Beta: Lekh AI lets Macs and iPhones run local LLMs

Local AI apps are getting real enough to feel less like hacker toys and more like normal software. That is the point of Lekh AI — a new app now being(betalist.com)rectly on a Mac or iPhone instead of sending prompts and files to cloud APIs. The pitch is simple: keep data on the device, cut latency, and use Appl(betalist.com)ke a compromise. BetaList’s listing and Lekh’s own app pages make clear this is not just chat — it (developer.apple.com)package. (betalist.com) ### What is Lekh AI, exactly? Lekh AI is a native Apple app for Mac, iPhone, and iPad that runs models locally for chat, image generation, text-to-speech, vision tasks, and document work. The company’s pages say it supports model formats including MLX and GGUF, while the iPhone and Mac versions emphasize offline use and no data leaving the device. The App Store listing frames the same idea even more bluntly — private, offline AI on Apple hardware. (lekhai.app) ### Why does “local” matter so much? Because cloud(betalist.com)y. If a model runs on your own hardware, prompts, files, and outputs do not need to leave the machine, which is a big deal for personal notes, drafts, PDFs, and voice input. It also changes the feel of the product. There is no waiting on a remote API and no per-call billing meter in the background. That does not make local models universally better, but it does make them attractive for everyday tasks where speed and privacy matter more than absolute frontier performance. (betalist.com) ### Why Apple hardware? Apple has spent the last few years turning on-device inference into a first-class feature of its stack. Core ML is built to use Apple silicon efficiently while minimizing memory and power use, and Apple’s tooling now explicitly talks about faster, more efficient on-device generative AI. That is the opening apps like Lekh are exploiting. The hardware is already in millions of laptops and phones, so the trick is no longer “can this run locally?” but “can this run locally without feeling slow?” (developer.apple. ([betalist.com)ml/)) ### What can you actually run? The short answer is: more than just one chatbot. BetaList and Lekh’s site name families like Llama, Qwen, Gemma, and Mistral for local LLM chat, then add Flux and Stable Diffusion image models, LTX video generation on Mac, and text-to-speech or audiobook-style workflows. The Mac product page also pitches this as a full “AI studio,” which is important — the product is trying to win on convenience, not just ideology. (betalist.com) ### Is th(developer.apple.com) App Store listing both say the app runs AI locally on iPhone, with Siri integration and shortcuts support highlighted in the broader product copy. But phones are still tighter on memory and thermals than Macs, so the practical experience will depend on model size and task type. Simple chat, voice, and lightweight generation are the obvious fit. Bigger creative workloads still look more natural on a Mac. That is not a flaw — it is just the physics of the device. (lekhai.app) ### What is the catch? Local AI still trades peak capability for control. The best cloud models remain stronger on many hard reasoning and multimodal tasks, and local generation can hit device limits fast — especially for video or larger models. Some of Lekh’s splashiest features also sit behind a Pro tier, including the broader creative suite. So the real comparison is not “local beats cloud.” It is “local is now good enough for a meaningful slice of daily work.” (betalist.com) ### Why is this showing up now? B(lekhai.app)up. Apple keeps improving on-device ML tooling, open-weight models keep getting smaller and better, and product directories like BetaList are now full of local-first apps instead of pure API wrappers. Lekh matters less as a one-off launch than as a signal. The center of gravity is shifting from “AI requires the cloud” to “the cloud is optional for a lot of tasks.” (betalist.com) ### Bottom line Lekh AI is a clean example of w(betalist.com)by default, fast enough to use, and broad enough to feel like software instead of a demo. The bigger story is not one app. It is that local AI on Macs and iPhones is starting to become normal. (betalist.com)