LiveKit is emerging as a key infrastructure provider for real-time AI voice agents, with this week’s news centered on deep integrations with Google’s Gemini and xAI’s Grok models. The company highlighted its open-source WebRTC stack and Agents SDK at Google Cloud Next 2026, focusing on turning Gemini-based prototypes into production-ready voice agents.
Claim 55% Off TipRanks
- Unlock hedge fund-level data and powerful investing tools for smarter, sharper decisions
- Discover top-performing stock ideas and upgrade to a portfolio of market leaders with Smart Investor Picks
LiveKit’s sessions at Google Cloud Next, co-presented by engineer Jesse Hall, showcased how its platform manages the real-time layer, audio transport, device compatibility, and scaling for Gemini Live and the Interactions API. By abstracting the media infrastructure, LiveKit aims to let developers deploy agents across phones, browsers, embedded devices, and robots without rebuilding communications layers.
The company also demonstrated a Gemini-powered VIP concierge application in the AI Garden’s Startup Hub, where a multimodal agent generates personalized conference agendas and keepsake videos. In this demo, Gemini handles reasoning while LiveKit delivers low-latency connectivity and media transport, underscoring its role as the real-time backbone for interactive AI experiences.
In parallel, LiveKit promoted an observer-pattern architecture for safety and compliance in voice agents, decoupling conversational logic from guardrail enforcement. A secondary large language model asynchronously reviews user turns and injects corrective context, with design features such as limited violation injections and concurrency controls implemented via LiveKit Agents.
Beyond Google’s ecosystem, LiveKit expanded support for xAI by enabling an end-to-end voice pipeline that combines xAI speech-to-text, Grok as the LLM, and xAI text-to-speech under a single LiveKit API key. The company emphasized a cascaded pipeline architecture for greater control, debuggability, and observability, along with support for tool calling like web_search, x_search, and code_interpreter.
Collectively, these developments reinforce LiveKit’s strategy to serve as a neutral, multi-model orchestration and real-time communications layer for AI voice applications. Increased alignment with hyperscalers and emerging model providers, plus a focus on safety and enterprise-grade control, may strengthen its competitive position and support usage-based growth as AI voice agents move into production. Overall, it was a visibility-rich and technically focused week for LiveKit.