On-Device AI Intelligence — Private, Offline, Zero Data Sent to Cloud

Every major AI assistant today, from ChatGPT to Google Gemini to Microsoft Copilot, works the same way: your text is sent to a remote server, processed by a large language model running on corporate infrastructure, and the result is sent back to you. This means that every question you ask, every message you want translated, every piece of text you want corrected is read, processed, and potentially stored by a third-party company. For anyone who takes communication privacy seriously, this is an unacceptable trade-off. OSHI takes a fundamentally different approach: all AI features run entirely on your device, powered by Apple Foundation Models, with zero data ever leaving your phone.

The Problem with Cloud AI

Cloud-based AI services present three fundamental privacy problems that cannot be solved by terms of service or privacy policies.

First, data transmission. When you use ChatGPT, Claude, Gemini, or any cloud AI service within a messaging app, the text you submit must travel across the internet to reach the AI provider servers. This transmission can be intercepted, logged by network intermediaries, or captured by government surveillance systems. Even with TLS encryption protecting the transport layer, the AI provider itself receives your data in plaintext. If you are translating a sensitive message or asking the AI about a confidential topic, that information is now in the hands of a third party.

Second, data storage and training. Most cloud AI providers reserve the right to use your inputs to improve their models. This means your private messages, questions, and text corrections may be stored indefinitely and used as training data. Even providers who claim not to train on user data typically maintain server logs, usage analytics, and content moderation records. The data you submit is never truly under your control once it leaves your device.

Third, metadata exposure. Even if an AI provider encrypts your data and never trains on it, the very act of sending a request reveals metadata: the timestamp, your IP address, the length and type of your query, the language pair for translations, and patterns of usage. This metadata can be subpoenaed, hacked, or analyzed to build a profile of your communication patterns and interests.

OSHI On-Device Approach

OSHI completely eliminates these risks by running all AI processing locally on your iPhone or iPad. The AI model lives on your device, processes your text on your device, and returns results on your device. At no point during any AI operation does any data leave your phone. There is no network request, no API call, no server-side processing, and no telemetry. The AI features work identically whether you are connected to the internet or completely offline.

This is made possible by Apple Foundation Models, a family of highly optimized language models designed specifically to run on Apple Neural Engine hardware. These models are compact enough to run on-device while being powerful enough to perform sophisticated natural language tasks including translation, text correction, tone adjustment, and conversational AI. Apple has invested billions of dollars in Neural Engine hardware specifically to enable this kind of local AI processing, and OSHI leverages this investment fully.

AI Features

Privacy Guarantee

OSHI privacy guarantee for AI features is not based on a promise or a privacy policy. It is based on verifiable technical architecture. Here is why you can trust it:

• Zero network calls: The AI features make absolutely no network requests. You can verify this yourself by enabling airplane mode and confirming that all AI features work identically. There is no degraded mode, no fallback to cloud processing, and no "phone home" behavior.
• No telemetry: OSHI does not collect any usage analytics, crash reports, or feature usage statistics related to AI processing. There is no tracking of which AI features you use, how often you use them, or what text you process.
• Open source: OSHI source code is publicly available on GitHub under the MIT license. Anyone can audit the codebase to verify that the AI features do not contain any network communication code, data exfiltration mechanisms, or hidden telemetry. What you see in the source code is exactly what runs on your device.
• Apple Foundation Models: The underlying AI models are provided by Apple as part of the iOS operating system. They run within Apple privacy sandbox and are subject to Apple own strict privacy guarantees. Apple has publicly committed that Foundation Models process data only on-device.

Why This Matters

The importance of on-device AI extends far beyond abstract privacy concerns. For specific groups of users, it is a critical safety feature.

Activists and Dissidents

An activist in an authoritarian country who uses a cloud AI service to translate a message about a planned protest has just sent the content of that message to a foreign corporation servers, where it may be accessible to government requests, court orders, or data breaches. With OSHI on-device AI, the translation happens locally. The message content never exists anywhere except on the activist own device.

Journalists Protecting Sources

A journalist communicating with a confidential source needs AI assistance to translate a message from a foreign language. Using a cloud translation service like Google Translate means the sensitive communication is processed on Google servers and associated with the journalist IP address and account. OSHI on-device translation ensures that the source communication remains completely confidential.

Business Users with Trade Secrets

Corporate communications frequently contain proprietary information, strategic plans, financial data, and trade secrets. Using cloud AI to process these messages creates data exposure risks that many companies compliance policies explicitly prohibit. OSHI on-device AI allows business users to leverage AI capabilities without any risk of data leakage to third-party servers.

Medical and Legal Professionals

Doctors, lawyers, and other professionals bound by confidentiality obligations cannot risk exposing client or patient information to cloud AI services. OSHI on-device processing ensures that privileged communications remain confidential and compliant with professional ethics requirements.