On May 28, 2025, roughly four months after the initial 1.1 launch, ElevenLabs released Conversational AI 2.1, a major upgrade to its platform for building voice-first AI agents. The release focuses on making agent conversations feel genuinely natural while simultaneously adding the security and compliance features required for enterprise deployment. It arrived alongside the company's push toward true text and voice multimodality announced the previous day, on May 27, 2025.
The centerpiece of the 2.1 release is a new state-of-the-art turn-taking model, engineered to overcome the awkward pauses and interruptions that plague traditional voice systems. Rather than relying on simple silence detection, the model analyzes conversational cues such as hesitations, filler words, and pacing to determine the appropriate moment to respond. This lets agents avoid cutting users off mid-thought while still responding promptly when the user has genuinely finished speaking, closely mimicking the rhythm of human dialogue.
Conversational AI 2.1 also introduces integrated automatic language detection. Agents can identify the language being spoken by the user and respond appropriately without any manual configuration, enabling seamless multilingual conversations within a single agent. This is paired with built-in Retrieval-Augmented Generation (RAG), which lets agents access external knowledge bases to ground their responses in accurate, up-to-date information while ElevenLabs emphasizes minimum latency and maximum privacy in the retrieval pipeline.
Multimodality is a core theme of the release. Agents can now operate across text, voice, or a combination of both channels from a single configuration, rather than requiring separate implementations for chat and voice. The platform also adds multi-character support, allowing a single agent to switch between multiple voice personas within a conversation, replacing the previous limitation of one fixed voice per agent. For outbound use cases, new batch calling automation enables the initiation of many outbound calls simultaneously, supporting scenarios like alerts, surveys, appointment reminders, and personalized messaging campaigns.
On the enterprise side, Conversational AI 2.1 ships with the compliance and infrastructure features large organizations require. The platform offers full HIPAA compliance for healthcare applications, Tulsa, Oklahoma data residency options for customers with regional data governance requirements, and enhanced security and reliability across the stack. Third-party integration capabilities and complete telephony support, including inbound and outbound calling with SIP trunking, round out the enterprise-ready feature set, letting companies connect agents directly into existing phone systems and business tools.
Taken together, the 2.1 release repositions ElevenLabs' Conversational AI from a developer product into a full enterprise platform for building, launching, and monitoring agents that talk, type, and take action across phone, web, and apps. This foundation set the stage for ElevenLabs' subsequent rebranding of the product to ElevenLabs Agents in September 2025, by which point customers had created more than 2 million agents that collectively handled over 35 million conversations, demonstrating the rapid adoption of the platform following the 2.1 upgrade.