Privacy
SoloLLM runs AI on your device. There is no account to create and nothing is ever sent to a SoloLLM server. The features below that do use the internet are opt-in or explicit, and the app asks before the first use. Get SoloLLM →
With a model downloaded, chat, OCR, translation, file tools and code execution work without any internet connection. On-device inference means your prompts are processed on your iPhone or iPad. Read the FAQ →
AI processing happens on your device unless you turn on Cloud Mode. SoloLLM never sends anything to a server operated by SoloLLM. The features below are the only ones that reach the internet; each is something you tap, something you switch on, or something the app asks you about before first use.
What is sent: A file download request to huggingface.co for the model you selected.
What is NOT sent: No personal data, no device identifiers, no conversation history.
When: Only when you explicitly tap "Download" on a model. The app asks for your consent before the first download.
Third party: Hugging Face, Inc. (privacy policy).
What is sent: Your conversations and settings sync to your own private iCloud account through Apple CloudKit, so they're available on your other devices.
What is NOT sent: Nothing goes to SoloLLM — the data stays inside your Apple account.
When: On by default. Turn it off anytime in Settings › iCloud Sync.
Third party: Apple iCloud (privacy policy).
What is sent: When the AI uses the Web Search tool, your search query is sent to DuckDuckGo. When the AI fetches a web page, a request is sent to that specific website.
What is NOT sent: No conversation history, no files, no personal identifiers.
When: Only when the AI uses a web tool during a conversation. The app asks for your consent before the first use.
Third party: DuckDuckGo, Inc. (privacy policy) and the website being fetched.
What is sent: Your messages, the conversation history, your system prompt, saved profile facts and — in Agent Mode — tool results, to the AI provider you choose, using your own API key.
What is NOT sent: Nothing until you turn on Cloud Mode, add your own key and allow that provider. SoloLLM asks before your first chat with each provider.
When: Opt-in and off by default. You can withdraw consent for a provider anytime in Settings › Privacy & Data › Cloud Mode Data Sharing.
Third party: The provider you choose — OpenAI, Anthropic, Google Gemini, Groq, OpenRouter, MiniMax, or a server you configure yourself. That provider's privacy policy applies.
What is sent: Whatever the AI sends to Model Context Protocol servers and custom HTTP tools you add — which can include text from your conversation.
What is NOT sent: Nothing to any server or tool you didn't add yourself.
When: Opt-in. Only servers and tools you explicitly add; tools that change things ask for your approval first.
Third party: Whatever endpoint you configure.
What is sent: In SoloLLM 1.7.3 and earlier, a reply containing math loads the MathJax renderer from cdn.jsdelivr.net — a standard request for the script file.
What is NOT sent: No conversation content and no personal data. Newer versions bundle MathJax inside the app, so no request is made.
When: Only when a reply contains math, in versions up to 1.7.3.
Third party: jsDelivr (privacy policy).
SoloLLM has no analytics, telemetry or tracking, and contains no third-party tracking or advertising frameworks. Versions up to 1.7.3 showed an opt-in "performance diagnostics" switch, but the diagnostics service it pointed to was never set up, so nothing was ever received; newer versions remove the switch. Crash logs are recorded on-device by Apple's MetricKit and are never uploaded by SoloLLM.
Questions about privacy: SoloLLM support. Last updated October 9, 2026.