A new category of wearable device has moved from niche to mainstream over the past two years: small pendants, clips, and pins that listen to your day, transcribe what they hear, and turn it into searchable notes, summaries, and reminders. Marketed as a “second brain you wear,” devices like Bee and Limitless promise to capture conversations, meetings, and ideas you’d otherwise forget.
In 2026, the category has consolidated significantly — Amazon acquired Bee, and Meta acquired Limitless — which changes the calculus on where your recorded conversations actually end up. If you’re considering one of these devices, or already wearing one, here’s what’s worth understanding.
How These Devices Work
The basic design is consistent across brands: a small device, worn as a pendant, clip, or pin, with a microphone that’s active for most or all of your waking hours. It doesn’t record full audio continuously in the way a voice recorder does — instead, it processes audio in short segments, transcribes speech, and sends the transcripts (and sometimes audio snippets) to a cloud service for summarization.
The output is a running log: summaries of conversations, extracted action items, names and topics mentioned, and a searchable history you can query later — “what did we agree on in that meeting Tuesday” or “what was the name of that restaurant my friend recommended.”
The value proposition is real for some use cases — people with memory difficulties, professionals in back-to-back meetings, or anyone who’s ever walked away from a conversation forgetting a key detail. But the mechanism that makes it useful — constant listening — is also what makes it privacy-sensitive in ways most wearables aren’t.
The Bystander Problem
A phone, laptop, or photo storage app holds data about you, by your choice. An always-on listening pendant holds data about everyone who talks near you — colleagues, friends, family, strangers in a coffee shop, a doctor during an appointment, a lawyer during a consultation — almost none of whom agreed to be recorded or know that they are.
This isn’t a hypothetical edge case; it’s the default mode of the device. Every conversation you have while wearing it becomes part of a transcript and summary stored on a company’s servers, regardless of whether the other person would have consented if asked.
Recording laws vary by location, and many require consent from everyone in the conversation. A number of U.S. states and countries require “two-party consent” — meaning every participant in a conversation must agree to being recorded, not just the person wearing the device. Wearing an always-on pendant in a two-party-consent jurisdiction can put the wearer in a legally gray area for every conversation it captures, regardless of intent.
Some manufacturers have responded with features like a “consent mode,” which mutes audio from unrecognized voices until they verbally opt in. This reduces — but doesn’t eliminate — the underlying issue: the device still has to listen to determine who’s speaking before it can decide whether to record them, and the practical reliability of voice-recognition-based consent in noisy, real-world conversations is unproven at scale.
Where the Data Goes — and Who Owns It Now
Processing a day’s worth of conversation into useful summaries requires real computing power, which means the audio and transcripts are processed on company servers, not on the device itself. That data — transcripts of your conversations, names and details mentioned in them, summaries of your day — lives on infrastructure controlled by whichever company operates the service.
The acquisitions that reshaped this category in 2025 matter here. When Amazon acquired Bee, and Meta acquired Limitless, the conversational data these devices collect became an asset within much larger companies — companies whose primary businesses include advertising, AI model development, and voice assistant ecosystems (Alexa, in Amazon’s case). It’s reasonable to ask, for any device in this category, exactly what the parent company’s data-handling commitments are, separate from the marketing of the device itself — and whether those commitments are likely to remain stable as the product gets folded into a larger platform.
Questions Worth Asking Before You Buy
If you’re considering an AI pendant or similar always-on wearable, a few questions are worth getting clear answers to before committing:
Is audio processed locally, or sent to a server? Most current devices in this category rely on cloud processing for transcription and summarization. If audio leaves the device, ask how long it’s retained, in what form (raw audio vs. transcript only), and whether it’s used for anything beyond generating your summaries — including AI model training.
What happens to recordings of other people? Ask specifically how the company treats audio of people who aren’t the device’s owner — whether it’s identifiable, how long it’s kept, and whether those people have any way to find out they were recorded or request deletion.
What’s the deletion process, and is it real? “You can delete your data” is a common claim. Whether deletion is immediate and complete, versus retained in backups or training datasets for some period, is the practical difference between a privacy commitment and a marketing line.
Has the company’s ownership changed, or might it? A device bought from a small startup with one set of privacy commitments can end up owned by a much larger company with different incentives — as happened twice in this category within a single year. Commitments tied to a specific company don’t automatically survive an acquisition.
Conversations Carry More Than You Think
A transcript of “everything you said today” isn’t just a record of tasks and reminders — it’s a record of whatever happened to come up in conversation, which for most people includes things they’d never deliberately write down. A venting conversation with a friend about a health scare. A frustrated call with a family member about money. A candid aside about a coworker, a relationship, or a decision still being weighed.
None of that is unusual — it’s just how people talk. But “how people talk” and “what people choose to put in writing” are normally very different things, precisely because speech is ephemeral in a way text isn’t. An always-on transcription device collapses that distinction: everything spoken becomes everything written down, searchable, and stored on a server, including the parts of a day nobody would have chosen to document.
This matters most for categories of conversation that carry their own legal protections when held by the right professional — conversations with a doctor, therapist, or lawyer, for instance, are often privileged or covered by health-privacy law when those professionals record them. A transcript of the same conversation, captured incidentally by a pendant worn by one participant, generally isn’t covered by those same protections, because the device and its operating company aren’t the professional bound by them. The sensitivity of the conversation doesn’t change; the legal framework around it does.
A Different Kind of Memory Tool
It’s worth distinguishing between two very different ideas that both get called “AI memory”:
Ambient capture — devices and features that record everything happening around you by default, and rely on you (or filters) to exclude what shouldn’t be kept. AI pendants, and features like Windows Recall on PCs, fall into this category. The privacy questions here are about other people’s data as much as your own, because the capture happens regardless of who’s present.
Deliberate storage — tools where you choose what to save: a photo you took, a document you scanned, a voice memo you recorded intentionally, a note you wrote. The privacy questions here are mostly about your own data and how it’s protected — encryption, who can access it, whether it’s used for anything beyond storing it for you.
daftei is built around the second model. It stores the photos, documents, voice notes, and files you choose to add — encrypted in transit with TLS 1.3 and at rest with AES-256 — without recording anything in the background, without listening to your conversations, and without capturing data about people who haven’t chosen to be part of it. Your voice notes are recordings you decided to make, for things you decided were worth keeping — not an ambient transcript of everyone you spoke to that day.
The Underlying Tradeoff
AI pendants solve a real problem — the frustration of forgetting useful details from conversations — by making a tradeoff that’s easy to underweight when a device is marketed primarily on convenience: continuous recording of everyone around you, processed by a company whose ownership and priorities can change, in exchange for searchable summaries.
For some specific use cases — accessibility needs, certain professional contexts with appropriate disclosure — that tradeoff may be worth it, made with eyes open and the people around you informed. But it’s a different category of product from a personal memory tool that stores what you deliberately choose to keep. Knowing which category a device falls into — before you put it on — is the most useful question to answer.