@wangyingqiThe day has finally come to ship Hemory. It’s hard to believe that in an age when AI coding is everywhere, what is, on paper, a recording-and-transcription app took us nine months to build.
Hemory started as a way to scratch my own itch. Later I talked Jun, the CEO of WizNote, into joining. Years ago, after ONES acquired WizNote, Jun and I often talked about building the next generation of note-taking. Jun has spent more than twenty years in note-taking tools and knows the history of every notes app on the market — Obsidian and Notion go without saying. What I admire most: he still swears by Roam Research to this day (never heard of it? doesn’t matter — only the old wizards know that one). He threw himself into Hemory, coding through the night, and most days I had to talk him into going home to sleep. I’d get up past eight to find that the code he’d pushed at five had already built clean. His exact words: “Note-taking software is dead.”
Back to what I needed. Like everyone who wants transcription, I started with meeting notes. First the software-only tools — Granola, Otter, and the rest — then hardware recorders. Plaud’s hardware design is excellent; I bought one the moment it came out. But once I’d tasted what AI transcription could do, my standards went up. A few pain points these products just couldn’t solve for me.
One — it has to listen and capture continuously.
I was always forgetting to start or stop the app — or the Plaud. Halfway through a meeting I’d realize it wasn’t on; other times I’d notice long after a meeting had ended that it was still recording, and several meetings’ worth of topics had blurred into one mush. I spend half to two-thirds of my workday in meetings, and having to constantly mind and toggle a recording app or device drained my attention. I thought about just recording nonstop, but most transcription apps cap a single file at four hours. And a day’s topics can be all over the place — I didn’t want them crudely lumped into one meeting file.
Two — recall what I’ve said and heard, from any angle.
An instant meeting memo didn’t matter much to me. What I wanted was the chance to get something back when I needed to recall it. And getting it back shouldn’t mean search in the traditional sense — I didn’t want to punch in keywords, land on a few memos, and then have to read them. No — those memos are already lossy. What I wanted was recollection that spans time and meetings, organized around a topic. With those memories in hand, I could have AI work them up into all kinds of things.
Here’s the experience I wanted: within a window I set in advance — say my workday, nine to eight — a service listens alongside me to everything that happens, and with zero effort on my part, automatically uploads, transcribes, and splits the day into separate meetings based on what was actually said. I can scroll a timeline of those segmented meetings and everyday chats to review the day’s topics and read the summaries; or I can use the agent I live in — Codex, OpenClaw — to talk to the service anytime, mining and working with everything it’s heard about my work and life.
I don’t want meeting transcription. I want a tool that captures all my conversations and automatically saves them as memory.
I looked around. Nothing like it existed. So we built one.
Hemory — from Hear and Memory. Heard, and written down.
The features are what I described above — all of them shipped. I’ve been building it and using it at the same time. For more than seven months now, Hemory has been on by default whenever I’m working. It’s just software: it runs on an iPhone, and on an Apple Watch too. On the Watch, Hemory is smoother than any third-party recording hardware can match — I’ll write a separate post on that.
The moment that surprised me most: after connecting it to Codex, I asked —
What interesting things did you hear yesterday?
What did you hear last month?
Write up the PRD for the new features we’ve been discussing lately.
Write me a half-year review based on what you’ve heard.
Take the pricing strategy we’ve talked over several times this month and write it up as a full competitive-analysis doc.
When I watched Codex pull Hemory’s memory, think for a moment, and start spilling out answers — I got goosebumps.
After that I never wanted to miss another day. I wanted to capture all of it. A personal second brain, a digital assistant, whatever you call it — without complete memory, it can’t know who I am, where I came from, or where I’m going. Calling Hemory an assistant feels too ordinary — it’s more like a clone of yourself. Once Hemory fully knows and understands how I express myself, and has heard the conversations I have with the people around me — I am him. He is me.
This is a small beginning. It may sound like a small product, but our little team of obsessive perfectionists rebuilt it from scratch three times before we dared to ship. Hemory’s mission is clear: remember everything you’ve said and heard, and prepare lossless context for AI.
“The limits of my language mean the limits of my world” (that one’s Wittgenstein) — language is also an expression of how a person sees the world. As AI grows more capable — seemingly able to do it all, and getting better at it by the day — what we think and what we say become the last thing that tells one human being from another. I once asked Hemory to guess my MBTI from everything it had heard me say. The answer was scarily accurate. Goosebumps again.
I hope Hemory can be a good companion for us, helping us look outward at the world and inward at ourselves — and, along the way, quietly getting the work and whatnot done too.
Download it at Hemory.com. I hope you like it.