# Privacy AI: Powerful chatbot — Hub for local and remote AI > AI fiction studio, character roleplay, and multi-model AI runner with extensive tools. (iPhone/iPad app by AcmeUp Inc..) - Source: https://appshunter.io/ios/app/privacy-ai-powerful-chatbot/id6738392421 (this page in markdown: same URL + `.md`) - Developer: [AcmeUp Inc.](https://appshunter.io/developer/1780566149) - Category: Productivity, Developer Tools - Price: Free with in-app purchases - Rating: 4.80/5 from 18 App Store ratings · 13 written reviews indexed - Age rating: 17+ - Requires: iOS 18.0 · 3.0 GB - Languages: American English, French (France), German (Germany), Italian, Japanese, Korean, Brazilian Portuguese, Russian, Chinese (Simplified, China), European Spanish and 1 more - Released: 2025-07-18 - Data updated: 2026-08-09 - Monetization: free with in-app purchases - User reviews in markdown: https://appshunter.io/ios/app/privacy-ai-powerful-chatbot/id6738392421/reviews.md ## What is Privacy AI? NEW: Email Assistant Triage your inbox on device. Connect any IMAP account, tap Process Inbox, and the AI flags what needs a reply, files the rest into folders, and can move junk. Each run writes a dated report you can undo one tap at a time, and your corrections teach it your preferences. Passwords stay in the Keychain, and content is never stored. Privacy AI is a general-purpose AI assistant for iPhone, iPad, and Mac. Run local models fully offline, or connect to the cloud, keeping conversations private. Run Models Locally or in the Cloud - On-device GGUF and MLX models, with vision and tool-calling - Connect to OpenAI, Claude, Gemini, DeepSeek, or any OpenAI-compatible server - Apple on-device Foundation Model (iOS 26+) - Switch local and remote models mid-conversation, keeping context Smart Chat - Set model tiers once (fast, capable, reasoning, local), then just type - Routes each message by complexity and privacy, and keeps sensitive text local System Keyboard - Bring Privacy AI into any app: dictate by voice, or Process to rewrite, translate, and reshape text in place - Build named pipelines (Polish then Translate) or use presets like Proofread and Email - Types in 11 languages with Pinyin, Japanese, and Korean input Assistants and Tools - Custom assistants with instructions, an avatar, and a knowledge base - Knowledge Wiki: a cross-linked knowledge base from your documents, with a graph and cited answers - LLM Council: query several models in parallel and synthesize one answer - 50+ built-in tools: web search, news, calendar, code execution, and more - Full MCP support with a marketplace, Siri, and Shortcuts Writing and Voice - Long-form writing assistant that drafts outlines and chapters - On-device text-to-speech, real-time voice chat, and live transcription Files and Content - Convert PDFs, Office files, EPUB, video, and audio to Markdown, with OCR - Media Studio: generate images with your provider Accessibility - Full VoiceOver coverage, sound and haptic feedback, and Verbose Announcements Free to Use. No Ads. Local model features are free, including the AI keyboard, knowledge wiki, writing assistant, custom assistants, 50+ tools, document reader, and image generation. Cloud API models and the MCP marketplace need a subscription. Privacy by Design Privacy AI runs on-device. No backend servers. No accounts. No data collection. Every API request is visible in the Protocol Inspector. Your models. Your device. Your data. Medical Disclaimer Privacy AI is a general-purpose AI assistant and is not a doctor. It is not a substitute for professional medical advice, diagnosis, or treatment. Any health-related content is provided for informational and educational purposes only. In addition to using this app, always check with a doctor or other qualified healthcare professional before making any medical decisions. Never disregard or delay professional medical advice because of something you read or generated in this app. In case of a medical emergency, call your local emergency number immediately. Terms of Use: https://privacyai.acmeup.com/docs/policy/tos.html Privacy Policy: https://privacyai.acmeup.com/docs/policy/privacy.html ## Key features - AI fiction novel writing studio - AI character roleplay studio - Run local and remote AI models - AI agents and 40+ built-in tools - On-device and cloud voice generation - Convert and export various file types - Full VoiceOver and haptic feedback support ## Pricing and in-app purchases Base price: Free. **Subscriptions** | Subscription | Price | Period | | --- | --- | --- | | Weekly Pro Plan | $3.99 | per 7 days | | Yearly Pro Plan | $99.99 | per year | | Monthly Pro Plan | $9.99 | per month | ## Recent user reviews (10 of 13) All indexed reviews: https://appshunter.io/ios/app/privacy-ai-powerful-chatbot/id6738392421/reviews.md ### 5/5 — Right direction *2026-08-07* **Edited: Dev responded positively and implemented fix right away, 5* Overall, I think this is on the right track, though one obvious problem is the text-to-speech feature... what's called "File TTS" isn't actually text to speech.... it's ASR/STT. That label should be corrected. On top of that, there's already a way to input text and preview how it sounds using the TTS models, so I don't understand why that same option isn't available for other text I'd want converted. Paying an API provider to run some massive model makes no sense to me when my own device has plenty of processing power for the job, and the drop in quality would be perfectly tolerable. **Developer response:** Thanks for the honest review, and you are completely right about that label. That tab is for file transcription, so it is ASR, not text to speech. Calling it "File TTS" was my mistake. It is now "File ASR" and the fix is in the next update. Your second point is a fair hit, and the credit is yours: I had not thought of it. You were right that it made no sense for the local voices to work in the preview box with no way to point them at anything else. The pieces turned out to be mostly there already and just needed wiring together, so I built it. There is now a "Text to Audio" feature that turns any text or document into an audio file using the speech models already on your device, processed in chunks so a long article or a book chapter works fine, and you can resume one you stopped partway. It ships in the next update. I agree with you on the last point. The phone has plenty of power for this job, the local voices sound good enough, and nothing needs to leave the device or cost anything to run. Thanks for taking the time to write this up. ### 5/5 — Hands down one of the best iOS harness/inference clients. *2026-07-01* Top two alongside Minis. You really can’t ask for more when it comes to running AI on iOS, and the development team is constantly pushing newer and newer updates on a regular basis. Just when you think it couldn’t possibly get better, they outdo themselves yet again. Now this is how app development SHOULD be like. Major respect to the team, and thank you. Infinitely appreciated. **Developer response:** Thank you so much for this. This kind of message is what makes the work worth it. It means a great deal to hear that the regular updates are landing well. My goal has always been to make running AI on iOS feel simple and private, and to keep improving it with every release. I have plenty more planned, and reviews like yours are exactly what keeps the momentum going. If you ever run into something you would like to see improved or added, I would love to hear it. Thank you again for the kind words and the support. ### 5/5 — Excellent *2026-06-06* This app is useful to me because it allows me to understand the capability of my device. Privacy AI is a little bit crowded, but that’s only because it lets you do everything under the sun. Because of this app, I know what models I will be using in apps I am building. **Developer response:** Thanks so much for the kind review, and for the five stars. I love that you use Privacy AI to see what your device can really do. That is exactly what I hoped for: a private place to try models hands-on before you pick one for your own apps. Fair point on it feeling crowded. It got that way because I wanted to fit in as much as I could. I am always working to make it easier to get around. If something ever feels buried, just let me know. Thanks again for the writing. ### 5/5 — What’s in a name? *2026-04-30* Using the app so far has been pretty mind blowing. Every other local inference wrapper application has a long way to go before they even touch the capabilities of this thing. It’s nothing shy of ambitious. I just wonder how they devs could put so much thought into the product itself but yet have such a…lackluster name. Anyways, top marks ti the mysterious makers of this masterpiece. **Developer response:** Thanks for the kind words, really glad the capabilities have impressed you! You are complete correct about the app's name. When I chose the name, I just wanted to express the idea that privacy protection is at the core of the app. But after it went live, I realized there are quite a few apps with similar names on the App Store, which makes it hard to find mine. I will admit that was a bad call on my part. Appreciate the top marks regardless. Thank you! ### 2/5 — Very cool app *2026-04-02* Has solely a subscription rather than one time payment in-app as an option and also collects tracking data, any data collection beyond typical LLM API usage makes the name pretty ironic. I get the purpose of the subscription model, and without a one time purchase it segments the potential customers into subscribers and those that will just pass on it entirely. Analytics are enabled by default too, so it is requiring users to opt-out by default. These can change in the future but show the direction to begin with and are often followed by typical dev review replies with attempted justifications. Doesn’t escape the contradictory. Solid UI. Not a lot of justification in contrast, no thanks. **Developer response:** You are right that "Privacy AI" sets a high bar, and I want to address your points honestly rather than talk past them. On analytics: the app has no server and requires no account registration. Your conversations, your AI outputs, and your API keys never leave your device. The only third-party data collection is Google Analytics, which captures aggregate feature usage, things like which features are popular and how often certain screens are visited. It does not touch your conversations or any personal content. I acknowledge the contradiction: even aggregate analytics sits uneasily under a privacy-focused name, and I understand why that bothers you. If you prefer no data is sent at all, you can disable analytics entirely in Settings and nothing will be transmitted. If you want to go further and verify that yourself, the app has built-in proxy support for mitmproxy and Charles. You can route all traffic through either tool and audit exactly what is sent out in plain text. Or if you prefer, you can use any network traffic monitoring software you trust, such as Wireshark, to inspect the same data independently. I am comfortable saying that openly because I have nothing to hide there. On the subscription: A lifetime purchase model would pull my attention toward supporting a growing base of one-time buyers rather than building new features. The subscription keeps me focused on the things that actually make the app better for everyone using it. Thank you for the kind words on the UI. That means a lot. If you have more questions or want to dig into any of this further, feel free to reach out at customercare@acmeup.com, on Discord, or on X at @best_privacy_ai. I am happy to talk through it. ### 5/5 — Powerful *2026-03-19* This App is powerful and professional, just as cherry studio & openclaw. The only shame is that it’s too complicated for beginners. And the local model can’t be deleted. **Developer response:** Thank you so much for the kind words and the five-star review! Regarding the app size, there are two embedded models included. The first is a Qwen3.5-0.8B-INT4 model (without vision), about 500MB. It is bundled to guarantee the app works out of the box. The second is the Kokoro TTS model at around 80MB, which powers the on-device text-to-speech feature. I am considering removing the Kokoro model in a future release to reduce the app size by 80MB. The rest of the size comes from various engine libraries including llama.cpp, MLX, JavaScript, and shell, which provide the tool calls and automation capabilities. I hear you on the complexity for beginners and will keep working on making things more approachable. And deleting local models is something I plan to improve as well. Thanks again for the support! ### 5/5 — Oh my god… *2026-03-07* This app is blowing my mind… This is everything that I’ve ever wanted in a local LLM app…and more. The advanced tools available in this app are kind of insane. I’m sorry but this app puts PocketPal and Locally AI to shame. 1 small BUG: I can’t seem to set the theme to “auto” mode. I’m stuck at either light or dark mode. **Developer response:** Thank you so much! This genuinely made our day! We're thrilled Privacy AI has become your go-to local LLM app, and we'll keep pushing to make the tools even better. Good news on the bug: the "Auto" theme mode fix and a new dark app icon are both coming in next version. The update is already submitted for reviewing. ### 5/5 — Better than I expected *2026-02-11* Good app for asking weird questions no other AI would ever answer. Completely private and after you download the models you need it work even offline. Only thing you have to know is that it uses up a lot of your devices resources. **Developer response:** Thank you so much for the five-star review! It really means a lot. You are absolutely right about the local resources. Local models are large files and they do take up significant storage space on your device. That is the trade-off for having full privacy and the ability to work completely offline. I am planning to add an option to offload models from iCloud Drive to local drive in a future release. This will give you more flexibility in managing your storage. Thanks again for your kind words and support! ### 5/5 — Most viable advanced local LLM client for iPad *2026-01-20* The scaffolding is here for this to become the go-to local LLM app for users who actually know what GGUF and MLX models are. On iOS, day-one support for the latest llama.cpp or mlx-lm updates is basically nonexistent across the ecosystem, so this app has real potential if the developer prioritizes keeping those libraries current. Right now, there still is not a clean way to import GGUF models with VLM capabilities. I have had to work around this by renaming and swapping in my own VLM models in place of the built-in ones, which is functional but clearly not ideal. I would very much like to see mlx-audio and mflux integrated, and more broadly, for the app to give the MLX stack more latitude. If the underlying MLX library is new enough to run a model, the app should largely stay out of the way. Features should not be artificially blocked due to UI assumptions or overly rigid integration when the backend already supports them. With faster library updates and fewer constraints at the UI layer, this could easily become the best serious local AI app on iOS. **Developer response:** Thank you so much for taking the time to write such detailed and thoughtful feedback. Reviews like yours are incredibly valuable to us and genuinely help shape our development priorities. Regarding VLM support for GGUF models, you're absolutely right that this isn't properly finished yet. We've actually dived into this issue, but the challenge is that mmproj files don't have a uniform specification and lack consistent URLs across different models. This makes it difficult to design a proper user facing workflow for customization. Additionally, we haven't received much feedback about VLM support until now, so we deprioritized it. But your feedback is exactly what we need to reconsider that decision. For the MLX stack, I want to make sure I understand your point correctly. When you say "If the underlying MLX library is new enough to run a model, the app should largely stay out of the way", could you clarify what specific limitations or UI assumptions you're experiencing? Are there particular models or features that the MLX backend supports but the app is blocking? Concrete examples would really help us understand what needs to change. Regarding mlx-audio and mflux integration: We haven't allocated time to these yet. For audio, we've already integrated sherpa-onnx for TTS and whispercpp for STT, and we weren't sure if adding mlx-audio would provide significant additional value. For mflux specifically, while it should theoretically be useful on mobile devices, our previous research showed that due to limited GPU and memory on mobile, the generated images were toy-level quality. We weren't sure if it would be valuable enough, but given your feedback, we're willing to give it a second try. If you have specific use cases or workflows where mlx-audio or mflux would be beneficial, we'd love to hear them. Please feel free to reach out to us at support@acmeup.com with more details. This kind of input directly influences our roadmap. We're committed to making this the best serious local AI app on iOS for advanced users like you, and we'll be rolling out improvements soon. Thanks again for your insights! ### 5/5 — Literally Years Ahead of Everything Else *2025-08-08* I've tried just about every AI client on the App Store, and this is the one. I stumbled on this app when it had zero downloads and found an absolute gem. This isn't just another chat wrapper; it's a full-on AI agent for your phone. You can plug in APIs for basically any model (Gemini, DeepSeek, etc.) and it has tools that let the AI actually do things—manage your calendar, get directions, search your contacts, the list goes on. The "Super Siri" feature that lets you use your own AI model with a voice command is a complete game-changer. Now, full disclosure, the app is BRAND new, so it's a little rough around the edges. You might hit a few quirks or a random crash here and there as the developer irons things out. But honestly, that's totally expected for something this ambitious. The potential here is just off the charts. Giving this 5 stars for the vision alone and for what it can already do. **Developer response:** Thank you for trying Privacy AI when it only had a few downloads. It means a lot that you see what we’re building, not just another chat app, but a full AI workspace and agent for mobile devices. We’re glad you tried features like broad API support and the “Super Siri” voice command. Both are designed to put you in control of your AI on your own terms. Our philosophy is simple: AI should run under users’ control, free from Big Tech lock-in, and work exactly the way users need it to. The app is still new and not perfect. But we’re shipping fast and improving every week. Thank you for the five stars and for recognizing the vision. Reviews like yours help more people discover Privacy AI and motivate us to make it even better. ## Frequently asked questions about Privacy AI ### What is Privacy AI's Novel Writer feature? Privacy AI's Novel Writer is a complete AI fiction studio that allows you to write full novels from a single premise. It supports 20 genres, builds structured outlines, and writes chapters with genre-tuned style, ensuring consistent voice and pacing. ### Can I use Privacy AI offline? Yes, Privacy AI supports offline GGUF and MLX inference on iPhone, iPad, and Mac via llama.cpp. All local model features are completely free with no advertisements. ### What AI models can Privacy AI connect to? Privacy AI can connect to OpenAI, Claude, Gemini, DeepSeek, or any OpenAI-compatible server. It also supports Apple's on-device Foundation Model on iOS 26+. ### Does Privacy AI have ads? No, Privacy AI is ad-free. All local model features are completely free with no advertisements. Cloud API models and the MCP Marketplace require a subscription. ### How does Privacy AI handle accessibility? Privacy AI offers full VoiceOver coverage across all screens and includes sound effects and haptic feedback for non-visual confirmation of AI events. A Verbose Announcements mode reads responses aloud on demand. ### What file formats can Privacy AI convert and export? Privacy AI can convert PDFs, Office files, EPUB, YouTube, and audio to clean Markdown. It can export content to Markdown, PDF, HTML, EPUB, or JSON. ### What are the subscription options for Privacy AI? Privacy AI offers a Monthly Access Plan for $9.99/month and a Yearly Access Plan for $99.99/year. These subscriptions are required for cloud API models and the MCP Marketplace. ### How frequently is Privacy AI updated? The latest version of Privacy AI is 1.9.1, and it was last updated on March 25, 2026. This indicates a relatively frequent update schedule. ## Version history (last 5 releases) ### 2.3.0 — 2026-08-04 FEATURES - Email Assistant A new mini app that triages your inbox. Connect any IMAP account with an app password, tap Process Inbox, and the AI reads unread mail, flags what needs a reply, and files the rest into folders under an AI label. Choose to summarize, organize, or also move obvious junk. Each run makes a dated report where every decision undoes with one tap, and your corrections teach your preferences. Passwords stay in the Keychain, content is never stored, and nothing is deleted. Mail can also become calendar events, reminders, and notes. - Name Your Pipeline Steps Give each step in an AI Keyboard pipeline its own name, so it shows your label instead of "Custom". - Reusable Saved Steps Save any named step to a personal library and reuse it across pipelines, each an independent copy. - Video on Local Vision Models MLX vision models can now understand video, not just images. Attach a short clip; set the frame count, rate, and size per model. - TurboQuant for Long Chats An opt-in option that keeps your whole conversation in on-device memory by compressing cached values, so long chats use less memory with a small quality trade-off. - Import MLX Models from ZIP Add your own MLX model from a ZIP on your device, no download needed. Tap Import From Local, pick a zip with the model folder, and it installs. IMPROVEMENTS - Local Engine Update (llama.cpp b10091) Two-bit (Q2_0) models now run on the GPU on Apple Silicon, and Qwen3-VL reads layouts more accurately. Adds Laguna code models and completes DeepSeek V4 support. - MLX Engine Update Major memory savings for vision models: Qwen3.5 VL and Qwen3 VL handle long and multi-image prompts with far less memory, and multi-turn image chats stay consistent. Generation also stops cleanly when the app backgrounds. - Engine-Aware Model Settings Settings now show only the options the model's engine actually uses, each tagged with its engine. MLX models now respect Top-K and Min-P. BUG FIXES - Memory Profile Stays On Topic The memory profile now appears only when memory is on and a sector is selected, and no longer blends unrelated sectors together. - Dismiss The Suggestion Bar The word suggestion bar can now be closed with an X, and no longer covers the pipeline picker. - No More Duplicate Imported Models Fixed a model imported from a local GGUF file reappearing as extra "Recovered from" copies after relaunch. Earlier duplicates are cleaned up. - Local Chat No Longer Fails When Memory Is On Local models with memory enabled no longer show a false error. With both models on-device, the chat keeps running and Apple Intelligence saves memories. - Local Models Remember Your Conversation Small on-device models no longer claim they cannot remember earlier messages. A hidden bookkeeping line fed to the model as an instruction is gone. - Long Chat Setting Stays Put Changing how a local model handles conversations that outgrow its context window now sticks, instead of reverting when set inside a chat. ### 2.2.0 — 2026-07-15 FEATURES - Open Keyboard On Microphone The AI Keyboard can open straight into dictation, ready to record the moment it appears. - Auto-Process After Dictation Stop recording and your pipeline runs automatically, returning the result in place with one-tap Undo. - Dictation Sound Cues Subtle sounds mark when recording starts, when processing begins, and when your text is ready. - Live Dictation Preview Your transcription previews live as you speak and is inserted the moment you stop. - Multi-Language Keyboard Type across all 11 app languages, with Chinese Pinyin and Japanese candidates, Korean Hangul, and European accents. - Inline AI Instructions Tap AI and type an instruction like "intro to Beijing"; it is filled in place while the rest of your text stays put. - Memory For The AI Keyboard Attach a memory sector so Process and inline AI use your identity, preferences, and style. Read on device, off until you opt in. - Knowledge Wiki Turn your documents into a personal, cross-linked knowledge base the AI builds and maintains, stored on device as plain Markdown. - Add Sources From Anywhere Build a wiki from PDF, Word, EPUB, web pages, Markdown, text, CSV, photos, videos, and links. - Ask Your Knowledge Base Ask in plain language and get an answer from your sources, linked to the exact pages. Choose Quick or Thorough. - Interactive Knowledge Graph See how everything connects in a live map: pinch, drag, and tap any point to open its page. - Automatic Overview and Insights Each wiki keeps a living Overview the AI rewrites as you add material, plus AI Graph Insights into the biggest themes. - Wikis In Your Language Pick a language and the AI writes every page and answer in it, across all supported app languages. - Git Sync For Your Wiki Back up and sync any wiki with GitHub or another Git service: publish, pull, clone, and restore full history. - Wiki Health Check A one-tap check finds pages that link nowhere, broken links, and missing pages, with an optional deeper AI review. IMPROVEMENTS - One Tap To Polish After Dictation After you dictate, VoiceOver focus moves to the Process button and reads it aloud, so polishing is one tap. - Keyboard Steps Keep Your Language Built-in steps like Rewrite, Casual, and Summarize now stay in your text's language instead of switching to English. - Rewrite Your Full Selection Process now rewrites the exact text you select, or Select All for the whole field, handling longer text reliably. - Local Engine Update (llama.cpp b9993) New model families including Hunyuan v3, DeepSeek V4, MiniCPM 5, and Unlimited-OCR, faster speculative decoding, and stronger model-file security. - MLX Engine Update (mlx-swift-lm) Gemma 3 prompts up to 2.6x faster on Apple Silicon, more reliable Qwen3/Qwen3.5/Qwen2.5 VL vision, plus Mixtral and Mamba2. BUG FIXES - Keeps Your Audio In High Quality While Dictating Keyboard recording no longer drops Bluetooth, AirPods, or the speaker to phone-call quality. ### 2.1.2 — 2026-06-29 FEATURES - AI Keyboard A new system-wide keyboard brings Privacy AI into any app where you type. Turn it on once, then in Mail, Messages, Notes, or anywhere else, tap the mic to dictate by voice, or tap Process to rewrite, polish, translate, or reshape what you have written, replaced in place with one-tap Undo. Build your own pipelines (ordered steps such as Polish then Translate) or start from ready-made presets like Proofread, Email, and Summary. Everything runs through the app on your device, so your text stays private and never leaves your phone. - Image Prompt Wizard Building a good image prompt is now much easier. Tap Prompt Wizard on the image generation screen to browse a curated library of prompt templates by their sample picture, pick one, and fill in its options. You can also type a single subject and have AI fill in every field for you in one tap, then preview the finished prompt before you generate. Each template credits its original author. IMPROVEMENTS - Local Engine Update (llama.cpp b9754) The on-device inference engine was updated to the latest build for faster and steadier local models. Multimodal models process attached images more quickly, large models now show real progress while they load, and more models support speculative decoding for quicker replies. Updated graphics and audio components improve performance and stability across the board. - MLX Engine Update (mlx-swift) The Apple MLX inference engine was refreshed for lower memory use and better stability. Vision models now read long prompts in smaller pieces, sharply cutting peak memory so large image chats run more reliably on device. Speculative decoding automatically steps aside when memory is tight, so it never slows you down. This update also adds the Gemma 4 12B Unified and Nemotron text-diffusion models and fixes loading of Qwen3.5, Qwen3-Next, and nomic-embed models. ### 2.1.1 — 2026-06-19 FEATURES - Smart Chat Automatic Model Routing Smart Chat picks the best AI model for every message automatically, so you never have to choose. Set up your model tiers once (a fast model for simple questions, a capable model for complex work, a reasoning model for hard problems, and a local model for anything private), then just type. Smart Chat reads each message for complexity, privacy sensitivity, and content type and routes it, with a badge on each reply showing which model answered and why. Personal or sensitive messages stay on a local on-device model, you can set per-setup spending budgets, and you can override any single choice. - Video Input For Local Vision Models Attach a video to a chat with a local vision model (such as Qwen-VL or Gemma) and it describes what happens across the clip. The app samples evenly spaced frames and feeds them to the on-device model, so it sees motion and multiple moments rather than a single still. This matches cloud models, fully on device and private. IMPROVEMENTS - Faster Sharing To Chat Sharing a file such as a video now opens the chat right away and prepares the file inside it, with a progress overlay, before sending automatically. No more waiting on a loading screen. If the file cannot be processed, the chat still opens and explains what went wrong. Works in regular chats and Smart Chats. - Resume Interrupted Model Downloads Large model downloads now pick up where they left off instead of restarting from zero. If your connection drops, you cancel and restart, or you leave the app and return, the download continues from where it reached, saving time and data on multi-gigabyte files. Applies to all model types. - Resend Keeps Both Replies Resending now keeps the original AI reply and adds the new one as a variant instead of replacing it. An inline picker (such as 2 of 3) appears on the latest reply so you can flip between versions and pick one. Switching a variant also updates what the assistant carries forward. Works in main chat and Smart Chats. - Local Engine Update (llama.cpp b9663) The on-device inference engine was updated to the latest build, adding video understanding for multimodal models, faster image processing when several pictures are attached, new model architectures, and audio and graphics fixes. - MLX Engine Update (Gemma 4 and Vision) The Apple Silicon MLX engine was updated for faster, steadier on-device models. Gemma 4 gains optional speculative decoding for quicker replies, and its image understanding uses less memory so long prompts with pictures no longer fail on iPhone and iPad. Qwen vision uses less memory too, and Gemma 4, Falcon H1, and LFM2 are more accurate. BUG FIXES - MiniMax-M3 Image and Video Input Fixed multimodal input for MiniMax-M3. Images no longer fail with an "invalid image detail" error, and attached videos are now sent in a format the model recognizes, so it describes the video instead of ignoring it. Videos over 50 MB show a note suggesting the convert-to-text option. ### 2.0.28 — 2026-06-12 FEATURES - Sign in with ChatGPT Use your ChatGPT Plus or Pro subscription instead of an OpenAI API key to route Codex models. xAI Grok OAuth also ships. - Live Performance Stats In Chats While streaming, a line shows elapsed time, decode speed, and context usage; an expandable row adds time to first token and token counts. - Exa Web Search Provider Exa joins Tavily, Serper, Brave, and LangSearch: neural search returning full page text, 1,000 free searches per month. - User Profile From Your Memories Each memory sector's Profile button asks an AI model to summarize its long-term memories into a short paragraph that prefixes every chat there. - One-Tap Memory Consolidation A Consolidate button finds near-duplicate long-term memories and merges each cluster into one entry. - Memory Maintenance In One Tap Profile and Consolidate first categorize Uncategorized memories and promote short-term notes to long-term. - Memory Audit History A new Audit History view lists every add, update, and delete with keyword search. - On-Device Model Benchmark A Benchmark button runs a llama-bench style test reporting speed and peak memory, then suggests optimal context, batch, and thread settings. IMPROVEMENTS - Text Selection In Fullscreen Chat Selecting text in the fullscreen chat input no longer slides the main menu open, so editing on iPad stays precise. - Choose Transcription Language When converting video or audio to text, pick the spoken language for every speech engine, or leave it on Auto. - Smoother Transcription Progress Audio-to-text progress now advances steadily instead of jumping backward. - Smarter Duplicate Detection Multiple memories from one turn are evaluated in a single AI call, and near-identical entries are skipped. - Memory Search Knows Profile From Facts The cached user profile is sent separately from matching memory rows for a cleaner prompt. - llama.cpp Engine b9553 Updated from b9279 to b9553 (269 changes), adding DeepSeek-OCR 2, Granite 4 Vision, Gemma 4 vision and audio, and DeepSeek V3.2 sparse attention. - MLX-Swift LM Engine tag-20260607 Faster, more accurate on-device vision: corrected Qwen2.5-VL attention, SmolVLM2 up to 9x faster, plus audio input and more reliable tool calling. - Sharper Camera Focus for Documents View Assistant locks focus more reliably on close-up subjects, and you can tap the live preview to focus. BUG FIXES - Soul Greeting Now Appears From the Model View Starting a chat with a Soul from model settings now shows its configured first message. - Attaching a Video to Chat Fixed Video Processing Error: file doesn't exist when attaching a video. - Empty Transcript From iOS Speech Fixed the iOS speech recognizer returning nothing on a clear recording. - Convert to Text Could Hang Fixed the iOS Speech Analyzer getting stuck at Analyzing audio; it now falls back to the system recognizer. - Legacy Memories Now Reachable Memories with an Uncategorized chip were invisible to Consolidate and Profile; all maintenance flows now pick them up. - Memory AI Calls No Longer Stall On First Event Fixed Profile, Consolidate, and Categorize calls returning empty when the first AI event was misread as the end. - Memory List Refreshes After Maintenance After Profile or Consolidate, Uncategorized chips on just-categorized rows disappear immediately. - VoiceOver Input Stays Reachable After Sending Fixed the chat input collapsing after sending; with VoiceOver it stays visible and focusable. ## Apps similar to Privacy AI | App | Rating | Price | Category | | --- | --- | --- | --- | | [Pal Chat - AI Chat Client](https://appshunter.io/ios/app/pal-chat-ai-chat-client/id6447545085) | 4.3 (416) | Free | Productivity | | [LatentChat - Assistant LLM](https://appshunter.io/ios/app/latentchat-assistant-llm/id6733216453) | 3.8 (28) | $4.99 | Productivity | | [LLM Pigeon](https://appshunter.io/ios/app/llm-pigeon/id6746935952) | 3.8 (4) | Free | Productivity | | [Local AI Studio](https://appshunter.io/ios/app/local-ai-studio/id6587560115) | — | $9.99 | Productivity | | [LLMConnect • AI Chat](https://appshunter.io/ios/app/llmconnect-ai-chat/id6737247302) | 3.5 (4) | Free | Productivity | | [AlevioOS - Local Ai](https://appshunter.io/ios/app/alevioos-local-ai/id6749600251) | 3.0 (7) | Free | Productivity | ## Related topics [privacy ai](https://appshunter.io/ios/topics/privacy-ai) --- *Data collected daily from the US App Store and indexed by [AppsHunter](https://appshunter.io/). User reviews are verbatim App Store reviews. Ratings, prices and chart positions refresh continuously; this snapshot is from 2026-08-09.*