Pinku

Overview

Pinku is an on-device artificial intelligence chat application developed by Saga Ltd. for iOS and macOS platforms. The software executes open-source large language models locally utilizing Apple's MLX framework and device GPUs, eliminating the need for cloud servers or user accounts. It enables users to process prompts, inspect photos, and analyze PDF documents entirely offline once models are downloaded.

Key Features

• Local model execution — Language models run directly on-device using Apple's MLX framework and Metal 3 architecture, ensuring conversation data never leaves the hardware.
• Open model library — Users can download up to 11 quantized 4-bit and 8-bit models ranging from 0.4 GB to 4.7 GB, including Llama 3.2, Qwen3, SmolLM3, and Gemma architectures.
• Multimodal document analysis — The application supports local querying and interaction with user-provided images and PDF files without cloud processing.
• Dynamic memory verification — Pinku evaluates available system RAM before loading a chosen model to prevent memory exhaustion and application crashes.
• Apple platform integrations — The software supports hands-free operation and automation via native Siri triggers and Apple Shortcuts.
• Local data management — Chat histories are retained solely on local storage, featuring support for plain-text exports, Markdown rendering, message editing, and response regeneration.
• On-device speech tools — Dictation and text-to-speech audio playback function using locally installed system voices without external network calls.

Who It's For

Pinku is designed for privacy-focused professionals, corporate researchers, and field workers who require generative AI assistance in air-gapped or restricted network environments. It serves organizations in compliance-heavy industries—such as legal, finance, and healthcare—where strict regulatory rules prohibit transmitting sensitive documents or proprietary queries to third-party cloud APIs. It also serves mobile professionals who require reliable AI productivity tools while traveling or operating without internet access.

Notable Strengths

Pinku distinguishes itself through a strict zero-server privacy architecture that eliminates cloud latency, subscription locks for baseline models, and off-device data exposure. Its support for 11 distinct quantized open-source models allows users to balance resource consumption and reasoning capability based on their specific Apple silicon hardware limits. Furthermore, offering both a functional free tier and a one-time lifetime purchase option delivers cost predictability over typical recurring SaaS AI tools.