# Off Grid AI - Private Local AI — user reviews > 21 indexed App Store user reviews of Off Grid AI by Wednesday Solutions, Inc, rated 3.00/5 from 24 ratings. - Source: https://appshunter.io/ios/app/off-grid-ai-private-local-ai/id6759299882/reviews - App overview in markdown: https://appshunter.io/ios/app/off-grid-ai-private-local-ai/id6759299882.md - Rating breakdown: 5★ 8 · 4★ 4 · 3★ 1 · 2★ 3 · 1★ 8 ## Newest reviews (2 of 21) ### 4/5 — My Current Main Local AI App *2026-05-25* I really like this app among the slew of other similar local ai apps in the App Store. But this one has a lot of the features I like and need. You can download models from their huggingspace search, including uncensored models or I can drop in my own models from the files app. Their text2image models doesn’t seem to work on my iPhone 17 Pro Max (2TB model), this wasn’t on my immediate needs but would like to revisit in the future. I like the UI, very clean and not flashy. Some gripes: When editing messages in chat, you cannot click on the text to select, it keeps making the keyboard disappear making it impossible to select text or move cursor manually. Editing text is the limited to using hold space to move cursor. But I can’t select, copy and/or pasting. For vision models, the + symbol for adding attachments, it add an unnecessary step to add from photos or camera. Instead of just two selections, photo and file, they should just add a third camera in one go. It makes you click photo, then I have to choose between camera or photo library. Make it seamless in one go. Suggestion to add Kokoro TTS, would make chats immersive. **Developer response:** Fair points on the edit prompt bug - we will fix this in the next release,Kokoro TTS is coming soon. ### 4/5 — Great UI- Still in Development *2026-04-19* The UI for this app is about the best I’ve found for a local AI interface, however the feature to add a document or load context is completely broken for me as of now, which is a pretty important aspect for me. Also, even for models that support higher context length (such as Qwen 3.5 which says it supports 262k) the app still seems to limit you to 32k, which is a shame. Great app otherwise, clean, customizable, and easy to use ## Most critical reviews (10 lowest-rated) ### 1/5 — It’s very good *2026-07-30* Nice and straightforward UI, and I really appreciate how customizable it is compared to other apps, it has everything that you could need, but 1 editing text is broken, 2 thinking constantly leaks into messages as well as the AI will abruptly restart its thinking or its output, 3 the app keeps auto downloading models I don’t want Edit: what I mean in 3 is that it will sometimes download SmolLM2-135M, by itself when I never downloaded it. And if I delete a llm, it either never gets deleted, or more likely, is re-downloaded again, maybe some local or memory issue? **Developer response:** thank you for the kind words. I was able to reproduce 2 issues. I'll work on those. Could you clarify what you mean by the the app keeps auto downloading models I dont want? ### 1/5 — toy models *2026-06-21* wont connect to ollama server if it thinks its open internet ip (even if device ip and server ip are in the same subnet, i assume it specifically whitlists 192.168 ip and maybe 10. but did you know thats not how ip actually works! shock.) no way to tell it “this ip is local, do it anyway” so only small local models. which is borderline useless. ### 1/5 — PRO the enshtfkashun *2026-05-26* Was an OK app, no tracking, no "PRO" junk. And nowhere near close to optimized. But, everyone wants to get rich off the open source and open weights; packaged in a PRO version for a price. Oh well, guess its on to the next app. **Developer response:** I understand your frustration - the intention for PRO is not some get rich idea but the notion of maintaining O.S.S - and for that PRO is our first step helps us sustain devs on this - I hope you understand - Btw there is no tracking, there never will be. ### 1/5 — Off Grid until it’s not *2026-05-21* Really solid app in general with a few gripes. The minor one is the edit function is totally borked. Attempting to select text aside from the very end of the prompt causes the text menu to instead move out of the way and close the keyboard. To actually edit anything earlier in the prompt, you have to delete everything after it. The really awful thing is that the app calls itself “Off Grid” but has a built in, untoggleable web search tool that it uses within your chats without direct user prompting. You could be asking your model a private question and it will just throw your prompt into a search engine without your consent. What about that is “off grid”? Why is it not clearly surfaced to the user and able to be disabled? **Developer response:** Both valid points. The text editing bug is on our list. On web search - you can disable it: tap the tools icon above the keyboard and toggle off web_search. We agree it should not be on by default for a privacy-first app. We're changing the default to off in the next release. Thanks for holding us to the standard. ### 1/5 — Glitchy, unusable, had high hopes *2026-05-16* Can’t load the only image model on iPhone Air (SDXL), get a missing model component error. Tried redownloading model. Black screens between tabs sometimes. Tooltips showing on wrong tabs after model downloads. First text model QWEN 3.5 XXS stuck in loops and reading first message when I ask a second question. UI is decent but developer focused. Limited model options. **Developer response:** We've solved this in our recent update, please take a look, would be awesome if you can test the same and if it works for you then please consider giving us 5 stars - looking forward to serving you. ### 2/5 — Doesn’t really work *2026-06-06* Iphone 15 pro and the app crashes half the time you try and load a model even if it should fit in your RAM. The image generation model simply doesn’t load at all either. It has potential but this thing needs bug fixes **Developer response:** Sorry for the inconvenience. This is not expected behavior. Since Off Grid doesn't collect analytics, logs, or crash data, I can't see what caused the issue. Could you share which models you're trying to load and when the crashes occur? If I may request your help in investigating this further, please join our Slack community: https://tinyurl.com/48ffcc2z. I can help diagnose this issue better there . ### 2/5 — Can't run Gemma 4? *2026-05-16* Really wish Gemma 4 would load. I have the 17 Pro Max and can't seem to get even the smallest G4 models to run. Runs flawlessly in Edge Gallery. **Developer response:** Do update to the latest version - Gemma 4 is now running on Offgrid. ### 2/5 — Looks great, but nothing works *2026-05-01, version 0.0.81* Connected to my local lm studio server running on windows fine and lists the models but that’s it. Selecting a model it doesn’t perform JIT loading. Loading a model manually and selecting the same model in the in the app it doesn’t work. Quick Look at the logs in lm studio and it seems to try to load models I’ve not even selected. Downloaded an image model within the app doesn’t work either. Comes up with an error. Great idea in theory just doesn’t work. I’ll test again and report back if the bugs are fixed. ### 3/5 — Has gone down hill *2026-06-05* This once was what I was looking for to run AI on IOS. However now sends incomplete prompts and is remarkably slow and stutters even with plenty of ram and small models. When I originally downloaded it was much more responsive and sent the full prompt you typed. Now it unpredictably cuts off part of your prompt at random and sends duplicate prompts that make it barely usable. Now that it runs like trash they have the audacity to ask you to purchase the pro version. **Developer response:** Sorry for the experience, and thank you for the detailed feedback. The latest iOS updates have primarily focused on fixing bugs and improving stability. Since Off Grid doesn't collect analytics or usage data, it's difficult for me to pinpoint what caused the slowdown, prompt truncation, or duplicate submissions on your device. I'd genuinely love to get the experience back to where you enjoyed using the app. If you're open to helping investigate, I'd greatly appreciate it if you could join our Slack community: https://tinyurl.com/48ffcc2z I can help diagnose it better there . ### 4/5 — It’s very good *2026-07-22, version 0.0.103* Nice and straightforward UI, and I really appreciate how customizable it is compared to other apps, it has everything that you could need, but 1 editing text is broken, 2 thinking constantly leaks into messages a lot as well as the AI will abruptly restart its thinking, 3 the app keeps auto downloading models I don’t want. Edit: what I mean in 3 is that it will sometimes download SmolLM2-360M, by itself when I never downloaded it. And if I delete a llm, it either never gets deleted, or more likely, is re-downloaded again, maybe some local or memory issue? ## Most praised reviews (9 highest-rated) ### 5/5 — It’s good *2026-07-13, version 0.0.102* Nice and straightforward UI, and I appreciate how customizable it is compared to other apps, it has everything that you could need, but editing text is broken. I also want per-message addable tools not just only menu wide. Also reasoning still leaks into messages a lot, or the AI will abruptly restart its thinking. ### 5/5 — It’s good *2026-06-28* Nice and straightforward UI, and I appreciate how customizable it is compared to other apps, it has everything that you could need, but editing text is broken. I also want per-message addable tools not just only menu wide. ### 5/5 — The best Edge AI for my iPhone XR. Haven’t seen anything better. *2026-05-24, version 0.0.93* Hosting llama from my MacMini works perfectly ! Better still this is the only app that seems to allow local Edge AI to work on my 7 year old iPhone XR, all the others just crash. Thank You !!! It runs: SmolLM-2-360M-Instruct at a VERY usable speed, although the answers are a bit limited. However, despite recommending not using the model, Llama-3.2-1B-Instruct-Q4_K_M I select Load anyway and it does actually work. It replies one word a second for a few seconds then, speeds up to, what I would say is a normal AI response speed !!! Excellent work !! I don’t know what you are doing differently to every other Edge AI App, but it works perfectly on a very limited phone. Highly recommended. ### 5/5 — Sdcpp for local image generation instead of local dream *2026-05-15* If we could get sdcpp we could use any model we want that’s supported instead of just sd15,20,21 and sdxl ### 5/5 — Nicely designed llm app *2026-04-09, version 0.0.81* Good UI, easy to use, good selection of models ### 5/5 — Fantastic *2026-03-26* Exactly what I was looking for to run AI on IOS. Even image generation. All privately and locally. Thank you! ### 5/5 — Amazing *2026-03-24, version 0.0.81* Amazing ### 5/5 — Great app! *2026-03-15* Latest update need work, broke the ability to load models. I’m sure they’re on top of it though. Seriously! Other than that, it’s amazing! This is the one! I deleted the others! Hands down! ### 4/5 — It’s very good *2026-07-16* Nice and straightforward UI, and I really appreciate how customizable it is compared to other apps, it has everything that you could need, but 1 editing text is broken, 2 thinking constantly leaks into messages a lot as well as the AI will abruptly restart its thinking, 3 the app keeps auto downloading models I don’t want. --- *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 2026-08-26.*