Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM/MemoryAI Review: Free Offline AI Chat for iPhone and Android
Mobile & Edge LLMs

MemoryAI Review: Free Offline AI Chat for iPhone and Android

·8 min read·By Hans Kuepper · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI
Read in:
🇺🇸en🇩🇪de🇫🇷fr🇯🇵ja🇨🇳zh🇪🇸es🇧🇷pt🇸🇦ar🇰🇷ko

Founder-reviewed. MemoryAI's founder reviewed this entry's technical specs and description for accuracy. The review itself remains independent PromptQuorum editorial content.

MemoryAI (version 1.0.56 on iOS and Android as of 23 October 2026) pairs an editable on-device memory with a PIN-locked Kid Mode in a free chat app that needs no account. Models are downloaded once from Hugging Face and then run offline. Usage events and crash reports without chat content do leave the phone, so read the privacy section before relying on the "no tracking" wording.

MemoryAI is an offline chat app for iPhone and Android from Alessandro Facchini (trading as Facchinits), an independent developer in Italy. It runs open-weight language models on the phone through two engines, llama.cpp and Google's LiteRT, and was first released on the App Store on 30 March 2026. This review reflects its store listings, website and privacy policy as checked on 23 October 2026.

MemoryAI Review: Free Offline AI Chat for iPhone and Android

Key Takeaways

  • Developer: Alessandro Facchini, a single developer publishing as Facchinits, with no company behind it beyond that and no public source code.
  • Maintenance: the App Store shows a release on 22 October 2026 and Google Play an update on 23 October 2026, so no stale-project warning applies.
  • Reach: Google Play shows 1K+ downloads and the App Store shows no ratings yet, so the app has little public track record.
  • Platforms: iPhone (iOS 16.4 or later) and Android; the App Store also lists Macs with Apple M1 or later, marked "not verified for macOS".
  • Languages: the app is available in English, Italian, German, French, Spanish and Portuguese.

📍 In One Sentence

MemoryAI is a closed-source iPhone and Android app that keeps an editable Agent Memory on the device and adds document questions, hands-free voice and parental controls.

💬 In Plain Terms

You install it, download a model once, and then chat, talk or ask about your own PDFs without an account; the optional memory it keeps about you lives on the phone, where you can read and delete it.

📌Note: This review is based on the App Store and Google Play listings, the developer's website and privacy policy, and the public feedback repository, checked on 23 October 2026. PromptQuorum has not installed, tested or benchmarked the app.

What Is MemoryAI?

MemoryAI is a mobile chat app that runs open-weight language models locally instead of calling a cloud service. According to its listings, no model is bundled: you pick one from the in-app list or import any compatible GGUF file from Hugging Face, and the first run needs a network connection to download it. Two inference engines are offered, llama.cpp for GGUF models and Google's LiteRT (MediaPipe) for models in LiteRT format. The developer says LiteRT is not limited to Gemma: it also runs Qwen 2.5 1.5B, DeepSeek R1 Distill 1.5B and Phi-4 Mini.

The App Store listing is titled "MemoryAI: Offline Private AI" in the US, Japanese and Korean storefronts, "MemoryAI: Private Offline-KI" in Germany and "MemoryAI: a IA que lembra" in Brazil; the Google Play package is com.ts.memoryai. Its "Agent Memory" is a feature of this app, not a separate product.

Get It

MemoryAI is installed from the two mobile stores; no desktop build, source code or sideloaded file is offered. Both store links below are the ones the developer's site and listings use.

Platform
Where to get it
iPhoneApp Store (iOS 16.4 or later, 198.7 MB)
AndroidGoogle Play
Websitealessandrofacchini.github.io/memoryai
Privacy policymemoryai-privacy (English, Italian version linked there; last updated 20 October 2026)
Bug reportsGitHub issues (feedback only, no source code)

Version reviewed: 1.0.56 on both platforms, as shown on the App Store listing (released 22 October 2026) and the Google Play listing (updated 23 October 2026). The developer states that Android 9 or later is required; the Google Play text read shows no minimum Android version.

How to Get Started

Setup follows the order described in the store listings; PromptQuorum has not run these steps.

  1. 1
    Install the app
    Why it matters: Download MemoryAI from the App Store or Google Play; the listings say no account is required and the phone should have 4 GB of RAM or more.
  2. 2
    Download a model on Wi-Fi
    Why it matters: Models are 0.3 to 4 GB each per the listings. The 18 October 2026 update says the first suggested model is now the lightest one that runs well on your phone, and that its download continues if you leave the screen. The 1.0.56 notes add that the app shows a model's size and the expected download time before you start, shows how much is left while it downloads, and picks the download up again by itself if the connection drops.
  3. 3
    Chat or talk
    Why it matters: Type, dictate, or switch on the continuous voice mode, in which the app listens, answers and reads the answer aloud without taps.
  4. 4
    Add documents or images
    Why it matters: Import PDF, TXT or Markdown files, or pick photos, and ask questions about them.
  5. 5
    Optionally turn on Agent Memory
    Why it matters: It is off by default and needs an embedding model; once on, you can open, edit or delete what it has stored.

Features Confirmed by the Sources

Every item below comes from the App Store and Google Play descriptions or the privacy policy.

  • Two engines and custom models. Choose llama.cpp or Google's MediaPipe/LiteRT per model, import any compatible GGUF file from Hugging Face, save a Hugging Face token for gated models, pause and resume downloads, and adjust temperature, top-k, top-p and maximum tokens per model.
  • Agent Memory. Optional; the AI stores what you tell it in your own words on the device, and you can view, edit or delete each memory. The 1.0.52 notes say requests such as "write an email" are no longer saved as memories.
  • Documents and images. Questions about PDF, TXT and Markdown files, image analysis with vision models (several images at once), and text recognition on photos through Google ML Kit.
  • Voice. On-device voice input, text-to-speech, and a continuous hands-free mode.
  • Everyday tools. Chat search, 20 persona templates, a speed benchmark in tokens per second, a context-window display, a screen showing whether the loaded model runs on the GPU or CPU, and backup to Google Drive on Android or iCloud on iPhone.

Kid Mode

The app offers a PIN-protected mode for children that its store descriptions present as a headline feature. Everything below comes from those descriptions; PromptQuorum has not tested how well the mode holds up.

  • Parent PIN. A parent sets a PIN that guards the mode and its settings.
  • Locked model. The active model is fixed, so a child cannot switch models or change settings.
  • Content filter. Inappropriate words are blocked before they reach the model, and a safety prompt guides age-appropriate replies.
  • Import warnings. The app warns before an unverified imported model can run in this mode.
  • Parental unlock. Settings and documents can be opened temporarily from the chat screen without ending the mode.
  • No notifications. The privacy policy says notifications are off while Kid Mode is active, and that turning the mode off restores the parent's choice. It also says usage statistics and crash reports may still be collected in this mode.

Privacy and Network Use

Chats, documents and memories stay on the device by default, and the developer says the app has no AI servers and sends no chat text. The Google Play Data safety section declares no data shared with third parties, with possible collection of app info and performance and of device or other IDs, encrypted in transit. The App Store privacy label lists search history, device ID and product interaction for analytics, and device ID, crash and performance data for app functionality, all marked not linked to you.

The privacy policy, updated on 20 October 2026, names what leaves the phone: usage events to TelemetryDeck, crash reports to Sentry, and model downloads from Hugging Face. Push notifications through OneSignal start only if you accept the in-app "Notifications" prompt, and turning them off in Settings unsubscribes the device. Google Analytics for Firebase, which recorded two events (onboarding completed, message sent) for ad-campaign measurement, ran only on Android versions before 1.0.55, and only if the earlier in-app prompt was accepted; version 1.0.55 and later do not contain it, and the policy says it is not used on iOS. Backup to your own Google Drive or iCloud is optional. The developer adds that usage events carry no chat content (for example, "a model was loaded") and that crash reports carry no chat or document content. The policy also says local data is not encrypted by the app and relies on the device's own protection, and that conversations leave the app only when you export them.

The store descriptions say "no in-app purchases", and the developer confirmed to PromptQuorum that the app has none; an earlier paragraph about purchases in the privacy policy was out of date and has been corrected. The "no tracking" wording does not mean that nothing leaves the phone: the analytics and crash reporting above still apply.

📌Note: The app is closed source, so none of these claims can be checked in code, and PromptQuorum has not audited network traffic. Anyone handling confidential data, or configuring parental controls for a child, should verify behavior on their own device.

From the Maker

Alessandro Facchini, the developer of MemoryAI, checked the draft of this review and sent the note below. It is presented as the developer's own words, not as PromptQuorum's independent editorial assessment. The corrections in the same message on purchases, data flows, engines and the Android version are applied in this review:

"I built MemoryAI for people who want an AI assistant without handing their words to a server. Once a model is downloaded, it answers with no internet connection, and the conversation stays on the phone.

Two ideas shape it. The first is a memory you can see: the app can remember what you tell it, in your own words, and you can open that memory, correct it or delete it at any time. The second is family use: Kid Mode locks the model and the settings behind a parent PIN and keeps replies age-appropriate.

It is free, with no account and no ads, and it speaks six languages. I build it in Italy with a small team, and two testers try new builds on real Android phones and iPhones. Next on the list: tools the model can use on the phone, now in testing, and on iPhone the faster first answer that Android users already have."

— Alessandro Facchini, developer

Who Should Use It

  • People who want a phone assistant that remembers them without a cloud account. The Agent Memory is stored locally and editable.
  • Parents who want a child-facing offline chat. Parental controls are included, with the caveat that their strength is unverified.
  • People who want to question their own PDFs and photos privately. Document and image questions are included.
  • Users who like to bring their own models. GGUF import and two engines suit anyone who already has a favorite model.

What We Could Not Verify

  • Hands-on performance. PromptQuorum did not run the app, so speed, battery use, answer quality and the real strength of the parental controls are not assessed.
  • Source code and license. The app is closed source; the feedback repository holds only a README and an issue tracker.
  • Analytics and traffic. The app is closed source and PromptQuorum has not audited its network traffic, so the data flows above are the account of the developer and the policy.
  • Android requirements. The developer states Android 9 or later, which the Google Play text read does not show; the iOS minimum is 16.4.
  • Track record. The app has been out for about seven months, with 1K+ Google Play downloads and no App Store ratings yet; the sources give no support commitment.
  • Not for users who need very large models. The downloadable models are small, so heavier work belongs on a computer.

Competitors and Alternatives

Private Mind

Platforms:
iOS, Android
Price / license:
Free / MIT
Key difference:
Open-source offline chat with on-device document Q&A and benchmarks
Articles about Private Mind (2)

Also mentioned in:

+2 more not shown

Off Grid

Platforms:
iOS, Android, Mac, Windows
Price / license:
Freemium / MIT
Key difference:
Open-source offline AI suite on phones and desktops

Competitor details change often; confirm each app's current price, license and platforms on its own listing. Of these four, MemoryAI is the only closed-source app.

Compare in directory

Frequently Asked Questions

Who makes MemoryAI?

Alessandro Facchini, who publishes under the name Facchinits; the privacy policy names Facchinits as the data controller.

Can I edit what the AI remembers?

Yes. Agent Memory is optional, stored on the device in your own words, and each entry can be viewed, edited or deleted. It needs an embedding model to work.

Can I use my own models?

Yes. You can import compatible GGUF files from Hugging Face and run them on the llama.cpp engine; the LiteRT engine runs Gemma and, per the developer, also Qwen 2.5 1.5B, DeepSeek R1 Distill 1.5B and Phi-4 Mini in LiteRT format.

Does it work on a Mac?

The App Store lists Macs with Apple M1 or later but marks the app "Designed for iPhone. Not verified for macOS", and there is no separate desktop build.

Where do I report a problem?

In the developer's public feedback repository on GitHub, or by the support email on the Google Play listing.

Verdict

MemoryAI is a free offline mobile chat app that keeps an editable memory of the user on the device and offers parental controls, alongside document and image questions, hands-free voice and two inference engines. It is also frequently updated, with a release the day before this review was checked. Against that, it is closed source, new, small in reach, limited to small on-device models, and its privacy claims cannot be checked in code. It suits readers who want these specific features and accept a young, closed app; readers who want auditable code can compare Private Mind or PocketPal AI.

Sources

← Back to Power Local LLM

MCP