Pocket built a note-taker for the conversations you actually have

The best conversation of your week probably did not happen on a video call. It happened in a car, or a kitchen, or the thirty seconds after a meeting officially ended and everyone finally said what they meant. None of it was captured. Pocket, a San Francisco startup in Y Combinator's Winter 2026 batch, built a small piece of hardware around that gap.

The product is a credit-card-shaped puck with three microphones and a single button. It has no screen. You press it, it records the room, and later the Pocket app hands back a transcript, a summary, speaker-attributed notes, and a list of the things you agreed to do. Snap it to the back of a phone and a contact microphone picks up both sides of a call. It works offline and supports more than 120 languages.

The framing is deliberate. "We thought every meeting notetaker was built for online conversations," co-founder and CEO Akshay Narisetti has said, "but nothing was geared towards real-life talk." That sentence is the whole company. Otter, Granola, Fireflies, and Read AI live inside your calendar and your Zoom window. Pocket lives in your pocket.

The case against the screen

There is a small act of rebellion in shipping a device with no display in 2026. Everything else on the market fights for your eyes. Pocket is designed to be forgotten. You are meant to stop noticing it, which is the only way a recorder becomes something you actually keep on you. Subtraction, in wearables, is a feature. The things a designer leaves out are usually the reason a product gets worn at all.

The one-button ritual matters more than it looks. Every extra step between a thought and its capture is a place where the thought escapes. No app to open, no meeting link to paste, no wake word to remember. Press, talk, walk away. The app does the rest when you are back on Wi-Fi. For anyone who has ever left a good meeting and realized halfway to the parking lot that they have already lost the one number that mattered, the appeal is immediate.

The screen-free choice also does something to the social dynamics of a recorded conversation. A phone laid on a table, glowing, is a small act of surveillance that everyone in the room feels. A card stuck to the back of that same phone reads as furniture. Pocket is betting that the least intrusive recorder is the one that gets used the most, and that people are more honest around a device that is not staring back at them.

How it works

The pipeline is simple to describe and hard to build, because the hard part is not the transcription - that is close to a commodity now - but the capture. Getting clean audio out of messy, overlapping, in-person talk is the actual engineering problem, and it is why Pocket sells a microphone array rather than just another model wrapper.

Who is actually buying it

The buyer list reads like anyone whose job is mostly talking to people: founders and salespeople, consultants and recruiters, doctors and researchers, students sitting in lectures. For teams, Pocket adds custom workflows, webhook support, and integrations with Google Calendar, Google Drive, OneDrive, Obsidian, and - notably for a company selling memory - Claude and Cursor. The enterprise pitch is accountability: turn the hallway chat into a trackable action item nobody can later claim they never heard.

The scale is what makes the story more than a demo. Pocket reports delivering over 30,000 units in its first five months and more than 130,000 since launch, at a roughly $27M annualized revenue run rate. Those are self-reported figures and worth treating as approximate. The number that is harder to manufacture is the 4.89-star average across hundreds of reviews. Growth can be bought. Satisfaction at scale, in hardware, where a bad batch means returns and a defect means a refund, cannot.

What buyers seem to be paying for is not transcription so much as recall. The promise is that you can walk into a follow-up two weeks later and know exactly what was said, by whom, and what you promised to deliver. Sales teams use it to keep their pipeline notes honest. Consultants use it so the billable insight from a site visit does not evaporate on the drive home. Students point it at a lecture and get a study guide. The common thread is people whose memory is the bottleneck in their work, not their calendar.

The pricing bet

Pocket charges $129 once. Basic transcription is included with the device, and the premium tier - unlimited AI summaries, assistant queries, daily highlights, file attachments - is an optional subscription around $200 a year. Bundles run up to roughly $327. In a market where nearly everything is a monthly bill, asking for money one time and making the subscription genuinely optional is close to contrarian. The wager is that people trust a product they own more than one they rent.

There is a second reason the one-time price is shrewd. Hardware buyers forgive a device they own in ways they never forgive a subscription. A monthly charge invites a monthly re-evaluation; a purchase invites attachment. By putting the transcription in the box and the AI features on an optional plan, Pocket lets the free tier do the marketing. The device earns trust first, and the upsell arrives only once the habit is set.

Where it sits in the pack

The AI note-taking category is crowded, and most rivals compete on the software: better summaries, tighter calendar hooks, smarter search. Pocket competes one step earlier, at the moment of capture. The contact mic for phone calls, the offline recording, the screen-free form factor - these are hardware decisions, and they are the ones that are hard to copy from a laptop.

Software-only tools like Granola and Otter are excellent inside the meeting they were invited to, and useless the moment the conversation leaves the call. Pendant and puck rivals such as Plaud and Limitless share Pocket's instinct that the world needs a physical capture device, which is the more interesting fight - it is a bet that a whole new category of always-available recorder is forming, not that any one app wins. Pocket's differentiation inside that group is the phone-call trick and a pricing model that does not require a subscription to be useful.

The people behind it

Narisetti is a hardware lifer. He says he built more than 100 robots as a kid, starting around twelve, and before Pocket he helped build Omi, one of the larger open-source AI wearables. That matters less as a résumé line than as a reason the reviews are good: shipping physical things over and over teaches you the parts software people never touch - supply chains, tolerances, returns, unit economics. His co-founder, Gabriel Dymowski, previously ran DoxyChain, a blockchain-based document infrastructure company for enterprises. One founder obsessed with the object, one with the system of record behind it.

The investors noticed the pattern. In June 2026 Pocket raised $11M led by Accel, with Y Combinator and Mati Staniszewski, the CEO and co-founder of ElevenLabs, joining in. A voice-AI founder putting money into a voice-capture device is its own kind of endorsement. Forbes later named Pocket one of the 21 most promising startups in the W26 batch.

The team is small - roughly a dozen people in San Francisco - which is part of why the review score is worth watching. A twelve-person company that has shipped 130,000 devices is running lean on a supply chain that would strain a team five times its size. That is a hardware discipline you do not fake, and it is the kind of operational detail that separates a viral gadget from a durable one.

The question hanging over it

A device that records real life carries a real question: consent. The most useful thing about Pocket - that it captures the unscheduled, unguarded conversation - is also the thing that makes people uneasy, and recording laws differ by place and situation. Pocket's answer so far is design rather than policy: the device is visible, the button is manual, and nothing records without a deliberate press. How it handles the etiquette of a recorder that is meant to disappear will shape whether it becomes a mainstream tool or stays a power-user habit.

The larger bet under all of it is about where real work happens. We have spent a decade optimizing the scheduled, on-screen part of the day and quietly ignored the unstructured conversations where most decisions actually get made. Pocket is a device for that second category. It is a wager that the next useful piece of personal AI is not another chat window but a small object that sits in the background and remembers on your behalf. Whether a screen-free recorder becomes a habit for millions or stays a power-user tool is the open question, and hardware categories have a way of looking obvious only in hindsight. For now, more than 130,000 people have decided it is worth keeping in a pocket.