Mirai
Category: AI Infrastructure
An on-device AI inference platform and SDK that enables developers to deploy and run AI models directly on Apple devices (iPhone, iPad, Mac) with zero latency, full privacy, and zero inference costs, eliminating cloud dependency. Mirai was founded in 2024. The company is led by Dima Shvets (Co-CEO). Based in San Francisco, United States. Team size: 11-50. Total funding raised: $14.0M. Latest round: Seed. Key investors include Uncork Capital, Garuda Ventures, Samsung Next, Sequoia Capital Scout Fund, Index Ventures Scout Fund, Z Venture Capital, Sarah Smith Fund, Allison Pickens Fund, David Singleton, Francois Chaubard, Marcin Zukowski, Mati Staniszewski.
- Founded
- 2024
- Headquarters
- San Francisco, United States
- Team size
- 11-50
- Total funding
- $14.0M
Value proposition
Enables near-zero latency AI inference, complete data privacy, predictable costs at scale, and offline capability by running models directly on Apple Silicon devices, eliminating per-request cloud computing costs and network latency while maintaining production-grade performance.
Products and solutions
uzu Inference Engine - High-performance Rust-based inference engine optimized for Apple Silicon (Swift/TypeScript/Python/Rust bindings), Model Optimization & Conversion Tools (lalamo) - Model conversion, quantization, and optimization, Mirai Quantization (Mirai-M 4-bit / Mirai-L 8-bit) - Co-designed quantized checkpoints for local LLMs on Apple Silicon (Qwen3.5 series), Local models catalog - Optimized models (Qwen, Llama, Gemma, GPT-OSS, LiquidAI LFM, etc.) for on-device run, Hybrid Cloud-Device Routing - Intelligent routing between local and cloud inference based on workload requirements
Unique value
Built entirely from scratch specifically for Apple Silicon architecture, not a wrapper around existing inference stacks. Owns the complete stack from model optimization and execution to memory management and deployment. Achieves up to 37% faster generation and 59% faster prefill compared to MLX and llama.cpp while remaining production-ready for real applications.
Target customer
Software developers, tech startups, and consumer technology companies building AI-powered applications who need low-latency, private, and cost-effective inference without cloud infrastructure dependencies
Industries served
Mobile Applications, Consumer Technology, AI-Powered Software, Enterprise Applications, Developer Tools
Technology advantage
Hardware-aware execution eliminates cross-platform abstraction tax, delivering predictable performance under real production constraints. Combines optimized inference, proprietary models, and deployment into a unified on-device foundation. Enables developers to integrate high-performance AI with minimal code changes while maintaining zero inference costs and complete data privacy.
How they differentiate
Built entirely from scratch for Apple Silicon, owns complete inference stack (uzu engine, model optimization tools, SDK, hybrid routing). Achieves up to 37% faster generation and 59% faster prefill compared to MLX and llama.cpp while remaining production-ready. Provides developer-first SDK for easy integration.
Main competitors
Apple MLX, llama.cpp, Baseten
Key partnerships
Uncork Capital (lead investor), Andreessen Horowitz (backing/support), Frontier model providers (architecture optimization collaborations), Chip manufacturers (hardware compatibility partnerships), Angel investors including David Singleton, Francois Chaubard, Marcin Zukowski, Mati Staniszewski
Major milestones
Founded in 2024 by creators of Reface (300M+ users) and Prisma (100M+ MAU, App of the Year), $4M pre-seed (Jul 2025) from Sequoia Scout Fund, Index Ventures Scout, Samsung Next, Z Venture Capital and angels, Launched developer-first SDK / uzu engine for on-device AI on iOS and macOS in July 2025, Secured $10M seed led by Uncork Capital (Feb 2026), Released Mirai Quantization (Jun 2026): co-designed 4/8-bit Qwen3.5 checkpoints; claims 40–60% more tokens/s vs Unsloth/llama.cpp and MLX at same quality, Team ~16; rebranded public identity as Mirai Labs (frontier on-device AI lab)
Growth metrics
Production-ready platform launched Jul 2025; ~16-person senior team; inference site claims up to 38% faster prompt processing vs MLX; Jun 2026 Mirai Quantization claims 40–60% more tokens/s at same quality vs Unsloth/llama.cpp and MLX on Apple Silicon
Market positioning
Early-stage developer tools provider for on-device AI inference, targeting iOS/macOS app developers seeking to eliminate cloud costs and latency. Positioned as 'Stripe for on-device inference' - simple integration layer.
Geographic focus
United States (San Francisco), United Kingdom (London), Global developer market
Patents and IP
No patents publicly disclosed for Mirai Tech (trymirai.com) as of February 2026
About Dima Shvets (Co-CEO)
Co-founder of Reface (AI face-swapping app with 200M+ users, backed by Andreessen Horowitz). Pioneer in Generative AI with over 300M total users across projects. Scout at Andreessen Horowitz (a16z). Board member at Endeavor. Successfully scaled consumer AI applications through hyper-growth phase with real-time face swap technology.
Latest news about Mirai
- Italian startup Mirai Robotics secured $4.2M pre-seed funding to develop autonomous surface vessels targeting the $2.5T blue economy, one of the largest rounds in Italy's robotics sector. With 80% of
- Mirai Tech has secured $10 million in seed funding led by Uncork Capital to simplify on-device AI deployment, enabling developers to tap smartphone and laptop NPUs with just a few lines of code. Found
- London-based Mirai just secured $10M seed funding led by Uncork Capital to accelerate on-device AI optimization amid surging cloud costs. Their Rust-based inference engine delivers up to 37% faster pe
More AI Infrastructure companies
Official website: https://trymirai.com/
