SHANGHAI, July 18, 2026 — iFLYTEK launched GuideX on Friday, an intelligent interaction agent for public service environments, at the World Artificial Intelligence Conference (WAIC) 2026.

Conventional digital humans answer questions and stop there. GuideX is built to complete tasks. It rests on three capabilities the company calls Omnimodal, Self-Regulation and Empathy-Driven, forming a single chain that spans perception, understanding, execution, memory and emotion.

“iFLYTEK has opened one new AI product category after another,” said Roy Lu, General Manager of Overseas Products and Solutions at iFLYTEK. “Public services are the next. GuideX is a new paradigm of interaction — an agent that does not stop at answering, but perceives, decides and gets the task done.”

Omnimodal perception addresses the hardest problem in a public venue: not whether AI can hear, but whether it can tell who is speaking, from where, and why. Instead of a single voice channel, GuideX fuses voice, face, position, lip movement and spatial context. With several people speaking at once, word loss stays below 1.3% and cross-talk below 1.7%; at -10dB noise, speech recognition accuracy reaches 92.1%. GuideX supports 30+ languages, with 97.35% English speech recognition accuracy and 95.12% end-to-end accuracy, alongside 200+ avatars and 100+ ultra-realistic voices. The full chain, from wake-up to avatar response, takes 0.42 seconds even under heavy concurrent load, and full-duplex communication lets users interrupt and follow up at will.

Self-regulation takes over once a real transaction begins. GuideX extracts business constraints from vague requests and fills in what the user leaves unsaid. Asked how to prove savings for a visa application, it identifies the underlying task — applying for an asset certificate — instead of matching keywords. A dual-track mechanism routes standard requests to deterministic policies and complex ones to adaptive reasoning, coordinating agents across enterprise systems to close the loop from inquiry to completion. Those actions come from SkillHub, iFLYTEK’s skills platform, where capabilities such as route planning, duty-free guidance and mall navigation are packaged as reusable skills that partners can apply, extend and co-create. SkillHub now carries more than 10,000 skills, extensible without code.

Empathy-driven interaction runs through all of it. GuideX reads emotional context and responds in kind, keeping its language steady and appropriate. With user consent, it combines face recognition with short-term scene memory to retain service context, preferences and past transactions, so returning users pick up where they left off rather than starting over.

GuideX has already been deployed across Southeast Asia, the Middle East, Central Asia, Latin America and Africa. It supports custom avatars, voices and interaction flows, and runs on smart terminals, all-in-one machines, transparent displays, web and mobile apps.

iFLYTEK also launched the GuideX Navigator Program at WAIC with eight ecosystem partners joining at launch, and said it would continue developing interaction and agent technologies with them across government services, transportation, hospitality, tourism and smart commerce.