Pocket AI — No Internet | Local LLM Chat

Why Pocket AI

Built different. On purpose.

Every design decision keeps AI inference local while making the free and Pro tiers clear.

Offline. Always.

Download a model once over Wi-Fi. After that, chat runs locally anywhere — a plane, a subway, a hospital, anywhere with zero signal.

Works in airplane mode
No internet needed for chat
No cloud AI inference

Private. Not just promised.

Your prompts, responses, and chat history stay on your iPhone. Usage analytics is content-free and can be switched off in Privacy settings; RevenueCat handles the optional Pro purchase without receiving your conversations.

No cloud AI processing
Analytics never sees your chats
RevenueCat handles Pro access

Free to start. Pro when you need more.

Download free with no account or credit card. The free tier includes daily messages and one downloaded model; Pro unlocks higher limits and advanced features.

Free download
15 messages per day included
Optional Pro upgrade

Inside the app

The real thing. No mockups.

Both shots below are the shipping app on an iPhone, exactly as it appears on the App Store.

Pocket AI chat screen answering on-device at 68 tokens per second — Chat, fully on-device

Ask anything. The model answers from your iPhone's own chip, with live tokens-per-second — no cloud, no account, no signal needed.

The daily brief summarising today's calendar and reminders — A daily brief, written on your phone

Each morning Pocket AI reads your Calendar and Reminders locally and writes you a short brief. Nothing is uploaded to build it.

Older iPhones Deserve AI Too

Runs on iPhone 12. And up.

Most "on-device AI" apps only work on the latest Pro models. We built this differently. If you own an iPhone 12 or newer, you have enough compute to run real AI.

Qwen 0.6B ~20–35 tok/s ~45–60 tok/s

Qwen 1.7B ~12–20 tok/s ~25–35 tok/s

Qwen 4B needs more RAM ~10–18 tok/s

Typical decode speed on a warm Q4_K_M model. Read across a row: the same model gets faster on a newer chip. Read down a column: a bigger model gives better answers and fewer tokens per second.

✈️

Airplane Mode

Full AI. Zero signal.

🔒

No Account

Not even an email.

⚡

Metal GPU

Apple Silicon optimized.

🎛️

System Prompts

Fully customizable AI.

Open Source Models

Top models. Completely offline.

Run open-source AI models directly on your iPhone. Downloaded models run locally without an internet connection.

Q

Qwen

Alibaba's powerful multilingual models. 0.6B, 1.7B, and 4B variants available.

Available Chat Code

L

Llama

Meta's flagship family of foundation models for general reasoning and long context.

Available Reasoning

G

Gemma

Google's lightweight, state-of-the-art models. Fast and efficient on Apple Silicon.

Available Fast

D

DeepSeek

Advanced reasoning and code generation models for power users.

Available Code Reasoning

All models use Q4_K_M quantization — best quality-to-size ratio for on-device inference.

FAQ

Common questions.

Everything you need to know before downloading.

Does it work without internet?

Yes. Once you download a model over Wi-Fi, chat inference runs on your iPhone's chip instead of a remote AI server. Model downloads and optional Pro purchase or entitlement checks may use a connection when available.

Is my data private?

Your prompts, responses, attachments, and chat history stay in your phone's private app storage and are not sent to cloud AI servers. The app does send content-free usage analytics — which features were opened, whether a download finished — to PostHog and Google Analytics, and you can turn that off under Privacy settings; it never includes your chats, voice audio, Calendar events or Reminders. RevenueCat receives only the anonymous purchase and device identifiers needed for Pro purchase and entitlement checks. Full detail is in the privacy policy.

What iPhone do I need?

Any iPhone 12 or newer. Older iPhones run our lighter models (Qwen 0.6B) at perfectly usable speeds (~20–35 tok/s; an iPhone 15 Pro does the same model at ~45–60). Newer Pro models handle our larger, more capable models. You do not need the latest hardware.

Do I need to create an account?

No. There is no account, no login, no email, no sign-up. Open the app, download a model, start chatting.

Is it free?

Yes. Pocket AI is free to download with no account or credit card required. The free tier includes 15 messages per day and one downloaded model; Pro unlocks unlimited messages, unlimited models, and advanced features.

How do I get started?

Download the app from the App Store
Go to the Models tab
Tap "Download" on any model (start with 0.6B for fast results)
Once downloaded, tap "Load"
Go to Chat and start typing

Can I customize the AI's behavior?

Yes. Go to Settings → System Prompt to write custom instructions for the AI. Make it a coding assistant, a writing editor, a language tutor, or anything else. This is one of the most powerful features for power users.

How much storage does it use?

Models range from ~460 MB (0.6B) to ~2.7 GB (4B). The app itself is small. You can download and delete models freely — your chat history is preserved independently of which model is loaded.