A New App Takes Meeting Notes, Turns Speech Into Text and Can Draft Emails

ElevenLabs has released ElevenAI, a new experimental app that takes meeting notes, turns speech into typed text and can connect to other tools. The company launched it in the week of November 13, 2026, during its 11/11 Summit in New York, according to TechCrunch.
The app is in preview. You cannot download it directly. You must sign up for early access. ElevenLabs is aiming it at teams rather than individuals. That shifts the focus from a personal transcript to a shared record of a conversation that several people can ask questions about.
For meetings, ElevenAI can take notes and tell the other people on the call that the conversation is being recorded. Users can search across many meetings. During a chat, the app can also help recall details from earlier conversations. The notes work less like a summary written at the end of a call and more like a searchable memory that covers many meetings.
The dictation feature turns speech into clean text in any application. It supports more than 90 languages. ElevenLabs says the system understands industry-specific terms in both dictation and meeting transcripts. This matters most in fields with heavy jargon, product names and abbreviations, which often confuse general speech tools.
Companies can also connect ElevenAgent to carry out follow-up tasks. The example given is writing an email based on meeting notes. Here the transcript is the starting point for another task, not the final product.
The launch follows a week in which ElevenLabs introduced v3 of its Scribe speech-to-text models. These offer live transcription and fine-tuning on company data, meaning the model can be trained further on a company's own material. Taken together, the announcements suggest the company is building a set of tools that covers speech models, an app and an agent layer. The products are separate, however, and each has its own way to get access.
The broader context here is consent and data handling. Recording rules differ by place and by type of call, so the notification feature is a visible sign that the issue is being recognized. Searching across meetings is useful, but it also creates a stored archive of what colleagues said. Administrators, auditors and, after a breach, attackers would all want access to that archive. Enterprise buyers are likely to check these points first.
In my view, the most useful test for ElevenAI will be whether the agent works reliably on messy meeting output. That means overlapping voices, incomplete transcripts and specialist terms the model has not seen. Note-taking and dictation are already common features. The harder claim is that a team can pass a transcript to an agent and get an action that needs no correction.
If those pieces hold, the gain is modest but real. People spend a noticeable part of their working time rebuilding what was agreed in a call. A tool that turns that rebuilding into a search, and a draft email into a single step, takes friction out of routine work. The preview period is when that claim can be tested against real use.


