Developers can test **Nativ** to determine which partner models fit their specific Mac environment. Key aspects to evaluate going forward include the accuracy of the hardware recommendation engine across different system configurations and the operational performance of large-context models like **North Mini Code** and **Gemma 4 E2B** when executed locally.