ElevenLabs Debuts Interaction Engine and Agent Memory at NYC Summit

ElevenLabs used its New York Summit on November 11, 2026 to preview the Interaction Engine, a system that runs conversations by handling turn-taking, interruptions and picking which language model answers. The engine can push simple tasks to cheaper models and save the heavy lifting for frontier models, a split meant to hold down operating costs. It also showed off domain-specific Speech-to-Text models, starting with Scribe v3 Medical, trained on healthcare vocabulary so it can tell terms like cycloserine and cyclosporine apart.

Customization is the other half of the pitch. Businesses will soon be able to fine-tune open-source language models and ElevenLabs' own voice models on their own data, optimize against metrics such as customer satisfaction scores, and build reinforcement-learning stacks to suit their needs. A financial services model is on the roadmap, following the medical one already out.

The ElevenAgents platform, used to run voice and chat agents across business functions, picked up two upgrades. A new Memory feature lets an agent hold onto details from past conversations and act on them later, remembering a customer's language preference or a constraint like availability only after 5 PM. Routines handles the follow-up chores that used to eat up manual effort once a conversation ends.

CEO Mati Staniszewski framed the whole push around a single idea: domain-specific solutions. “Our mission at ElevenLabs is to solve human-technology interaction, and for the first time, it's solvable,” he said. He predicted conversational AI will end up as essential to enterprises as cloud computing or payment systems. “Every business will need to reinvent how it interacts with its customers, employees, and the wider world,” he said.