AI Built for Audited Work

No software can make you compliant on its own — compliance depends on how your organization uses it, and there is no official HIPAA certification for software. SafeRag is not certified; what it gives you are technical safeguards that support your HIPAA and GDPR work: every AI model runs on your Mac, everything SafeRag stores is kept in an AES-256 encrypted disk image (chat messages and backups are also encrypted with AES-256-GCM), admin and user accounts control access, an audit log records key activity, and admins can export or erase data. Your documents and prompts are never sent to us, to any third party, or to a cloud AI model. The app goes online only to download the AI models you choose, check daily for SafeRag updates, activate and periodically re-check your license with LemonSqueezy, and load the Feature Requests window while you have it open. There is no crash reporting and no analytics.

We do not sign BAAs. SafeRag never sends your documents, prompts or chats to us or to any cloud AI service, so we never handle PHI on your behalf. Whether you need a BAA for your setup is a question for your compliance advisor.

SafeRag's audit log records sign-ins, sessions, admin actions, data exports, backups, and document uploads, queries and deletions — documents are logged by id and file type, never by file name or content. Admins can export accounts, chat history and the document list, and the compliance dashboard shows the status of each safeguard. The log is kept in SafeRag's encrypted disk image on your Mac until you delete it.

Yes. Once your models are downloaded, chat, document search and OCR run without an internet connection. A Pro license needs internet to activate and must be re-checked online at least every 30 days.

Cloud AI services like ChatGPT and Claude process your data on remote servers, creating compliance risk for regulated industries. SafeRag runs every model on your Mac in a bundled Ollama runtime with Ollama Cloud switched off, and it refuses cloud models. Your documents and chats never leave your device, and no third party processes them.

Upload documents and SafeRag creates local vector embeddings. When you ask questions, the app finds relevant document sections and includes them as context for AI responses.

SafeRag offers Qwen3.5 (4B, 9B), Gemma 4 (E4B, 12B), Phi-4, Mistral Small 3.2, Qwen3 30B-A3B, Granite 4.0 and DeepSeek-R1 8B, plus other local Ollama models. On macOS 26 or later you can also use Apple Intelligence. Cloud models are refused. You can download and switch between models within the app.

No. SafeRag bundles its own Ollama runtime, so there is nothing else to install. The app guides you through downloading your first AI model.

Yes. With Pro, Multi-User mode supports up to 5 accounts with admin and user roles. Each user's chat sessions and uploaded documents are private to them; documents from watch folders (set up by an admin) are shared with every account on that Mac. Perfect for shared workstations in clinics, law firms, or enterprise environments.

SafeRag requires macOS 14.0 or later and an Apple Silicon Mac (M1 or later) with at least 8GB RAM. Intel Macs are not supported.