Onkery is a phone-only social app with no profiles. You film when asked. Then you watch a reel of strangers who were asked the same thing.
I built this because every feed I used was a performance for everyone and no one. Onkery does one thing: put your clip next to people who pointed at the same kind of thing. Names, likes, and typing stay out. Live at onkery.vercel.app. Phone only.
Two minutes of the outdoors path on a real phone in a real park. Welcome, camera, "I'm outside", a warm-up, two prompts, a reel of strangers, then your turn. No cuts and nothing sped up. It is silent: Onkery records no audio, and I stripped the recording's park noise to keep the file light.
Tap to play. 2:12, phone screen, nothing sped up.
I was using feeds that treat people as inventory. Instagram, TikTok, even BeReal once it grew a following count. The more people I "met," the less I met anyone. I wanted a visit that ends.
I used the apps I was already opening, then tried to steal one mechanic and throw the rest away.
A daily prompt, then a feed of friends. The prompt is good. The following count ruined it. People started staging the two-minute window. I kept the timed ask and dropped the profile.
Infinite. The algorithm decides who you sit next to. You never know why this clip is next to that one. Onkery's reel is only people who were asked the same line.
No names, but the keyboard is the product. A typed wound on a phone reads as a caption. A pan reads as attention. I killed typing.
I stole the prompt from BeReal and the anonymity from boards. I stole the infinite scroll from nobody.
Feeds are huge and the contact is zero. You scroll past thousands of people and brush against none of them. I wanted the opposite, briefly, then you put the phone down.
I tried typing first. "What were they carrying?" On the phone it read as a random quote. People will film a pan. They will not type a wound into a prototype.
"Show me your hands" is a scavenger hunt. "Show me something you keep meaning to take care of. Do not touch it" is a relationship to a thing. A kept plant and a neglected plant look the same in a photo. The prompt is what changes.
If you don't see other people in this visit, there is no product. I also built it on a laptop for too long. Landscape chrome pretended to be a pocket. It lives on a phone.
Two verbs. Film, then watch.
I wrote a lot of prompts that were scavenger hunts. Hands, shoes, a red thing. They photographed well and meant nothing. The ones that survived asked for a relationship to an object: something you keep meaning to take care of, a mess you made, weather you have been under. You cannot fake that with a pretty still. You can refuse it, which is also information.
Film-only came from the same pass. Typed answers on a phone looked like captions. A pan looks like attention.
Most of the first week was subtracting. Each of these looked like a product. Each of them made the visit about something else.
I shipped a version where you answered in words. On a phone it looked like a quote graphic. People wrote one clever line and left. Nobody filmed.
"Show me your hands." "Find something red." Pretty stills. Zero relationship. I can stage a red mug in ten seconds. I cannot stage neglect.
The second you can follow someone, the reel becomes a talent show. I wanted the clip, not the person attached to it.
Landscape chrome pretended to be a pocket for too long. Reviewers nodded at a laptop and the first honest session, standing up, one thumb, made half the chrome look fake.
These are not survey findings. They are what happened when I used the thing, then asked other people to film from where they were standing.
A sentence is a caption. People will film a plant they keep meaning to water. They will not type that they have been ignoring it.
A kept plant and a neglected plant look identical in frame. Only the ask tells you which. If the line is weak, the reel is a pile of objects.
If you do not see other people in this visit, there is no product. Filling that with my own live camera is a demo talking to itself.
Record, redo, OK and skip had to fit one thumb with the camera already open. If they need a second hand they are decoration.
After the permission screen there is no page, only your own live view with a line of text on it. Buttons are hand-drawn rectangles sitting on that camera feed.
I thought the hard part was the camera. The hard part is other people in the pile. A social product with one real plant clip and a lot of live-camera stand-ins is a loop that works and a meeting that does not. I would not add profiles to fake density.
You don't follow anyone. You get the same prompt as a stranger. Both clips go in the same reel. One of them is yours. Nothing says which.
Your clip is in here. So is theirs. Nobody is labeled.
One loop on a phone. Inside goes A, B, C. Outside goes D, E, F. After three, you pick inside or outside again. Welcome and camera permission only happen once.
Tap a row for the line on screen. Record is tap the ring, auto-stop, then OK, redo, or X.
Pulled from the recording above. Every button is a hand-drawn rectangle, including the shutter, because a wobbly stroke asks less of you than a filled pill. Text sits over your own camera feed from the second screen onward. There is no chrome to learn.
Nine screens from one visit. Six of them are text over your own camera.
Four kinds, in a circle. A kept thing and a neglected thing look identical in frame. Only the prompt tells you which. A loop only borrows from the kind next to it. Keeping never fills in for justdone. Those pictures fight.
Something you keep meaning to take care of. Next to keeping and residue. The plant lives here.
keeping → delay → residue → justdone, then around. Tap a word.
A to F walks the circle, inside first, then outside. Each loop asks two things, then shows you other people who were asked the same. Show me anything is always the same line. Show me yours is always the second prompt.
A and B both end on delay, so the plant shows up twice inside.
The end of the recording: loop D closes and E opens with "Show me something out here that is not yours." The walk keeps going.
Start, then inside or out. Inside walks A, B, C. Outside walks D, E, F. After three you pick again. A loop only borrows from the kind next to it on the ring. Keeping never fills in for justdone. Those pictures fight.
The flowlow. Inside and outdoors stay separate. Neighbors can share a clip. Opposites cannot.
Clips go to Blob. The pile index sits in Redis. Vite ships the phone shell. Nothing is recorded until the shutter. No audio. Desktop visitors get "Turn the phone upright" and that is the whole page.
The loop works. I got the moment your clip sits in a reel of strangers. I did not get people you recognize twice. Most "someone else" slots are still your own live camera. The pile needs other people in it. I stopped there.
The pile will only be a pile if strangers use it.
I designed it on a laptop for too long. Landscape chrome pretended to be a pocket. The first honest session was standing up, one thumb, camera already open. The shutter, redo, OK and skip had to fit that grip or they were decoration.
Eleven prompts never shipped. They were scavenger hunts I was proud of until I filmed them. Hands, shoes, a red thing. They photographed well and meant nothing. The ones that survived asked for a relationship to an object.
Next is other people in the pile. Until then the reel is a demo of a social product talking to itself. I would not add profiles to fake density. I would not open desktop to make the critique easier. The visit has to happen now, on a phone, or it is a different product.