Guides and explainers on running an AI assistant privately on the iPhone: what the words mean, how to check claims yourself, and what to expect.
What Orvena 4B was trained to do, how it scored on 547 phone tasks in 30 languages, how fast it answers on an iPhone 17 Pro Max, how to switch to it, and where the open weights are.
Say "wait" and it stops, say "mm-hm" and it carries on, ask something new and it answers that instead. What each phrase does, the replies that skip the model, and how often it gets it right.
Siri AI's chat app is a beta in six languages, not on EU iPhones, but other apps can use Apple's on-device model. What you need, where to pick it in Orvena, its tools, offline use, languages and limits.
Orvena connects the iPhone to nine listed MCP services and to MCP servers you add by HTTPS address. Setup steps, URL rules, the approvals you will see, how tools fit Apple Intelligence's window, and fixes for common errors.
Apple says the DMA keeps Siri AI off iPhones in the EU, with no timeline. What EU users miss, which Apple Intelligence features iOS 27 still brings, and how to use Apple's models with your phone's tools today.
The exact Siri phrases for Ask Orvena and Talk to Orvena, which engine answers in each case, when Orvena opens instead, the Shortcuts action with three recipes, and why iOS won't let you replace Siri.
What Apple promises about Private Cloud Compute, in its own words: the five requirements, what researchers can check, the expansion to Google Cloud, the daily limit and iCloud+, and how apps such as Orvena use it.
Hearing, deciding and speaking, all on the phone. How to start a session from the Action Button, how a turn works, what runs offline, and the limits, starting with the wake word.
A side-by-side table with the numbers: where your words go, offline use, model size, conversation memory, monthly cost and hardware. What each does well, and a way to decide.
A feature-by-feature comparison of ChatGPT and an on-device assistant: price, account, offline use, phone actions, voice, memory. What changes, what you give up, and a five-minute test.
Which iPhones have the memory, how much storage it takes, a five-step setup, the full list of what works with no connection, and a table of what still needs one and where it goes.
The exact Qwen variant Orvena offers, why four bits, what the MLX runtime caches, the 16k context window, thinking budgets, tool calling, and how it compares with running Qwen on a Mac.
The full list of 31 spoken languages, how to set the recognition language, the voices you can choose, the two downloads, how the voices are made, and notes for Norwegian, Chinese and Arabic.
Apple documents which Siri requests stay on the phone and which reach its servers, under what identifier and for how long. How to check your own settings, and how a no-server assistant compares.
Private means a policy, a setting or an architecture. Seven checks you can run on your own iPhone in minutes, including the App Privacy Report built into iOS, and what Orvena shows on each.