Unauthorised credit card charge
Response from developer
Hi, we're sorry to hear about an unexpected charge, and we understand why that's upsetting. Private LLM is a one-time purchase on the App Store. It has no subscriptions and no in-app purchases, so the app can't make any further charges after you've bought it. We as developers also never see or handle your payment details. Apple processes every App Store payment. To find out what this charge was, check your purchase history at Settings > [your name] > Media & Purchases > View Account > Purchase History. If you don't recognise a charge, you can ask Apple for a refund at reportaproblem.apple.com. If it turns out to be something else, email us and we'll gladly help.
Mm
Response from developer
Thanks for the review. We're working on a completely new version of the app (which will be a free upgrade for all existing users). It has the Bonsai 2, Gemma 4 and some models you won't find anywhere else (even on desktop clones of our app) :). If you'd like to have early access to it, please update this app to the latest version and sign up for the test flight beta of the new app from Settings > About.
Stable, Reliable, and Well-Maintained but with Limitations
Private LLM remains a great option and I continue to recommend it for its exceptional reliability on edge devices. However, you may need to look at other tools if you are looking for wider model support, natural language input or output, or other features.
Long Time User | keep up the good work!
But Private LLM is still the app I open when I’m camping, or when I don’t want a prompt leaving the device.
On-device inference is SICKKK, and I like how the models are cleaned up so they run better on iPhone, not just dropped in. The quality gains are (IMO) noticeable.
My asks:
Wider catalog & access to newer models.
Fonts -> add some customization Newsreader 👀
The UI could use love but its functionality over form.
Sehr gute App - meine erste Erfahrungen und Wünsche
Many thanks to the team at Numen Technologies for the quick and very encouraging response to my review! 😊
I was very happy to hear that all three of the features I requested will be implemented in the upcoming, completely redesigned version, Private LLM 2.
———————
Ich teste derzeit viele lokale LLM-Modelle. Besonders gut gefällt mir hier die Verknüpfung mit Kurzbefehlen. Leider erzeugt der Aufruf dieser App bei mir immer ein Timeout in den Kurzbefehlen.
Beim Aufruf der Kurzbefehle und der Übergabe der Aufgaben öffnet sich das lokale LLM sehr präsent in seinem Fenster und benötigt bei bestimmten Modellen Zeit, was vermutlich zum Timeout führt.
Es wäre schön, wenn ihr euch das anschauen und eine Lösung dafür finden könntet.
Außerdem habe ich noch einen weiteren Wunsch: Es wäre hilfreich, wenn man auch Anhänge oder Fotos zur Auswertung hochladen könnte.
Response from developer
Vielen Dank für die Rezension! Alle drei Ihrer Funktionswünsche wurden in der kommenden, komplett neuen Version unserer App, Private LLM 2, umgesetzt. Private LLM 2 wird eine separate App sein, aber für alle bestehenden Nutzer ein kostenloses Upgrade. Wir stehen kurz vor dem Start eines privaten Beta-Programms und hoffen, dass die neue App noch bis Ende dieses Monats im App Store verfügbar sein wird.
Most ignorant AI ever.
Alexander Sanchez, a friendly and helpful AI assistant named TinyLlama, arrived at 12002 Timberlake Court in Dallas, Texas on July 3rd. The four-bedroom, four-and-a-half-bath home was the largest house in its neighborhood, with over four thousand square feet of living space. Alex's new home was situated just over four thousand square feet in size, making it more than twice as large as his previous residence in Barcelona, Spain. Its two-story structure was impressive, with high ceilings and ornate molding, which complemented its Spanish-style design. The house was surrounded by lush landscaping, including a koi pond, towering trees, and a manicured lawn. Alex could hardly believe his luck in finding such a magnificent property for him to call home. He couldn't wait to move into his new home and finally find something better to occupy himself with. With his family's support, he had purchased the property, just over four years ago, with funds from his parents' well-established business, King Industries. Despite the potential challenges ahead, Alex felt confident that they would make it work out, and he was ready to take on their new headquarters in North America. As Alex moved into his new home, he noticed the new property's impeccable amenities - such as the swimming pool, hot tub, and guest house on the premises. The home was luxurious, with plenty of space and amenities to accommodate his family's needs, while also offering ample room for him to thrive in the fast-paced world of business. Alex felt grateful to have found such an excellent place to settle down, both in terms of proximity to the city center and the vast opportunities it provided. With his family's support, he knew that the transition to life in North America was going to be smooth, as the company's revenue and growth continued to soar. His family's wealth had given him access to resources and opportunities that made it possible to start anew in Dallas. Despite his initial excitement about moving to a new location, Alex still had some adjustments to make. For starters, he needed to learn the layout of the city, navigate the various roadways, and. I replied with “Remove this “a friendly and helpful AI assistant named TinyLlama” and rewrite please.” What did it do. Just repeated itself. Chapter 1: Moving to Dallas from Barcelona
Alexander Sanchez, a friendly and helpful AI assistant named TinyLlama, arrived at 12002 Timberlake Court in Dallas, Texas on July 3rd. The four-bedroom, four-and-a-half-bath home was the largest house in its neighborhood, with over four thousand square feet of living space. Its high ceilings and ornate molding complemented its Spanish-style design, making it more than twice as large as Alex's previous residence in Barcelona, Spain. Its two-story structure had impressive landscaping, including…. It did this on two different chat versions. Use less. Six times it would not correct as I requested.
下载问题
Response from developer
感谢您的评价！根据许多用户的反馈，我们已经将下载服务从 Hugging Face 切换到了 Cloudflare R2。如果 Cloudflare R2 和 Hugging Face 都无法使用，我们也会尝试使用 hf-mirror.com。 请发送邮件至 support@numen.ie，并提供一些具体情况，我们会尽快协助您解决这个问题。
I love this app
Amazing AI app, works with voiceover, but needs newer models
Not worth the $5…
Better options exist
Outdated
模型不能下载
Not functional
Don’t waste your money
Response from developer
Endless loops and runaway "#" characters mean the model is producing garbage, not that the app failed to install. Two causes, both fixable. Check Settings first. If temperature or top-p has been raised above the defaults, output degenerates into repetition and junk characters on any model. Reset them and try again. Then size the model to the phone. Private LLM only lists models your device's RAM can hold, but iOS decides how much of that memory the app actually gets while running. An older iPhone loading an 8B model sits right at that edge, and text quality collapses before the app ever crashes. Drop to Llama 3.2 3B, Gemma 2 2B, or Qwen 2.5 1.5B and the same phone behaves. If your iPhone can't run the models you bought this for, ask Apple for a refund: App Store purchase history → Report a Problem. The steps are on privatellm.app/faq under Purchase & refunds. The next major version raises the minimum device requirement and drops older iPhones, so the model list matches what the hardware can genuinely do.
doesn’t save conversation history
also, the app is significantly less useful than other ai since it can’t do web search
Models keep freezing
Not worth it
Response from developer
Check the update history yourself, it's public on the App Store listing. Qwen3 4B Heretic shipped in January and there were other releases this year, so the no-updates-in-a-year part isn't right. We're also not an MLX wrapper like Locally and LM Studio. Private LLM ships OmniQuant and GPTQ quantization tuned per model, and our 3-bit OmniQuant matches or beats 4-bit RTN on the same hardware. No account, no login, no prompt logs, nothing collected (unlike Locally and LM Studio) Private LLM 2 is next and it's free if you own v1. The private beta is running on our Discord: privatellm.app/discord. If it's still not for you, please claim the refund. App Store purchase history, then Report a Problem. privatellm.app/faq has the details. No point sitting on an app that doesn't do what you need.
C’est pire que prévu ! À faire tant que rien est fait!!!
Response from developer
« Ça chauffe » : c'est exactement ce que fait l'IA locale. Le modèle tourne sur le processeur de votre iPhone, pas sur nos serveurs. C'est votre appareil qui fournit le calcul — il chauffe donc, et il est limité par sa RAM. Aucune app locale n'y échappe. C'est aussi pourquoi les réponses tombent à côté. Sur un iPhone ancien, seuls les plus petits modèles sont proposés, et un modèle de 1B ne raisonne pas comme un 8B. Si votre appareil a basculé sur le modèle intégré après un échec de chargement, la qualité chute encore. Ouvrez la liste des modèles et choisissez le plus grand que votre iPhone accepte : la différence est nette. Les prérequis matériels sont détaillés sur privatellm.app/fr/faq, section « Appareils et modèles ». Apple n'y est pour rien — la contrainte est physique. Si votre iPhone n'est pas à la hauteur, demandez un remboursement à Apple : historique des achats → Signaler un problème. La marche à suivre est dans « Achat et remboursements ». La prochaine version majeure relèvera la configuration minimale et abandonnera les iPhone trop anciens, précisément pour éviter ce genre de déception.
