How Buoy works, honestly
Buoy asks you to share a window with a stranger. That deserves a plain description of what happens, not a trust badge. Here it is, in the order things happen.
The circle
Buoy is a small floating avatar that sits above every app and every Space on your Mac. Click it and a panel opens. The panel does three things: shows whether you're available to help, lets you ask for help, and shows the current session if you're in one. That's the whole app.
Asking
You type or dictate what's wrong. Dictation runs on your Mac using Apple's speech recognition; audio never leaves the machine for that step. Your description goes to our server, which boils it down to a one-line summary and a few tags. Helpers see the summary and tags. They do not see your description, your screen, or your name until a session starts.
Matching looks for people who are available right now and a level or two above you. It widens the range in rings (first the closest levels, then wider) and gives up after three minutes if nobody's free. You can say "Human," "Bot," or "Either." If you say Human, you get a human. If a bot ever answers, it's labeled as one.
Helping
If you're available, you get a card: the seeker's first name and level, the tags, the one-line summary, a rough time estimate, and what you'd earn. You have 45 seconds to accept or pass. Passing costs nothing and isn't held against you.
The session
When someone accepts, voice connects automatically through a media server we run. Nobody's camera is ever used. Either person can share a screen, and sharing means one window you pick from a list, or a whole display if you choose that deliberately. The picker puts AI tools first because that's what people share. Stop sharing whenever you like; the other person sees a blank.
Right now what you share is sent as-is. We're building on-device masking that blurs things that look like tokens, keys and email addresses before the frame is encoded, so they never reach the server. Until that ships, share the window, not the password manager.
Credits
You start with 100 credits. Getting help costs a base rate of 4 credits a minute, multiplied by how far above you the helper is (one level up is cheapest). The first minute of every session is free. Credits are held when a session starts and settled when it ends, rounded to the minute, and the helper gets what you spent. There's no money in the system today. When there is, we'll write a post like this one about it first.
Ratings and levels
After a session both sides answer two questions: solved (yes, partly, no) and thumbs up or down. Those move levels slowly. Levels are coarse on purpose; the point is to find someone slightly ahead, not to rank people.
What we store
Your LinkedIn name, photo, email and a per-app ID. Your requests and their summaries. Session metadata: who, when, how long, how many credits, how it was rated. We do not currently record audio or video. If we start recording sessions for safety review (we expect to, with consent and a visible banner), the recording will be scrubbed and this post will change before the feature does.
What isn't built yet
Annotation (drawing on the shared window), on-device masking, session recording, the helper web console, Windows, enterprise accounts, and paying anyone real money. We'd rather tell you that than imply it.
Where it runs
One Linux box we rent, running the API, the database and the media server. It's enough for the early-access group. The site you're reading is static files, so it will survive a popular day; the matching server is the part we'll scale when it needs it, and the status link in the footer is the real health check, not a picture of one.
Questions, bugs, or "this isn't how it behaved for me": email Ben. A person will answer.