crash crash crash
an other update with hopes crushed by constant crashes, voice mode crashes when you use the auto mode after the first answer, it does not switch into listening mode, just simply crashes no matter what voice model or ai model you use... on an M2 ipad pro with 16 gigs of ram... this app was never stable, pretty much unusable, no response from developer... don't spend money on this app, not worth it until they fix the bugs instead of packing more and more features into it.
Show more
Response from developer
Hi trancemuter, thank you for using the app. We apologize for the inconvenience. It was a bug that has now been fixed in the latest version. You should be able to long-press and copy the model response in iOS, and select text on macOS. Please update to the newest version to try it out.
High quality and well thought out
Excellent implementation of features and very robust. This app is more advanced than most which allows you to do so many neat things. Highly recommend to pay the small fee and unlock everything
Response from developer
Thank you for supporting!
hard crash!!!
today I saw that there is an update, I was happy to try it out, maybe they fixed the problems it had, I sent emails to them about it, (no response btw) well, I open it up and starts loading the last model I was using then hard crashed immediately, not just a normal crash, it was blank screen crash!!! it was stuck on blank screen for a while unresponsive to anything, then my iPad restarted by itself, thanks God! this is the only app that hard crashes my iPad and I have over a thousand apps, I even beta test apps, I am a power user, no other app ever crashed my iPad as this app is crashing it, not only now, this app had these hard crashes many time despite the updates! I have a maxed out iPad pro M2, 16gb ram 2 tb storage, I have about 700 gigs free space on it, I was only running this app on a freshly restarted machine... this is not normal! idk what else to say but this developer is not a pro! no response from support either for weeks now, I have a lifetime membership... I think there is a new owner or something, I had good working relationship with the developer before, I was testing for them, he was responsive to me before, not anymore... yeah... š¤·āāļø idk...
Show more
Response from developer
Hi trancemuter, thank you for using the app. We apologize for the inconvenience. It was a bug that has now been fixed in the latest version. You should be able to long-press and copy the model response in iOS, and select text on macOS. Please update to the newest version to try it out.
Keine Reaktion nach dem Sendenā¦
Weder das lokale noch eine Downloaded AI reagieren auf die User-Eingabeā¦ lost Time & Energy!
A Game-Changer for Personal AI Assistants
Iāve tested countless AI apps over the years, and OnDevice AI truly stands out as a powerhouse. Its expansive feature set covers everything you need to build and customize your own AI assistant, making it incredibly versatile for both casual and advanced users. I primarily rely on it for voice-to-text functionality across various models, which enables seamless, hands-free operation thatās a total game-changer for productivity on the go.
Iāve also dived deep into the chat flow feature, where you can assign multiple ācharactersā to role-play interactions between different AI models. Itās endlessly creative and funāperfect for brainstorming, storytelling, or even simulating debates. The built-in search tool is another highlight; I use it frequently to pull in real-time information without breaking my workflow.
What I love most is running OnDevice AI strictly on my MacBook (with 64GB RAM) as an endpoint via LM Studio. This setup lets me handle the largest AI models effortlessly, without any performance hiccups or memory overload on my devices. The app integrates beautifully with Kokoro voices or Appleās default voices, delivering natural, fluid conversations that feel truly unencumbered.
I seamlessly use this app across my iPhone, iPad, MacBook Pro, and Vision Proāitās a unified experience thatās hard to beat. If I could, Iād rate it 10 stars!
Suggested Features and Improvements
While OnDevice AI is already exceptional, here are some thoughtful enhancements that could elevate it even further:
ā¢ Automated Multi-Character Conversations: Introduce an option to select two or more characters and let them engage in ongoing, autonomous discussions on a given topic. Right now, users must manually alternate between characters for each exchange, which can feel tedious. As a workaround, Iāve experimented with setting up an iPhone and iPad to ātalkā to each other via voice-to-voiceāitās surprisingly immersive, like listening to a live story unfold!
ā¢ Character Set Saving and Syncing: Add functionality to save, export, and import character configurations across devices. Currently, I have to recreate them manually using copy-paste and Appleās Continuity features, which disrupts the seamless multi-device experience.
ā¢ Streamlined Search Integration: Refine the search tool to integrate more fluidly into chats without interrupting the flow (e.g., eliminate the toggle that converts the chat field into a URL input). Extend voice-mode access for quick, on-the-fly queriesālike checking a productās priceāwithout switching sections or resuming the conversation separately afterward.
ā¢ UI Optimization for Icons: Consolidate key elements like the microphone and speaker icons into a single, more prominent area, and enlarge them slightly for better accessibility, especially on the Vision Pro where spatial interactions demand intuitive design.
ā¢ Remote Endpoint Support for Multi-Agents: Expand the multi-agent (character) system to include remote endpoints, in addition to local devices. This would align perfectly with OnDevice AIās strengths in handling distributed AI workloads.
ā¢ HomeKit Integration: Incorporate basic controls for smart home devices, such as adjusting lights or playing music. While Siri handles this natively, OnDevice AI could offer a more customizable, AI-driven alternative for enhanced automation.
ā¢ Improved Documentation: Provide clearer guides on importing models from sources like Hugging FaceāI havenāt been able to do this successfully yet, and step-by-step tutorials would be invaluable for users experimenting with custom setups.
ā¢ Expanded Agent Input Window: Enlarge the text input area for defining agents/characters; itās currently quite restrictive, especially when this should be the focal point for creative customization.
ā¢ Per-Agent Model Assignment: For each agentāparticularly in remote or cloud-based scenariosāadd the ability to assign specific model sets. My MacBook can handle running multiple models simultaneously, and while this might be a niche feature, it would greatly aid in evaluating whether a particular model is suited to the task at hand.
These are just a few ideas to build on an already outstanding appāI have more in mind, but this covers my top priorities for now. Keep up the fantastic work!
Thank you!
Iāve also dived deep into the chat flow feature, where you can assign multiple ācharactersā to role-play interactions between different AI models. Itās endlessly creative and funāperfect for brainstorming, storytelling, or even simulating debates. The built-in search tool is another highlight; I use it frequently to pull in real-time information without breaking my workflow.
What I love most is running OnDevice AI strictly on my MacBook (with 64GB RAM) as an endpoint via LM Studio. This setup lets me handle the largest AI models effortlessly, without any performance hiccups or memory overload on my devices. The app integrates beautifully with Kokoro voices or Appleās default voices, delivering natural, fluid conversations that feel truly unencumbered.
I seamlessly use this app across my iPhone, iPad, MacBook Pro, and Vision Proāitās a unified experience thatās hard to beat. If I could, Iād rate it 10 stars!
Suggested Features and Improvements
While OnDevice AI is already exceptional, here are some thoughtful enhancements that could elevate it even further:
ā¢ Automated Multi-Character Conversations: Introduce an option to select two or more characters and let them engage in ongoing, autonomous discussions on a given topic. Right now, users must manually alternate between characters for each exchange, which can feel tedious. As a workaround, Iāve experimented with setting up an iPhone and iPad to ātalkā to each other via voice-to-voiceāitās surprisingly immersive, like listening to a live story unfold!
ā¢ Character Set Saving and Syncing: Add functionality to save, export, and import character configurations across devices. Currently, I have to recreate them manually using copy-paste and Appleās Continuity features, which disrupts the seamless multi-device experience.
ā¢ Streamlined Search Integration: Refine the search tool to integrate more fluidly into chats without interrupting the flow (e.g., eliminate the toggle that converts the chat field into a URL input). Extend voice-mode access for quick, on-the-fly queriesālike checking a productās priceāwithout switching sections or resuming the conversation separately afterward.
ā¢ UI Optimization for Icons: Consolidate key elements like the microphone and speaker icons into a single, more prominent area, and enlarge them slightly for better accessibility, especially on the Vision Pro where spatial interactions demand intuitive design.
ā¢ Remote Endpoint Support for Multi-Agents: Expand the multi-agent (character) system to include remote endpoints, in addition to local devices. This would align perfectly with OnDevice AIās strengths in handling distributed AI workloads.
ā¢ HomeKit Integration: Incorporate basic controls for smart home devices, such as adjusting lights or playing music. While Siri handles this natively, OnDevice AI could offer a more customizable, AI-driven alternative for enhanced automation.
ā¢ Improved Documentation: Provide clearer guides on importing models from sources like Hugging FaceāI havenāt been able to do this successfully yet, and step-by-step tutorials would be invaluable for users experimenting with custom setups.
ā¢ Expanded Agent Input Window: Enlarge the text input area for defining agents/characters; itās currently quite restrictive, especially when this should be the focal point for creative customization.
ā¢ Per-Agent Model Assignment: For each agentāparticularly in remote or cloud-based scenariosāadd the ability to assign specific model sets. My MacBook can handle running multiple models simultaneously, and while this might be a niche feature, it would greatly aid in evaluating whether a particular model is suited to the task at hand.
These are just a few ideas to build on an already outstanding appāI have more in mind, but this covers my top priorities for now. Keep up the fantastic work!
Thank you!
Show more
Best app for ai in 2025
From being on device too haveing a advanced search feature too text to voice this is one of the best ai applications Iāve found the old ui was a lot more difficult too use but with recent updates the ui has never been easier too understand and get around thank you for polishing this application 10/10 would reccomend
Show more
Response from developer
Thank you so much for your support!
I loved this app but it is broken now for me.
I really like this app, but now whenever I hit enter to send my prompt and get a response, it immediately crashes. Up until this started happening I had been very much enjoying this app and had several very complex contexts that I was loving being able to run local on my iPad with it. Maybe it was a recent update? PLEASE FIX IT! I want to come back and give it 5 stars.
Show more
Response from developer
Thank you so much for taking the time to share this feedback! I really appreciate you letting me know about this issue, and I'm sorry to hear that the app is crashing Could you please provide me with a few more details to help me troubleshoot this: - Which model are you currently using when the crash occurs? - Does the crash happen with all models, or just a specific one? Please feel free to contact me directly at developer@ondevice-ai.app with these details, and I'll work on getting this fixed as quickly as possible. I'm committed to resolving this so you can return to the experience you were enjoying. Thank you again for your patience and for wanting to give the app another chance!
Crashes at launch
I'm surprised that this app crashes at boot seeing I have a 12g RAM iPhone Air.
The design is clunky with settings strewn throughout the app, making finding anything a challenge. There is no manual, and suggested AI models are far too large to run.
This app is a carbon copy of other AI app. If you're gonna come up with an AI app at least tell people where all the buttons and settings are, because again they are not all in one place.
I subscribed yesterday, I guess it's time to get my money back.
The design is clunky with settings strewn throughout the app, making finding anything a challenge. There is no manual, and suggested AI models are far too large to run.
This app is a carbon copy of other AI app. If you're gonna come up with an AI app at least tell people where all the buttons and settings are, because again they are not all in one place.
I subscribed yesterday, I guess it's time to get my money back.
Show more
Good Stuff
If it could update my iOS calendar and reminders too, Iād probably use it every day!
Amazing and works with the voiceover screen reader
Thank you so much for developing this app! This is by far the easiest app that I have found to locally run AI models. And it is completely accessible with the voiceover screen reader as well. The only problem that I am running into, and it might just be the models that I have tried, is I can only send messages and receive messages that end up being 100 tokens and after that I have to make a new conversation and start over again. I have changed the settings for tokens so responses wonāt be as long, but no matter what, when it gets to the hundred tokens, I have to start over.
Show more
Response from developer
Thanks so much for the kind words! Really glad to hear the app is working well with VoiceOver. That 100 token issue shouldnāt be happening. Could you let me know what device youāre using and which model youāve tried? That would help me figure out whatās going on. You could send me email at developer@ondevice-ai.app Also, we recently added options to connect to cloud providers or local servers, might be worth trying those to see if it helps with your situation. Thanks again for the feedback!
No Custom MLX?
The app is very good, though it crashes with large context sizes. You can download custom GGUFs, though the app runs MLX models. I am hoping custom MLX models can also be downloaded soon. Other apps like Mollama do allow custom MLX.
Upgrading my review to 5 stars...
UPDATE: The Developer's prompt response below took care of the problem after I unplugged my iPad from the charging brick, while running the app, my device no longer overheated. I really like this app and appreciate the rapid response support. Having access to LLMs offline (when wi-fi connectivity isn't available) is a much appreciated gift. Thanksšš¼ ............................. ORIGINAL REVIEW: How can I use this app on my ipad pro (12.9" m1 fifth gen 16mb device with 1TB memory) but without it overheating my device??? For example, I just ran a simple search and analysis of cutlery "super-steels" and my device screen promptly heated up to the touch. No other apps I have used elicit this speedy and high a temperature increase. I am concerned that continuing to use it will damage my iPad. Developer, please discuss and identify how to fix this problem. Please note that I downloaded only one of the two LLMs recommended by the app for this iPad's capabilities.
Show more
Response from developer
Thank you for your feedback! Since the AI models run locally on your device for privacy, intensive processing does generate heat, especially during long tasks. This is particularly noticeable when your device is connected to power during use. Immediate suggestions: 1) Disconnect from power cable during AI processing when possible 2) Take breaks between longer tasks to let device cool down I'm going to adding a thread control setting that will let you adjust processing threads. Lower thread count = less heat but slower processing. Higher thread count = faster but more heat. You'll be able to choose the balance that works for your needs. Thanks for helping make the app better!
Well, a treasure and a nightmare
Well, it is very impressive that this app has the models in your device and completely offline, the thing is that it wonāt even answer a question, it will turn black and then opt you out of the app, I think not even a quarter of the cellphones and iPads un the world can handle this app.
screen blocked
With the latest updates I have to force restart my phone each time I start the app, because the screen shows some wierd stuff and you cannot control it at all.
This app is causing my iPad Pro to overheat...
How can I use this app on my ipad pro (12.9" m1 fifth gen 16mb device with 1TB memory) but without it overheating my device??? For example, I just ran a simple search and analysis of cutlery "super-steels" and my device screen promptly heated up to the touch. No other apps I have used elicit this speedy and high a temperature increase. I am concerned that continuing to use it will damage my iPad. Developer, please discuss and identify how to fix this problem. Please note that I downloaded only one of the two LLMs recommended by the app for this iPad's capabilities.
Show more
Amazing
It is an absolutely amazing on device AI. I would like to ask: what is the exact AI model of the ādefault modelā?
Good app
Feature rich, large variety of models.
Pretty Good
Cant get the search to work as all the LLMs I have tried all say DuckDuckGo or Bing dont have any information or that the page is just the search engines privacy policy or some error happens (AI_on_Device.WebBrowsā¦
If this did work it would be a 5 star app for sure.
It also seems to be missing the huggingface search that other apps are starting to include which gives a good option to search model cards and find specific quantizations.
Though it is really awesome that MLX models are offered in the downloads area.
I like the UI, though the sidebar buttons sometimes lag or dont register taps.
Iād buy lifetime if this had Kokoro and/or VITS text-to-speech.
If this did work it would be a 5 star app for sure.
It also seems to be missing the huggingface search that other apps are starting to include which gives a good option to search model cards and find specific quantizations.
Though it is really awesome that MLX models are offered in the downloads area.
I like the UI, though the sidebar buttons sometimes lag or dont register taps.
Iād buy lifetime if this had Kokoro and/or VITS text-to-speech.
Show more
Token problem
The idea is great but I can not use in a full posinilities. Why do I need tokens? Where I can exchange my text promt? What is this restrictions? On my iPhone it runs great, but why do I need some tokens to exchange text? Where I can get it???
insanely open and configurable
The amount of freedom this app gives you with changing the model and configuring it is much more than even some desktop LLM apps. The indication of what the LLM is doing or what links itās scraping in the search mode is also a nice touch. It can even run some decent models on 6gb devices like the iPad a16 or iPhone 13 Pro Max.
Show more
Response from developer
Thank you for the support! Iām continuously working to improve the app. Please let me know if you have any suggestions!
