Pinku

About Pinku

Pinku is a private AI chat application for iPhone, iPad, and Mac that runs open models locally using Apple's MLX framework. Users download a model once from Hugging Face and can then chat offline, including in airplane mode, with no account required. The app's core mechanic is that conversations never leave the device-only model downloads, the model list, and Pro purchases generate network calls.

Review

Pinku addresses a specific need: asking an AI about health questions, personal drafts, or notes about people without the data going to a cloud service. It does this by shifting the entire inference workload onto the user's own hardware. The trade-off is that the available models are smaller and less knowledgeable than cloud-based alternatives, which the maker acknowledges directly.

Key Features

• Runs more than a dozen open models locally, from Qwen3 0.6B up to Qwen3 8B, using Apple's MLX.
• Functions fully offline after a one-time model download, with no account creation or server-side processing of chats.
• Pinku Pro unlocks larger models, custom instructions, on-device dictation, and a read-aloud function.
• The free tier includes the Llama 3.2 1B model with no time restrictions.

Pricing and Value

Pinku is free to start with the Llama 3.2 1B model. Pinku Pro is an optional paid upgrade that enables access to larger models and additional features like custom instructions, on-device dictation, and read-aloud. The maker has not published specific Pro pricing in the available materials. The free tier doesn't require payment details or an account.

Pros

• Chats stay on-device with no server processing, confirmed by a published list of every network call the app makes.
• Works without an internet connection after the initial model download, which is practical for flights.
• Model selection spans multiple open-source families, giving users a choice between speed and capability.

Cons

• Smaller local models have less factual knowledge and reasoning depth compared to large cloud-based language models.
• Pro feature pricing is not yet defined in public documentation, making the full cost unclear.
• The app is not well suited for users who need up-to-date information from the web or the broad knowledge base of a GPT-5-class system.

Pinku fits users who prioritize privacy for sensitive prompts and can work within the limits of smaller on-device models. It's practical for drafting, rewriting, brainstorming, and explaining concepts while offline. Anyone who regularly pastes personal or health-related text into AI tools and hesitates about where that data goes will find the local-only approach directly relevant.