The best offline AI apps for iPhone: AI chat that runs on the phone

Several iPhone apps run AI language models on the phone itself, so they keep working in airplane mode and your chats stay on your device. The main ones as of November 2026 are Modelbase, PocketPal AI, Private LLM, Locally AI, Liquid Apollo and Enclave. Most are free; Private LLM costs $4.99 once; some offer optional paid extras. They differ in how many models they offer, whether they understand photos, voice and documents, and whether they can also connect to your own server or the cloud.

We make Modelbase, so we’ve held its entry to the same standard as the others: every fact comes from the app’s App Store page or official site as of November 2026, with links. Prices and features change, so check the App Store before you decide.

The apps at a glance

App	Price (Nov 2026)	Requires	Photos	Voice	Documents	Also connects to	Privacy label
Modelbase	Free	iOS 18	Yes	Yes, speech recognition on device	PDFs and documents	OpenRouter, your own server	Data Not Collected
PocketPal AI	Free, in-app purchases	iOS 15.1	Yes	Yes; hands-free mode with Plus	Not stated	OpenAI-compatible servers	Data Not Collected
Private LLM	$4.99 once	iOS 17, A12 or later	No (text only today)	Through Siri	Not in the current version	Not stated	Data Not Collected
Locally AI	Free	iOS 18.1	Yes	Yes, on device	Not stated	LM Studio (LM Link)	Data Not Collected
Liquid Apollo	Free	iOS 18	Through OpenRouter or your server	Read-aloud	Not stated	OpenRouter, LM Studio or Ollama	Data Not Collected
Enclave	Free; Pro $9.99/month	iOS 27	Yes	Yes, on device	PDFs, text files, code	Cloud models with Pro	Data Not Linked to You

“Privacy label” is what each app’s App Store page shows under App Privacy.

Modelbase

Modelbase is our free app for iPhone and iPad (iOS 18 or later) and Macs with Apple silicon. It runs models with Apple’s MLX and comes with LFM 2.5 (1.2B) by Liquid AI installed, so it answers offline from the first launch, with no download and no account. Its catalog has 53 open models, including Qwen 3.5, Gemma 4, Llama 3.2, Phi 4 Mini, Granite 4.0, Ministral 3, DeepSeek R1 and vision models, each with its download size and the iPhones it’s recommended for.

• Offline: chat, questions about photos (with a vision model), PDFs and documents, voice mode with speech recognition on the device, and your calendar and reminders.
• When online, if you want: web search through DuckDuckGo with numbered sources, deep research, and live cards for weather, stocks and places.
• Your own models: connect OpenRouter with your own key, or a server on your network such as Ollama or LM Studio.
• Also: thinking levels, assistants with custom instructions, iCloud sync through your own iCloud, 15 languages.

Limits worth knowing: the catalog is curated, so you can’t load any model from Hugging Face, and it doesn’t run a server for other apps.

PocketPal AI

PocketPal AI is open source (MIT, GitHub) and runs GGUF models with llama.cpp. You can search Hugging Face for GGUF files from inside the app, which gives it a very wide choice of models. It supports vision models, voice input, web search (You.com without a key, or Brave, Tavily, Exa or Firecrawl with yours), Shortcuts, and connections to OpenAI-compatible servers such as LM Studio and Ollama. It’s also on Android.

It’s free to download. As of November 2026, the App Store lists in-app purchases including PocketPal Plus, a one-time purchase offered at several prices from $3.99 to $44.99, which adds features such as a hands-free voice mode and a local API server for other apps, plus optional add-on assistants. It needs iOS 15.1 or later.

Private LLM

Private LLM is a paid app: as of November 2026, $4.99 once, with no in-app purchases, and one purchase covers iPhone, iPad and Mac and works with Family Sharing (up to six people) (site). It uses an engine based on mlc-llm with OmniQuant quantization; the site lists more than 140 models, including DeepSeek R1 Distill, Phi 4, Qwen 3, Llama 3.2 and Gemma 3. It works with Siri and Shortcuts and can export chats to Markdown or PDF.

The current App Store version supports text-based models only; vision, documents and web search are in a beta of “Private LLM 2” on TestFlight. It needs iOS 17 and an A12 Bionic chip or later.

Locally AI by LM Studio

Locally AI is free and now part of LM Studio. It runs models such as Llama, Gemma, Qwen, DeepSeek and LFM on the phone with Apple MLX, analyzes images, and has an on-device voice mode plus Siri, Shortcuts, Control Center, Lock Screen and Action Button access (site). Its LM Link feature connects to LM Studio running on your computer, so you can use bigger models from there. It needs iOS 18.1 or later.

Liquid Apollo

Liquid Apollo is free and made by Liquid AI, the company behind the LFM models. It runs Liquid AI’s small models on the device (through its LEAP platform), connects to OpenRouter with your own key, and can use a model running on your computer in LM Studio or Ollama as a “custom backend”, including image attachments for vision models. It can read answers aloud and works with Siri and Shortcuts. It needs iOS 18 or later.

Enclave

Enclave runs models such as Llama, DeepSeek, Gemma, Qwen, Phi and Mistral on the device, with on-device voice chat, chat with PDFs, text files, images and code, custom assistants, and Siri and Shortcuts. Local AI is free. As of November 2026, the Pro subscription is $9.99 a month and includes credits for cloud models from providers such as Anthropic, OpenAI and Google (pricing); web search works with cloud models. It requires iOS 27 or later. Its App Store privacy label lists data not linked to you (user content, identifiers, usage data and diagnostics).

Other apps you may come across

• MLC Chat (App Store), from the open-source MLC LLM project, runs models on the device, but as of November 2026 its last App Store update was in September 2024.
• LLM Farm (GitHub) is an open-source llama.cpp app; as of November 2026 it isn’t available on the App Store, and its README says so.
• Enchanted (App Store) is a client for an Ollama server and doesn’t run models on the phone, so it needs your computer to be reachable. More in Ollama GUI apps.

How to choose

• A model included, so it works from the first launch: Modelbase.
• Photos and voice on the phone: Modelbase, Locally AI, PocketPal AI or Enclave.
• PDFs and documents: Modelbase or Enclave.
• The widest choice of models, or open source: PocketPal AI (GGUF files from Hugging Face, MIT license).
• You prefer to pay once, and use Siri and Shortcuts on iPhone, iPad and Mac: Private LLM.
• You run LM Studio or Ollama at home and want it on your phone too: Modelbase, PocketPal AI, Liquid Apollo, or Locally AI for LM Studio.
• You want Liquid AI’s own models and app: Liquid Apollo.

Whatever you pick, the model you run matters as much as the app. A phone holds models of a few hundred million to a few billion parameters; our guide Which AI model should you run on your iPhone? helps you pick one that fits your iPhone’s memory, and Can you use ChatGPT offline? explains what works without a connection.

FAQ

Is there an AI app that works offline on iPhone?

Yes. Modelbase, PocketPal AI, Private LLM, Locally AI, Liquid Apollo and Enclave all run language models on the iPhone, so they answer in airplane mode once a model is on the phone. Modelbase comes with a model installed.

What is the best free offline AI app for iPhone?

It depends on what you need. Modelbase, PocketPal AI, Locally AI and Liquid Apollo are free to download and use on-device models; PocketPal AI and Enclave sell optional extras. Compare the table above against what you want to do.

Do offline AI apps collect my data?

Five of the six show “Data Not Collected” on their App Store pages; Enclave lists data not linked to your identity. Models that run on the phone don’t need to send your chats anywhere, but features that use the web or the cloud do.

Which iPhone do I need?

Each app has its own minimum iOS version (see the table). For the models, more memory means bigger models: many apps, Modelbase included, show which models suit your iPhone.

Can these apps understand photos?

Modelbase, PocketPal AI and Locally AI run vision models on the phone, and Enclave lets you chat about images. Liquid Apollo handles images through OpenRouter or your own server. The current Private LLM release is text only.