Ari Helper vs SoloLLM

Introducing Ari - Your Easy-to-Use AI Assistant! Meet Ari, your personal AI assistant, right on your device! Ari runs its models on the device itself, so it works entirely offline — no internet connection required. Your privacy is our priority – Ari doesn't collect any of your personal data. Start Ari anywhere with Siri and App Intents — just say "Ask Ari," speak your question, and hear the answer back. Power users get a full Shortcuts catalog: Ask Ari with any model (Apple Intelligence, local, or cloud), start chats, generate and edit images, analyze PDFs, speak replies, change settings, queue Code Session tasks on Mac, and dispatch prompts to other devices — all chainable in Shortcuts. Ari is on your wrist too: the Apple Watch app remote-controls Ari on iPhone. Dictate from the "Ask Ari" complication or Smart Stack, watch answers stream in, hear them spoken, and hand long replies to iPhone. Questions queue when the phone is away; private chats stay off the watch. Conversations show up in Spotlight — including by what's in their images. Ari reads text and objects in chat pictures on-device so you can search for "that screenshot with the receipt". Private media is never indexed. With Sensitive Content Warning on, flagged images blur behind "Show Anyway" (verified adults can opt out). More to love: Text-to-Speech (device voice or Ava); long-press to copy or retry across models; syntax highlighting with one-tap code copy; Previous Chats synced via iCloud. On Apple Intelligence (iOS 26 / macOS 26 and later), Ari uses the on-device foundation model by default — private, no API key, no network. Attach images: on iOS 27 and macOS 27 the model sees them directly; on OS 26 Ari describes them on-device with the Vision framework (text, tables, barcodes, subjects). Prefer full vision models? Pick one from the menu. Download on-device models with a tap — Google's Gemma 4 for chat, Stable Diffusion converted for Core ML for pictures — or import a model folder of your own. Self-host with Ollama or any OpenAI-compatible server (LM Studio, vLLM, llama.cpp, LiteLLM), or generate images with Draw Things over gRPC. New in Ari 7: Ari makes pictures and films, and repairs the ones you already have. Films is a studio — write a film as a list of shots, render them one at a time, keep the ones that work, and watch the cut come together in a player above the timeline. Pictures get a canvas editor: brush over a crease and heal it, paint a leash or a stranger out of the frame, or treat the whole photograph — repair, denoise, sharpen camera shake, colorize, upscale four times. Every restoration model runs on your device and downloads on first use rather than shipping in the app. Point the camera at something and ask about it. Look at Something sends the frame to whichever model you have selected — on-device, self-hosted or cloud — captions the answer under the picture and speaks it if you like. Talk to it and it listens, answers, then listens again. It sits beside the microphone, and goes on the Lock Screen, Control Center, the Action Button or Siri. Cloud providers with your own keys: OpenAI (GPT-5 family and GPT Image), xAI Grok (4.5 plus Grok Imagine), Anthropic Claude, Google Gemini (plus Imagen and Veo), and AWS Bedrock. Image-capable models accept photos; import text and source files. Cloud use reminds you each provider's privacy policy applies. On Mac, open a Code Session on a project folder with Grok for agentic tools — read, edit, search, git, shell, browser, web, and MCP — so Ari's GUI agent matches a terminal coding agent. Note: For strongest privacy with iCloud sync, enable Advanced Data Protection. Without it, conversations stay limited to your iCloud account but are not fully end-to-end encrypted. Optionally powered by Gemma 4 (HuggingFace) for offline chat and image understanding. Download Ari and keep your AI assistant at your fingertips!

SoloLLM runs powerful AI models entirely on your iPhone and iPad. No cloud. No API keys. No subscriptions. Just fast, private AI that works even without the internet. Your conversations never leave your device - because they never have to. IT DOESN'T JUST CHAT. IT ACTS. SoloLLM gives the AI 15 built-in tools to get real work done - reading PDFs, searching the web, running code, extracting text from photos, translating between languages, writing files, and more. All on-device. All private. Available tools:
 • PDF Reader: extract and analyze content from any PDF
 • Document Reader: read text files and documents
 • File Writer: save AI output directly to your Files app
 • File Extraction: pull text from images and documents
 • Web Fetcher: fetch and summarize any webpage
 • Web Search: search the web without leaving the app
 • Code Interpreter: run JavaScript on-device
 • Data Analyzer: process and visualize structured data
 • Image Analyzer: analyze and describe image content
 • OCR: extract text from photos using Apple Vision
 • Clipboard: read from and write to your clipboard
 • Translation: translate between dozens of languages using Apple Translate - fully offline
 • Siri Bridge: trigger voice-driven workflows through Siri
 • MCP Tools: connect external services via Model Context Protocol CHOOSE YOUR MODEL SoloLLM ships with 13 models ready to download: Gemma 4, Llama 3.1 and 3.2, Qwen 3, Mistral 7B, DeepSeek R1, Phi-4, and more. Use a small, fast model for quick tasks or a larger one when you need more depth. Switch between models at any time. TRULY PRIVATE SoloLLM uses llama.cpp - the engine behind leading open-source AI tools - to run models entirely on your device. No account required. No data collected. No one reads your conversations. What you ask stays on your phone. DESIGNED FOR iPHONE AND iPAD Native experience on both. iPad gets a full sidebar with split-view navigation. iPhone keeps the familiar tab bar. Both work offline. ONE-TIME PURCHASE Pay once and own it. No monthly fees, no usage limits, no subscriptions - ever. REQUIREMENTS Requires at least 4GB of RAM:
 • iPhone 12 or later
 • iPad mini (6th generation) or later
 • iPad Air (M1) or later
 • iPad Pro (M1) or later Performance depends on your model choice and available memory.