Clerk AI 如何藉助 LiveKit 大規模處理數百萬段企業對話

Clerk AI is an enterprise conversational AI platform for voice and messaging agents. The company enables businesses to deploy AI agents that make and receive calls, send and respond to SMS, RCS, and WhatsApp messages at scale. Founded by Alexander Haque, Igor Boshoer, and Will Chertoff, Clerk AI serves enterprises that need to handle hundreds of thousands of calls and millions of messages daily while maintaining natural conversation quality and built-in compliance.

"Low-latency voice is existential for conversational AI. If there's a noticeable pause, callers hang up. We needed infrastructure that could deliver sub-500ms response times at enterprise scale."

Will Chertoff, Co-Founder, Clerk AI

Enterprise customers demand both performance and reliability. Long hold times and rigid automation scripts create friction for customers and limit conversion rates. Clerk AI saw an opportunity to solve this with truly conversational AI agents, but delivering natural-feeling voice experiences at scale required robust real-time communications infrastructure that could handle thousands of concurrent calls with minimal latency.

The platform's agents are purpose-built for outbound marketing and customer engagement at scale. They detect intent and sentiment in real time, integrate with CRMs like Salesforce and HubSpot, book meetings, qualify leads, and transfer to human teams when needed. The company's proprietary ScreenSense technology detects call screening on both Apple and Android devices with 97% accuracy, while TrueReach voicemail detection exceeds 98% accuracy.

From the start, the team required complete control over their stack, including the ability to choose and swap out speech-to-text (STT), large language models (LLMs), and text-to-speech (TTS) providers. Native SIP support was essential for phone system integration, and the platform needed to scale globally without compromising performance or increasing operational complexity.

LiveKit's open-source framework and global edge network provided the foundation Clerk AI needed. The framework abstracted low-level WebRTC complexities while still giving the team transparency and control over voice pipeline behavior. LiveKit Cloud's distributed architecture ensured consistent low-latency media transport between agents and callers, regardless of geographic location.

"We evaluated multiple platforms, but LiveKit was the only one that could deliver the combination of sub-500ms latency, 1,000+ concurrent call capacity, and the flexibility to bring our own AI models. The open source framework meant we could customize anything we needed."

Will Chertoff, Co-Founder, Clerk AI

The company's agent platform now handles over 1,000 concurrent voice sessions without pre-provisioning, with natural conversation quality maintained through InstantRecall caching and edge-optimized inference. Agents respond in under 500 milliseconds, eliminating the awkward pauses that cause caller drop-off. The platform supports 30+ languages natively, with automatic language detection and switching mid-conversation.

Today, Clerk AI's customers deploy voice and messaging agents that drive measurable business outcomes across sales, marketing, and customer success. The platform's SOC 2 Type II certification, combined with built-in TCPA compliance, enables enterprise adoption without security or regulatory concerns.

"LiveKit lets us focus on building the best conversational AI platform rather than managing real-time infrastructure. When we need to scale, it just works. When a new model provider emerges, we can integrate it without rearchitecting our voice pipeline."

Will Chertoff, Co-Founder, Clerk AI