Five iPhone apps now make local AI useful for more than a spot of novelty: they can tidy dictated notes, work with files or run voice chats without an internet connection. MakeUseOf’s Bryan M. Wolfe tested them on an iPhone 17 Pro Max, and the practical takeaway is to start with a smaller model, not the biggest name on the menu.
Watch Desk analysis
What happened
Wolfe’s guide covers Locally, Private LLM, PocketPal AI, Google AI Edge Gallery and Enclave. His tests included using Private LLM with Siri and Shortcuts to clean up dictated text, and asking Enclave questions about a PDF while the phone was in Airplane Mode. He also found that Locally can connect to LM Studio on a home computer, letting the Mac run a larger model while the iPhone acts as the remote.
The trade-offs differ. PocketPal lets users browse Hugging Face for GGUF models and select a quantisation level to suit their phone’s memory. Google AI Edge Gallery is more of a Gemma test bench, with tools such as Wikipedia lookups available through skills. Enclave supports local voice chat and documents, but can also connect to cloud models, so users should check which model is selected before sharing sensitive material.
Why it matters
Running a model locally can keep quick tasks on the device and make some features available without a connection. It does not make a phone a pocket-sized data centre: larger models need more memory and may respond more slowly, and locally run models lack live web access unless an app adds a tool.
Wolfe’s beginner’s advice is refreshingly concrete: try a 2B to 4B model first, check available memory before downloading, and look for Q4 quantisation as a smaller-file option. Model files can also take up space more than once because each app stores its own copy.
Our read
The useful shift is not that every phone can now replace a cloud assistant. It is that local models can handle specific jobs, from cleaning up notes to searching a document, without making every task a trip to the cloud. Start small, check where processing happens, and keep an eye on storage. Your phone’s free space is not an infinite AI pantry.
What to watch
- Whether more apps make local models easy to reach through Siri, Shortcuts or device buttons.
- How performance varies across phone models and model sizes.
- Whether apps make it clearer when a task is processed locally or sent to a cloud model.
Discussion spark: Would you trust a small on-device model with routine tasks, or do its limits make a cloud assistant worth the privacy trade-off?
Sources and evidence
- Source update (17 October 2026, 19:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.