Personal GPT

Personal GPT occupies the category of AI apps that trade cloud convenience for control. Rather than posting your prompts to a remote server, it loads quantised open-source models such as DeepSeek R1 Distill, Llama 3.3, Qwen3, Phi 4 and Google Gemma directly onto your Apple hardware and generates responses there. That design has three practical consequences: your conversations are not logged or reviewed by a vendor, the app keeps working on a plane or in a remote area, and heavy usage does not rack up per-token API bills.

The app covers the whole Apple ecosystem. A single purchase unlocks it across iPhone, iPad and Mac, and Family Sharing can extend access to up to six people. On macOS it also hooks into the system text services, so you can select text in almost any app, right-click and have it rewritten, summarised or corrected on-device. On iOS it integrates with Siri and the Shortcuts app, which means AI steps can be dropped into automations that summarise text, draft content or feed results into other apps via the x-callback-url specification.

It is built by a small bootstrapped team based in the EU and sells on a one-time purchase model rather than a subscription. The app offers a large library of downloadable models with guidance on which ones suit a given device, RAM budget or use case, and it uses OmniQuant and GPTQ quantisation methods to try to preserve output quality at reduced memory footprints. Anyone evaluating it should treat it as a local-first alternative to cloud chat assistants rather than a full replacement for the largest frontier models.