Vertebrae: Privacy-First AI Notetaker

Vertebrae, a privacy-first AI notetaker from the team behind OpenTools, launched in beta on iOS, offering end-to-end encrypted voice calls and outbound phone calls with local transcription on each participant's device. The app aims to give AI agents context while ensuring consensual recording, with inference run on-device or in a trusted execution environment and hardware-signed attestation for verification.

Why we exist We believe Vertebrae is the best way to both give your agents great context and respect your right to speak freely. On the one hand, AI that can “listen in” on work discussions is vastly more useful than AI that can’t. Agents equipped with the right company context can act at the speed of thought; they push code<sup 1 fn-1 </sup and update systems of record as soon as the decision is made. They work magically when equipped with the tacit knowledge that has historically evaded being written down. On the other hand, the modern workplace has accidentally become a surveillance state. Every call you join is assumed to be recorded<sup 2 fn-2 </sup . Meeting capture tools today allow a ton of parties to eavesdrop: there’s the AI notetaker service itself, all of their subprocessors, the hackers who compromise those subprocessors, the lawyers who subpoena the data, etc. When recording is the default, “any statement you make may be used for or against you.” We solve two acute issues with the status quo. 1. Vertebrae is verifiably private. Calls are end-to-end encrypted<sup 3 fn-3 </sup . All inference is run either on-device or in a trusted execution environment<sup 4 fn-4 </sup . Your device verifies a hardware-signed attestation of the exact environment before sending any data. We publish the attested measurements, so you can confirm the running environment matches what we claim we’ll release a how-to guide soon . 2. Recording is consensual. We believe the data belongs to all participants, not just to whoever records. On calls between Vertebrae users, both sides see a transcription notice during the call, and each participant’s device transcribes the call locally — so everyone gets the transcript, not just whoever chose to keep notes. Disclosure and delivery are part of the network by design. We aim to be just as usable to humans as to agents. Our MCP server makes use of everything we learned as the team behind OpenTools<sup 5 fn-5 </sup , one of the original server registries in the space. Today, we’re live in beta on iOS and support two types of meetings: - End-to-end encrypted voice calls with other Vertebrae users - Outbound phone calls from your existing phone number We’re working on supporting in-person meetings in the app once the speaker diarization meets our quality bar. Following that, we’ll release our native desktop app, which will support meeting platforms like Google Meet, Microsoft Teams, Zoom, etc.