OneLLM vs Local LLM

Local AI. Online models. Your computer, within reach. OneLLM brings everyday AI and your computer’s agents into one workspace on iPhone and iPad. CHOOSE HOW YOU CHAT Download a compatible model for on-device, offline chat, or use your own credentials with supported online providers. Connect to Ollama or OpenClaw on a reachable server. Different tasks call for different models. BIGGER LOCAL MODELS An optional 35-billion-parameter MoE package uses expert streaming with about 3B active parameters. Smaller text and vision models are also available. Model capability, speed and memory needs vary by device. YOUR ONELLM OFFICE, ON YOUR PHONE Pair with a compatible OneLLM Office computer using its connection code. Continue a conversation, choose an available local or online model, follow task progress and respond to approvals. Open task files from a separate Files view. The desktop preview must remain running and reachable; desktop release availability may vary. OPENCODE MOBILE CONNECTOR Connect to OpenCode on your computer from iOS. Continue sessions, choose models and agents, browse files, review changes and respond to permissions. Scan the matching connection code or enter the service address. DEEPSEEK HARNESS CONNECTION Keep DS Harness conversations and workspaces close. Send messages and supported attachments, follow agent activity and continue from your phone. Use the DS Harness connection code in its matching module. PHOTOS, FILES AND VOICE See image previews in chat, check attachments before sending and remove them with a tap. English voice input uses bundled local transcription. Notes and on-device text recognition are also included. Local models require a separate download, storage and compatible hardware. The optional 35B package needs about 16.6 GB of storage; 12 GB RAM is recommended. Remote connections require a running computer and a reachable trusted LAN, Tailscale network or supported secure service address. Computer services use the model accounts configured there. Direct online chat requires your own provider credentials; provider subscriptions and API charges are not included. Review AI output before using it.

MITHRIL is a private AI assistant that runs on your iPhone and iPad. Your conversations are processed and saved on your device, or on your own Mac if you pair one. No account, no cloud, no tracking. TWO KINDS OF MODELS • Apple's on-device model, built into iOS 26 on devices with Apple Intelligence. Nothing to download. • Open models from Hugging Face in GGUF format, from companies like Alibaba, Google, Meta, Microsoft, Mistral, DeepSeek and Liquid AI. Browse by company, see whether a model fits your device's memory before you download it, and switch any time. AN ASSISTANT THAT CAN DO THINGS It knows today's date, the time and your city from your device, and asks before it acts. Texts and emails open as drafts for you to review and send. Choose which abilities it can use: • Reminders and calendar events • Texts, email and phone calls • Look up your contacts • Directions and places nearby • Run your Shortcuts • Web search: Wikipedia, DuckDuckGo, or Brave with your own key • Search your own documents • Today's date and exact math PHOTOS, YOUR PHONE AND YOUR MAC • Ask about a photo: attach one, take one or paste a screenshot, and models that read images answer on the device. • Search my phone: ask about the text in your photos, screenshots and the folders you pick, indexed on the device. • Use your Mac: pair your own Apple silicon Mac and chat with the bigger models on it over your Wi-Fi, encrypted between your devices. • Run your coding agents or a terminal on your Mac from here. PRIVACY YOU CAN SEE The Privacy map shows what runs on your device, what goes online, and exactly what is sent. A web search sends only the search words, never your conversation, and your location is only turned into a city name. A live list shows every network request MITHRIL makes. VOICE Dictate with Apple's speech recognition, or download NVIDIA Parakeet for more accurate on-device transcription. Hear answers in natural voices generated on the device. The floating assistant stays with you in Picture in Picture while you use other apps. FOR THE CURIOUS Tap the speed line under an answer for a live engine dashboard: tokens per second, time to first token, context use, cache hits and memory. REQUIREMENTS iOS 26. Apple's on-device model needs a device with Apple Intelligence. Open models run on any supported iPhone or iPad. Larger models need more memory, and MITHRIL shows what fits before you download. Models are made by their publishers and downloaded from Hugging Face. MITHRIL doesn't train or host models.