Pinku: Private Local AI Chat1.0.8

Pinku is a private AI chat app that runs open-source language models entirely on your iPhone or Mac. Download a model once and chat with it offline. Your messages are processed on your device and are never sent to a server. No account. No subscription required.

HOW IT WORKS
Pick a model, download it once and start chatting. It runs locally with Apple's MLX framework on your device's GPU. After the download, Pinku works in airplane mode or anywhere without a connection.

MODELS
Pinku includes 12 open models, all 4-bit or 8-bit and sized for phones and laptops:
. Qwen3 0.6B (0.4 GB)
. LFM2 1.2B (0.7 GB)
. Llama 3.2 1B (0.7 GB)
. Gemma 3 1B QAT (0.8 GB), the default, strong in many languages
. Qwen3 1.7B (1.0 GB)
. DeepSeek-R1 Distill Qwen 1.5B (1.0 GB)
. SmolLM3 3B (1.8 GB)
. Llama 3.2 3B (1.8 GB)
. DeepSeek-R1 Distill Qwen 1.5B 8-bit (1.9 GB)
. Qwen3 4B (2.3 GB)
. Gemma 3n E2B (2.6 GB)
. Qwen3 8B (4.7 GB)
Qwen3, DeepSeek-R1 and SmolLM3 are reasoning models that think step by step before they answer. Pinku only lists the models your device has enough memory to run, and it checks available memory before loading a model so the app does not crash.

FEATURES
. Chat history saved only on your device; delete any chat at any time
. Copy, share or regenerate any answer
. Edit a message you sent and get a new reply
. Share a whole conversation as plain text
. Markdown formatting for code, lists and tables
. Siri and Shortcuts: start a new chat by voice
. Accent color and font options
. App available in English, Spanish, German, French, Japanese, Korean, Simplified Chinese and Russian

PINKU PRO
An optional one-time purchase. No subscription. It unlocks:
. The larger models over 1 GB (SmolLM3 3B, Llama 3.2 3B, DeepSeek-R1 8-bit, Qwen3 4B, Gemma 3n E2B, Qwen3 8B)
. Custom instructions that shape every answer
. Voice dictation, transcribed on your device
. Read aloud, using the voices installed on your device
Restore your purchase at any time from the app.

PRIVACY
The AI runs on your device. Pinku has no account, no analytics and no server of its own that receives your chats. The only time Pinku uses the internet is to download a model you choose, from Hugging Face. Purchases are handled by Apple.

FAQ

Is there a private AI assistant for iPhone that works offline?
Yes. Pinku runs models like Llama, Qwen, Gemma and DeepSeek-R1 directly on your iPhone or Mac, so your conversations stay on your device and it works without internet once a model is downloaded. On-device models are smaller than cloud models, so they are best for writing, questions, brainstorming and everyday help.

Does it work without internet?
Yes. You need a connection only to download a model. After that, every chat runs offline.

Is it free?
Downloading Pinku and chatting with the six models up to 1 GB costs nothing. No account and no subscription are required. Pinku Pro is an optional one-time purchase for the larger models, custom instructions, dictation and read aloud.

What models does it have?
12 open models: Qwen3, Llama 3.2, Gemma 3, Gemma 3n, DeepSeek-R1 Distill Qwen, SmolLM3 and LFM2, from 0.4 GB to 4.7 GB.

Is my data sent anywhere?
No. Your prompts and answers are processed on your device and stored only there. Pinku does not collect them and does not use analytics.

REQUIREMENTS
iPhone with iOS 18.6 or later and Metal 3 support, or a Mac with Apple silicon running macOS 14 or later. Larger models need more memory.

Website: https://aisagalab.com/pinku.html
Privacy Policy: https://aisagalab.com/privacy.html
Terms of Service: https://aisagalab.com/terms.html
Terms of Use (EULA): https://www.apple.com/legal/internet-services/itunes/dev/stdeula/

Version History Bug fixes and performance improvements. 1.0.8 4d ago Bug fixes and performance improvements. 1.0.7 6d ago Bug fixes and performance improvements. 1.0.6 Sep 10 Bug fixes and performance improvements 1.0.5 Aug 29 Bug fixes and performance improvements. 1.0.4 Feb 10 Bug fixes and performance improvements. 1.0.3 Jan 31 Bug fixes and performance improvements 1.0.2 Jan 21 Bug fixes and performance improvements 1.0.1 12/15/2025 1.0.0 11/05/2025