Nativ: Run Frontier Open Models Locally on Your Mac
Nativ: Run frontier open models locally on your Mac
I built Nativ to put frontier intelligence directly on your Mac without cloud dependencies. This 100% open-source app runs models from Google and Cohere natively on Apple Silicon, offering real-time telemetry and full privacy. You get a free, forever solution with no subscriptions or accounts, ensuring you own your data and workflow end-to-end.
"Stop renting intelligence. Run it locally, in the open."
HN discussion
- Critics argue the post is astroturfing for Apple hardware, noting that running frontier models like Qwen 3.6 or Gemma 4 at useful token-per-second speeds requires expensive Mac Studio setups or dual 3090 GPUs, making cheaper laptops with cloud subscriptions a more cost-effective alternative.
- Practitioners report that on M1 Max chips, MLX-based tools offer little to no performance advantage over GGUF models in llama.cpp, suggesting MLX benefits are likely confined to M3 and newer Apple Neural Engine architectures.
- Users clarify that 'frontier open models' refers to the Pareto Frontier of intelligence versus cost, where models like DeepSeek V4 Pro represent the best performance for their price point rather than the absolute largest models available.
- While Nativ is a wrapper around MLX that may face portability bottlenecks, users highlight its potential for integrating audio and image generation modalities, with some finding small local models surprisingly capable at tool calling and specific tasks like photography analysis.