Las Vegas, Nevada - January 13, 2026 — At CES 2026, Tuya Smart (NYSE: TUYA, HKEX: 2391), a global AI cloud platform service provider, officially unveiled Hey Tuya, the AI Life Assistant. Built on a multi-agent collaborative architecture, Hey Tuya is designed to move AI beyond screens and into real-world environments and physical spaces.

Simultaneously, Tuya unveiled its self-developed Physical AI Engine (PAE) system architecture. As the foundational engine powering Hey Tuya, PAE delivers a stable, secure, and evolvable system framework and directly addresses the complex challenges AI hardware faces when deployed in physical environments. Through a comprehensive architectural redesign, PAE enhances real-world user experience while significantly strengthening data security, privacy protection, and compliance capabilities.

With PAE, developers can significantly lower the barriers to AI hardware development—reducing R&D costs while building safer, more reliable, and more stable AI-powered devices. This accelerates the entire journey from proof of concept to large-scale deployment. More importantly, PAE enables large AI models to transition from the digital world into physical spaces—deeply integrating into every device and every scenario to continuously improve and solve real-world problems. The concept of “AI for the Real World—Every Device, Every Space” is becoming a reality.

OmniMem: Individual Long-Term Memory Engine Redefining AI Memory

To address the industry-wide challenge of AI systems that “forget” after each conversation and lack emotional continuity, PAE integrates Tuya’s proprietary OmniMem Individual Long-Term Memory Engine.

OmniMem is not simply a conversation log storage system. It is a memory hub with temporal reasoning, dynamic updates, and cross-device collaboration capabilities, enabling AI to build a persistent, evolving individual memory system for real-world usage scenarios.

Compared with mainstream open-source memory solutions, OmniMem achieves significant improvements across multiple key metrics through architectural optimization and algorithmic innovation:

• Memory: OmniMem supports both short-term and long-term memory, ensuring real-time responsiveness while achieving long-term retention of high-value information and enabling appropriate forgetting.

• Response Efficiency: OmniMem reduces long-term memory retrieval latency to the millisecond level, breaking the common second-level delay bottleneck of open-source solutions and enabling stable real-time dialogues and continuous interactions.

• Accuracy: OmniMem Pro achieves over 80% accuracy on the open-source LoCoMo benchmark dataset, reaching industry-leading performance and effectively mitigating memory confusion in multi-turn and long-term conversations.

OmniMem supports AI in truly understanding and continuously remembering user preferences, habits, and interaction history—forming a humanized emotional memory foundation for Hey Tuya and enabling continuous self-evolution.

AD-RTN & T-RTC: Real-Time Communication Engines Accelerating AI Interactions

Leveraging years of global technical expertise, Tuya has built and deployed the AI-Device Real-Time Network (AD-RTN)—a global edge acceleration network covering major regions worldwide. Built on AD-RTN infrastructure, Tuya developed Tuya Real-Time Communication (T-RTC), a highly reliable, low-latency, and resilient real-time communication system that enables seamless interactions between AI, devices, and users.

Designed specifically for complex network environments, T-RTC maintains stable performance even under extreme conditions with 30–70% packet loss, ensuring smooth and continuous interactions across mobile networks and weak connectivity scenarios.

Powered by Tuya’s seven global data centers, real-time media acceleration networks across dozens of countries and regions, and intelligent scheduling across 1,300+ edge nodes, T-RTC delivers: global average latency below 86 ms; audio/video stalling rate under 0.8%; and overall service availability as high as 99.9%. This provides a solid real-time communication foundation for large-scale AI interactions.

DOA: Intelligent Agent Orchestration Engine Empowering Developers for AI Applications Deployment

To further enhance development efficiency and system controllability, PAE introduces an innovative intelligent orchestration engine, the Dynamic Orchestration Agent (DOA), designed for rapid AI application development and iteration.

With DOA’s visual orchestration interface, developers can, through simple drag-and-drop operations, quickly build, validate, and optimize complex AI workflows by combining modules such as speech recognition, intent understanding, large-model reasoning, knowledge base retrieval, and device control.

Like building blocks, developers can easily construct multi-agent collaborative processes. DOA integrates shortest-path decision-making and multi-path parallel execution mechanisms to ensure maximum efficiency while maintaining flexibility and scalability.

With DOA, AI hardware solution validation cycles can be shortened from months to as little as one day, significantly reducing development and innovation barriers.

In addition, DOA supports standardized MCP and A2A protocols, allowing customers to integrate proprietary models, knowledge bases, cloud services, and on-device capabilities into a unified end-to-cloud system. Tuya’s proprietary Adaptive Expert System (AES) dynamically schedules and constrains agents and MCP capabilities—enhancing system reliability and timeliness, and significantly reducing hallucination risks in complex scenarios.

Full-Stack AI Capability Chain for Diverse Scenarios

To support natural voice and multimodal interaction needs across different chipsets, hardware forms, and application scenarios, PAE integrates a full-stack AI capability chain, offering developers flexible, open, and customizable model combinations for rapid deployment and scalable growth:

• Conversational AI Engine: This engine integrates high-precision Voice Activity Detection (VAD) with 300 ms “instant interruption” and 800 ms “graceful interruption” capabilities. Automatic Speech Recognition (ASR) covers over 60 languages, delivering industry-leading accuracy for major languages. Text-to-Speech (TTS) supports over 60 languages, over 300 voices, and emotional enhancement. The engine also supports multimodal end-to-end models, selectable based on business needs.

• Vision AI Engine: This is an end-to-end visual understanding engine for AI hardware, supporting image recognition, content analysis, and scene understanding—enabling AI cameras, companion robots, and smart frames to truly “see” and understand the physical world.

• IoT Intelligence Engine: A centralized intelligence hub for physical device communication and control, this engine unifies product and firmware management, device panels, group management, behavior analysis, and multi-protocol adaptation—providing a reliable foundation for efficient interaction between AI technology and the real-world.

With the debut of Hey Tuya at CES 2026, Tuya’s PAE system moves from behind the scenes to center stage—demonstrating how it drives AI hardware evolution from single-point connectivity to autonomous intelligence, and from conversational AI to proactive, scenario-driven services, truly making the intelligent experience of “AI everywhere” within reach.