ElevenLabs made its biggest move yet beyond voice generation on November 11, 2026. At its New York Summit, the company unveiled a package of 13 new products and upgrades that turns it from a voice toolmaker into an enterprise AI platform: businesses will be able to fine-tune models on their own data, a new ElevenAI app will act as an AI teammate for employees, and voice agents will get memory, routines and a smarter system for routing conversations. An expanded partnership with NVIDIA adds serious compute behind the push.
The message from CEO Mati Staniszewski was blunt about where this is heading. He said conversational agents would become “critical infrastructure for every enterprise, as essential as cloud or payments” within a few years. That is a big claim, but the breadth of the announcement backs it up: this is not one feature, it is a whole enterprise stack, from the models underneath to the marketplace where partners sell finished applications.
Custom Models Tuned on Your Own Data
The most consequential announcement is model customization. In the coming months, selected enterprises will work directly with ElevenLabs researchers to fine-tune selected open-source language models and ElevenLabs’ own voice models using data from their live deployments. Companies will be able to optimize for the metrics they actually care about, such as customer satisfaction or reliability, and build reinforcement-learning stacks tailored to their own context.
That matters because generic models, however capable, do not know a company’s customers, products or rules. A bank’s voice agent needs different behavior from a hospital’s, and a retailer answering support calls needs different phrasing from an insurer qualifying claims. Because the tuning data comes from live deployments, the models can keep improving as interactions accumulate, rather than being frozen at the point a general industry model was trained.
ElevenLabs is already applying the same idea in narrower form. Scribe v3 Medical, its speech-to-text model for healthcare, is built to recognize specialized medical terminology, and a domain-optimized financial-services model is also in development. The new program generalizes that approach: instead of waiting for ElevenLabs to ship an industry model, a company can help shape one for its own operation.
There is a catch worth noting. The customization work is rolling out to selected enterprises first, and it begins “in the coming months” — it is a roadmap item, not something every customer can sign up for today. Pricing has not been disclosed either, so the real test will be whether mid-sized companies can afford it or whether it stays a large-enterprise offering.
The Interaction Engine: Picking the Right Model for Each Moment
Alongside customization, ElevenLabs previewed the Interaction Engine, a system designed to orchestrate a conversation rather than just generate speech. It manages turn-taking and interruptions, and it can assign different language models to different parts of an exchange — a cheaper model for triage, a frontier model for the complex part of a task.
That routing is as much about cost as quality. Running a frontier model for every second of every call is expensive, and most of a routine exchange does not need it. By matching the model to the moment, ElevenLabs aims to keep routine exchanges cheap while still bringing in the heavyweight model when the conversation gets hard. It is the same cost-versus-capability balancing act shaping the wider agent market, from Google’s Gemini 4 Carbon for long agent work to the always-on AI coworker Databricks is building with Genie.
ElevenAgents Gets Memory and Routines
ElevenAgents, the company’s platform for deploying voice and chat agents across sales, support, recruitment and operations, is getting two features that push agents from answering questions toward doing work. Memory will recognize returning authenticated customers and distill each conversation into a structured record of preferences, constraints and future occasions — so a customer who prefers Spanish and can only talk after 5 p.m. would not need to repeat those details on the next call. The customer relationship management system remains the primary customer record; Memory is the conversational layer on top of it.
Routines, also forthcoming, is designed to automate multi-step work before, during and after conversations, connecting with a company’s existing tools and systems. The examples ElevenLabs gave are concrete: pre-screening a mortgage applicant and calling back within minutes, reminding patients about visits and offering canceled slots to waitlisted patients, emailing reports, writing to a CRM or initiating calls. Together, Memory and Routines move ElevenAgents from a smart answering service toward an agent that retains context and completes the work surrounding a conversation — a direction echoed by Anthropic’s move into spreadsheets, meetings and shared whiteboards with Claude.
ElevenAI: An AI Teammate That Listens to Meetings
For employees rather than customers, Staniszewski previewed ElevenAI, a new app that listens to spoken conversations — including in-person meetings — and turns them into notes, summaries and action items. It can also provide live assistance during a call and pass context between people and agents. Free early access opened with the announcement.
Dictation and meeting notes in ElevenAI use Scribe v3, which ElevenLabs describes as its recorded-audio transcription model covering 102 languages; a separate Scribe v3 Live variant is built for low-latency, real-time audio. The company says the app takes meeting notes with participants’ consent and includes notice and consent mechanisms — an important detail, since an app that listens to in-person meetings will face obvious questions about privacy and workplace trust. How those consent mechanisms work in practice, and whether employees can opt out easily, will matter as much as the transcription quality.
ElevenAI also fits a broader pattern this week: AI is moving onto the device and into the meeting room, not just the chat window. Google’s November Pixel Drop brought Gemini Intelligence, including voice-driven multi-step tasks, to the Pixel 10, and the race is now about which assistant actually does the follow-up work, not which one takes the best notes.
Partners, Marketplace and the NVIDIA Deal
ElevenLabs reports more than 550 partners across 44 countries, including Google Cloud, IBM, Deloitte, DXC and Genesys. Its new Partner Program lets consultancies, systems integrators, software vendors and agencies integrate ElevenAgents into enterprise technology stacks or work with ElevenLabs’ forward-deployed engineers on complex deployments. An alliance with McKinsey and QuantumBlack, AI by McKinsey, pairs the platform with McKinsey’s enterprise-transformation work.
The accompanying Marketplace offers a curated catalog of partner-built applications for healthcare, financial services, insurance, education and retail. Enterprises can browse the applications and contact their developers today; purchasing through existing ElevenLabs commitments is planned to start in 2027. Launch partners include Salient, which automates borrower servicing calls, Gail by Lula, which handles and qualifies insurance inquiries around the clock, and Zendesk, whose agent resolves customer service requests from start to finish, according to ElevenLabs.
The infrastructure news may prove the most durable part of the day. ElevenLabs and NVIDIA expanded their partnership, making NVIDIA’s Nemotron 3.5 Super VL available in ElevenAgents, and ElevenLabs plans to add 15 MW of NVIDIA Vera Rubin compute capacity on top of its existing Blackwell Ultra systems. ElevenLabs says Nemotron 3.5 Super VL achieved leading end-to-end task-completion accuracy — including strong performance in financial calculations and billing analysis — at 70% lower costs compared with other leading open models. The added capacity is intended for large deployments with security or data-residency requirements, and to give ElevenLabs researchers access to optimized compute. That compute story connects directly to the hardware side of the AI boom: AMD’s new EPYC 6 Venice server CPUs, built for the AI agent era, show the same demand for agent-ready infrastructure from the chip side.
The expanded collaboration also covers a sovereign deployment: an on-premises voice agent for Centralna e-Rejestracja, the central appointment system in Poland’s national e-Health platform, with EasyCall supporting delivery. The agent calls patients who only have landlines to remind them about appointments and help them reschedule. ElevenLabs says the central system handles 40 million appointments a year with no-show rates of 10% to 20%, and the deployment uses NVIDIA Confidential Computing to keep citizen data and audio within government infrastructure.
What to Watch Next
The Summit package makes ElevenLabs’ enterprise strategy dependent on more than model performance. Tuning and Memory draw on data accumulated in live use, Routines connects conversations to business systems, and the partner program and Marketplace create routes into enterprises that a voice API alone could not reach. The central execution challenge is now twofold: moving the previewed and early-access products into broader production use, and progressing the Marketplace from browsing to actual purchases through existing ElevenLabs commitments.
For businesses evaluating voice AI, the questions to ask are practical ones. Is model customization available to you, or only to selected enterprises? Does Memory’s consent and data handling fit your privacy rules? And does the Interaction Engine actually cut token costs on your call mix, or only on paper? ElevenLabs has laid out an ambitious enterprise vision; the next few months will show how much of it ships to ordinary customers.
Conclusion
ElevenLabs’ New York Summit was a coming-out party for an enterprise AI company, not just a voice company. Custom models tuned on live data, an Interaction Engine that routes conversations sensibly, agents with Memory and Routines, the ElevenAI meeting teammate, a partner Marketplace and a bigger NVIDIA deal add up to a coherent bet: conversational agents become infrastructure, and the winners will be the ones who own the whole stack. The announcements are real, but several of the most important pieces are previews or early access — so watch for the production rollouts before judging the bet.
