Mirai

Convert, optimize, distribute & run your models with the fastest inference engine on Apple Silicon

Mirai Labs builds uzu, an inference engine written in Rust that runs large language models on iPhone, iPad, and Mac. The tagline sums up the workflow: convert, optimize, distribute, and run your models on Apple Silicon. Alongside the engine, the company publishes 4-bit and 8-bit versions of Qwen, LFM, and Muse-Glimmer models, with benchmarks against MLX and llama.cpp.

The product lineup on the site includes the inference engine, a model conversion tool, a macOS app, and a CLI tool, and Android inference is marked as coming soon. uzu is on GitHub, and docs, metrics, and research pages are available. The stated goal is on-device AI that is interactive, ambient, and continuous, with the team targeting speeds around 1,000 tokens per second so that models can act as interfaces rather than chatboxes.

This directory places Mirai under chatbots and assistants, but it is really infrastructure for developers who want to ship on-device models. Pricing is listed as a free tier with paid plans; the homepage shows the open-source engine and a contact form rather than a price list, so reach out through the site if you need commercial terms. Developers building private, low-latency AI features for Apple devices should look at the uzu repository first.

Category: Chatbots & Assistants. Pricing: freemium. Visit website

Alternatives to Mirai in Chatbots & Assistants

• HeyAIBuddy AI Girlfriend — Chat with an AI girlfriend by voice or text. Free to start, safe-for-work
• Jev Knows — Free AI yes/no answers for any "Should I…?" decision
• AutoReception — 24/7 AI receptionist for Australian service businesses
• HeyAIBuddy AI Boyfriend — Chat with an AI boyfriend by voice or text. Free to start
• Avenyora — AI self-reflection for astrology, tarot, numerology and journaling
• Cognira — Better thinking, better answers.