Understand the five real-time interaction patterns in Web3 and fintech apps — and the architecture behind them — in one article.

Every great product deserves an easier way to connect with its users. 💬 Love what @Knocketai is building for makers and small teams! 🚀

Tencent RTC Conversational AI now supports integration with all major full-duplex real-time speech models, with the full stack around them. First batch live now. More in the near term.

"You should not have to pick a language before you speak." Brad Mattick, Advisor at @soniox_ai , on our Voice AI panel in Menlo Park. It sounds obvious until you look at what the first generation of voice models actually asked of people: pick a language, stay in it, and hope nobody says a phone number or a brand name mid-sentence. That is not how multilingual people speak. They switch inside a single utterance, and the older systems broke completely. Full session in the video.

A model can sound human. An agent has to be good at a specific job. In our Voice AI panel with @KylanGibbs , CEO and co-founder of @inworld, we discussed what that changes: role-specific benchmarks over generic naturalness, emotional range as a design decision, out-of-distribution conversations, graph-based systems that once took hundreds of people, and an effort-to-quality curve that keeps moving. The line that stuck with me: find a high-performing human in that role, study how they interact, then build the agent to replicate what they do well. Full session in the video.

A demo proves the voice can sound human. Production proves the agent can hold a workflow. At our Voice AI panel in Menlo Park, moderated by Jean Wang, Kylan Gibbs (@inworld ), Brad Mattick (@soniox_ai ), and Jackson Fu (@TencentRTC ) worked through what sits between those two: measuring "human-like" by role-specific performance rather than voice quality, systems that clear a POC and then buckle at 100 calls over a weekend, voice data that legally cannot leave the jurisdiction where it was collected, and multilingual input where the user never has to pick a language before speaking. Thanks again to Kylan Gibbs, Brad Mattick, and Jackson F. for the excellent sharing and valuable insights. Voice AI is entering its full-stack era, and “sounding human” is now the relatively easy part.The distance between a demo and true production readiness tests the entire product system — real-time transport, multi-region deployment, and the orchestration of ASR, LLM, and TTS as one unified system. Closing that distance is the problem we at Tencent RTC are focused on solving together with our partners.

Google's two new Gemini Live models: 3.8 Live — the answer just arrives. Scheduling, order status. Extended Thinking — multi-step work across systems. Rebooking flights. Building voice agents with Gemini 3.8 Live? Try TencentRTC Conversational AI.

Big day for our partner @Inworld AI: the GA launch of their new flagship voice model family, Realtime TTS-2 and Realtime TTS-2 Flash. TTS-2 is now available in the Tencent RTC console, and we've integrated it into our joint demo over our global realtime network. Try it in our joint demo today: trtc.io/demo/homepage/#/inwo…

It’s a wrap on Building the AI-Native Startup Stack in Menlo Park, and what a turnout. Full house, great conversations, and a demo area packed with founders, engineers, and AI builders exploring what’s next for production-ready AI applications. A big thank-you to everyone who joined our Voice AI panel, Advancing Voice AI: Emerging Capabilities and the Next Wave of Applications, and to our panel speakers: Jean Wang, Senior Product Manager, @tencentcloud — Moderator Kylan Gibbs, CEO & Co-founder, @inworld Brad Mattick, Advisor, @soniox_ai Jackson Fu, Senior Solution Architect, @tencentcloud The discussion covered where voice agents are heading next, from real-time interaction and multilingual AI experiences to the RTC, ASR, TTS, LLM, and agent orchestration stack needed to build more natural and responsive voice products at scale. Our demo area was also packed throughout the event. It was great to see builders testing live voice agent workflows, asking technical questions, and exploring how to bring real-time AI experiences into production. Thank you to our co-hosts, partners, speakers, and everyone who joined us in Menlo Park. #VoiceAI #RealtimeAI #AIStartups #TencentRTC #TencentCloud #InworldAI #Soniox

Join us at Building the AI-Native Startup Stack on Aug 14 in Menlo Park, California. Tencent RTC will be part of the event’s Voice AI track, featuring live demos and a panel discussion on how Voice AI is moving from early experimentation to production-ready applications. Our panel, Advancing Voice AI: Emerging Capabilities and the Next Wave of Applications will bring together leaders from @TencentRTC , @inworld , and @soniox_ai to discuss the next wave of voice agents, real-time interaction, multilingual AI experiences, and the infrastructure needed to scale Voice AI products. Register here: luma.com/aug14_ainative

We're genuinely happy to see Knocket helping builders around the world connect with their users Developers in different countries, building for completely different industries — AI tools, rental platforms, logistics, travel blogs, consulting. Same one line of code, same result: their visitors can finally reach them. 🤖 AI tool developers — visitors ask questions right on the product page. No docs, no Discord. 🏠 House rental platforms — tenants message from the listing page instead of filling out forms. 📦 International logistics companies — clients check shipment details through live chat. ✈️ Travel bloggers — readers reach out from the blog post they're reading. 💼 Business consultants — potential clients start a conversation before booking a call. What is Knocket? A 100% free, zero-barrier contact toolkit. One line of code adds live chat to any website — so the people visiting your site can actually reach you, in real time. 💬 Web Widget — one line of code. Works on WordPress, Shopify, Webflow, Framer, Next.js, and more. 📱 Mobile Widget — embed in iOS, Android, Flutter, or React Native via WebView. 🔗 Contact Page — a shareable link with live chat and social links. No website needed. 📥 Unified Inbox — Telegram, email, and live chat all in one place. Reply anywhere, and the message goes back through the original channel. 📱 Mobile Widget — embed in iOS, Android, Flutter, or React Native via WebView. Free. No strings. Just a real connection between you and your users. If you're building something people use, Knocket is ready. → knocket.trtc.io/

See you at IBC — drop by and let's explore what AI and real-time interaction make possible.