Open Source and 100% Private
This is not Claude or ChatGPT. Ensu runs on your phone or desktop with limited GPU and memory. It does not run in a datacenter.
Yet, the fact that it runs on device means you can feed it any personal information. For example:
- a PDF of your tax forms
- a photo feed of your home security cameras
- any photos or information involving your family or friends
We need more ethical products like this. I’m excited to see how this evolves!
Yet, the fact that it runs on device means you can feed it any personal information. For example:
- a PDF of your tax forms
- a photo feed of your home security cameras
- any photos or information involving your family or friends
We need more ethical products like this. I’m excited to see how this evolves!
Show more
Not great
If it doesn’t know the answer, it shouldn’t offer one. It thought Patrick Swayze died many years before he did and of a completely different illness. Despite telling it was wrong, it was very adamant.
Good thinking but not sure what its use would be for me.
Good thinking but not sure what its use would be for me.
Direkt wieder gelöscht.
Die App hat auf sämtliche meiner Anfragen, ausschließlich blühenden Unsinn in Halluzination.
Great app, but keep limitations in mind
It is a great idea to have a local AI chat . No privacy concerns or subscription fees.
Having an offline AI tool on my iPhone/iPad is most welcome.
The problem is, as another reviewer stated - that Ensu does not know how to say “I don’t know” and gives wrong answers to some questions.
So, double check what it tells you - especially the really important stuff.
Having an offline AI tool on my iPhone/iPad is most welcome.
The problem is, as another reviewer stated - that Ensu does not know how to say “I don’t know” and gives wrong answers to some questions.
So, double check what it tells you - especially the really important stuff.
Show more
Unusable
keeps repeating and gets stuck in loops
Totally unreliable
The idea of the app is solid and well executed but the model used is so inaccurate it’s dangerous. I couldn’t find a *single* factual question it could answer accurately. Not one! It’s dangerous to release such an inaccurate tool to people who might trust the results. Accuracy matters.
GPT-4
Currently used GPT-4 under the hood and is fast on an iPhone 16. While perhaps not as powerful as a cloud-based system like ChatGPT, Gemini, Grok, etc it does well for general use.
Der Feinschliff für Barrierefreiheit fehlt noch
Hallo liebes Entwicklerteam,
ich habe eure App heute getestet und bin mir bewusst, dass sie sich noch in einem frühen Entwicklungsstadium befindet. Ich selbst bin blind und nutze auf meinem iPhone die Bedienungshilfen, insbesondere VoiceOver. Grundsätzlich lässt sich eure App bereits gut bedienen, was ich sehr positiv hervorheben möchte.
Allerdings sind mir einige Elemente aufgefallen, die noch nicht vollständig zugänglich oder eindeutig verständlich sind. Es wäre sehr hilfreich, wenn ihr den Bereich Barrierefreiheit noch einmal überarbeiten könntet, um die Nutzung insgesamt klarer und intuitiver zu gestalten.
Insbesondere bei Schaltflächen fehlt teilweise eine eindeutige Beschriftung. Zum Beispiel ist bei der Schaltfläche zum Absenden eines eingegebenen Textes nicht klar erkennbar, welche Funktion sie erfüllt. Ähnlich verhält es sich bei weiteren Buttons, die häufig nur mit allgemeinen Bezeichnungen wie „Icon“ versehen sind, was ihre Bedeutung schwer nachvollziehbar macht.
Es wäre daher wünschenswert, die Beschriftungen präziser und aussagekräftiger zu gestalten, damit Nutzer sofort erkennen können, welche Funktion hinter einer Schaltfläche steht.
Insgesamt handelt es sich um kleinere Anpassungen, die jedoch einen großen Unterschied machen können. Eine verbesserte Barrierefreiheit würde dazu beitragen, dass wirklich alle Nutzer die App problemlos verwenden können.
Vielen Dank für eure Arbeit und viel Erfolg bei der weiteren Entwicklung.
ich habe eure App heute getestet und bin mir bewusst, dass sie sich noch in einem frühen Entwicklungsstadium befindet. Ich selbst bin blind und nutze auf meinem iPhone die Bedienungshilfen, insbesondere VoiceOver. Grundsätzlich lässt sich eure App bereits gut bedienen, was ich sehr positiv hervorheben möchte.
Allerdings sind mir einige Elemente aufgefallen, die noch nicht vollständig zugänglich oder eindeutig verständlich sind. Es wäre sehr hilfreich, wenn ihr den Bereich Barrierefreiheit noch einmal überarbeiten könntet, um die Nutzung insgesamt klarer und intuitiver zu gestalten.
Insbesondere bei Schaltflächen fehlt teilweise eine eindeutige Beschriftung. Zum Beispiel ist bei der Schaltfläche zum Absenden eines eingegebenen Textes nicht klar erkennbar, welche Funktion sie erfüllt. Ähnlich verhält es sich bei weiteren Buttons, die häufig nur mit allgemeinen Bezeichnungen wie „Icon“ versehen sind, was ihre Bedeutung schwer nachvollziehbar macht.
Es wäre daher wünschenswert, die Beschriftungen präziser und aussagekräftiger zu gestalten, damit Nutzer sofort erkennen können, welche Funktion hinter einer Schaltfläche steht.
Insgesamt handelt es sich um kleinere Anpassungen, die jedoch einen großen Unterschied machen können. Eine verbesserte Barrierefreiheit würde dazu beitragen, dass wirklich alle Nutzer die App problemlos verwenden können.
Vielen Dank für eure Arbeit und viel Erfolg bei der weiteren Entwicklung.
Show more
Model does not download
Model does not download
Its perfect for me
For I generally use chat bots for, this is literally perfect
Thank u🫡
Thank u🫡
Tried twice. Unable to load model on iPhone 17
Tried this app twice. Can’t even get it to load the model both times. Tired in previous initial version and this updated version. Same error, unable to load model despite it downloading. Tried on iPhone 17 and older iPhone 14 Pro, and still, can’t get it to load. I’ve tired deleting the app and trying again. Tried disabling adblockers. I’ve tried restarting the phones. At a loss why it’s not loading. GitHub shows others having model loading issues on the .AppImage but no real place to ask questions outside of discord which I’m avoiding.
Show more
Good start
Nice idea, needs some improvement for context stability, it crashes after 5 prompts and needs an app restart. Improvement would be a memory setting with basic info about the user that is imported in new chats. Also please add a icon switcher. I liked the old one.
Dysfunctional?
From my first test, this doesn’t work at all, I can’t send the message I’m typing…
Is this another vibe coded slop?
Bug: I have been hitting enter for the last several minutes to start a conversation without seeing any other input button and finally realise all those other UI interactive elements including ‘send’ are simply invisible.
Model: It feels weirdly awkward to use GPT-3 back from 2020 against GPT-5.5 in 2026 when multiple open weight models from DeepSeek, Qwen to Sarvam exists because it’s just plain useless even against Mistral AI.
App: No native integrations like Apple Shortcuts as opposed to ChatGPT where personalisations are also a standard now.
Being able to directly grab models from Hugging Face like with Draw Things in addition to downloading one separately is what would make this app truly appreciable.
Perhaps, start using this app yourselves?
Model: It feels weirdly awkward to use GPT-3 back from 2020 against GPT-5.5 in 2026 when multiple open weight models from DeepSeek, Qwen to Sarvam exists because it’s just plain useless even against Mistral AI.
App: No native integrations like Apple Shortcuts as opposed to ChatGPT where personalisations are also a standard now.
Being able to directly grab models from Hugging Face like with Draw Things in addition to downloading one separately is what would make this app truly appreciable.
Perhaps, start using this app yourselves?
Show more
Fast
It’s fast and local, lots of potential for new features.
Love Ente; please slow down
What We Tested and What We Found
We put Ente’s new AI assistant, Ensu, through six demanding questions spanning probability puzzles, philosophical reasoning, and real-world emergencies like wilderness medical crises, car breakdowns, and a sick child in the middle of the night. Across every category, Ensu demonstrated the same core weakness: it could recognize that a situation was serious without understanding why it was serious. It dressed up its answers in confident formatting, numbered steps, bold headers, organized sections, which made the responses look authoritative while the actual content was frequently wrong or dangerously incomplete. In the medical scenarios especially, it missed critical diagnoses entirely, identifying unequal pupils as a possible eye infection rather than a brain emergency, and describing a textbook meningitis presentation in a child without ever mentioning meningitis.
Should You Trust Ensu in a Pinch?
In a word: no, at least not for anything that matters. This is a painful conclusion to reach about a product from Ente, a company that built its reputation on being genuinely trustworthy with things people care about. Ente Photos passed independent security audits and earned real loyalty by doing one thing exceptionally well. Ensu does not carry that same standard forward. It scores adequately on presenting information clearly, which makes it feel more reliable than it actually is, and that gap between feeling trustworthy and being trustworthy is the most important finding here. For low-stakes tasks like drafting a note or answering a general question, it would likely serve you fine. But if you were considering using it as an offline reference during a camping emergency, a roadside breakdown, or a medical scare, our testing suggests Ensu could actively steer you wrong at the exact moment you need accurate guidance most. The danger isn’t that it says “I don’t know” — it’s that it never does.
We put Ente’s new AI assistant, Ensu, through six demanding questions spanning probability puzzles, philosophical reasoning, and real-world emergencies like wilderness medical crises, car breakdowns, and a sick child in the middle of the night. Across every category, Ensu demonstrated the same core weakness: it could recognize that a situation was serious without understanding why it was serious. It dressed up its answers in confident formatting, numbered steps, bold headers, organized sections, which made the responses look authoritative while the actual content was frequently wrong or dangerously incomplete. In the medical scenarios especially, it missed critical diagnoses entirely, identifying unequal pupils as a possible eye infection rather than a brain emergency, and describing a textbook meningitis presentation in a child without ever mentioning meningitis.
Should You Trust Ensu in a Pinch?
In a word: no, at least not for anything that matters. This is a painful conclusion to reach about a product from Ente, a company that built its reputation on being genuinely trustworthy with things people care about. Ente Photos passed independent security audits and earned real loyalty by doing one thing exceptionally well. Ensu does not carry that same standard forward. It scores adequately on presenting information clearly, which makes it feel more reliable than it actually is, and that gap between feeling trustworthy and being trustworthy is the most important finding here. For low-stakes tasks like drafting a note or answering a general question, it would likely serve you fine. But if you were considering using it as an offline reference during a camping emergency, a roadside breakdown, or a medical scare, our testing suggests Ensu could actively steer you wrong at the exact moment you need accurate guidance most. The danger isn’t that it says “I don’t know” — it’s that it never does.
Show more