How pong works

Most assistants are a chat window over someone else's servers. pong is closer to a person with their own computer: it has a machine, it keeps things running on it, and it texts you when there's something to say.

Your own computer#

When you sign up, pong provisions a computer just for you. It's an isolated cloud machine with its own disk, its own browser profile, its own files and its own memory. Nobody else's pong can reach it, and it can't reach anyone else's. Everything else in these docs, the browser, the workspace, hosting, happens on this machine.

You

iMessageWhatsApp💬SMS

Your pong's computer

Isolated, just yours

BrowserWorkspace and filesMemoryVault

Your apps

Mail and contactsCalendarGitHub, Slack, Notion…

trypong.app/dashboard looks at the same computer: live browser, files, routines, crew

What a turn looks like#

1. 1

   You text

   pong puts a 👀 on your message. That's how you know it picked it up.

2. 2

   A short acknowledgement

   For anything that takes more than a moment, one short line that says what it's checking. Then it goes quiet and works.

3. 3

   Work happens in parallel

   Research and browsing are split into up to three parts that run at the same time, each in its own browser tab. A trip becomes the stay, getting there and things to do. A price hunt becomes one site per worker.

4. 4

   One answer

   When the parts are back, pong compares them and texts you the pick and why, then the real contenders numbered, each with its link and a screenshot of the page. Long jobs get at most a couple of progress updates along the way.

You can keep texting while workers run. Add a detail ("actually make it two rooms") and pong steers the running job instead of starting over. Say stop and it stops.

Sleep and wake#

Your computer goes to sleep when nothing is using it, after about half an hour of quiet, or a few minutes after the last job finishes. A text wakes it in about a second, and so does opening the computer page on the dashboard. Messages that arrive while it sleeps are queued, never dropped. Routines and reminders live outside the computer, so they fire on time and wake it when they need it.

Work that outlives the text#

An install, a build, a scrape, a server: pong starts it on your computer and ends the turn. The process keeps running. When it's a run you're waiting on, pong also sets itself a reminder to look again and brings the result back without you asking. A server that should stay up just stays up.

What pong does on its own#

pong can start routines and watches by itself when a job needs a later look, and new mail can wake it once your mailbox is connected. When pong is working on a turn you didn't start, it acts like a good executive assistant: it puts things on your plate (a reminder, a watch, a draft, a calendar check) and never does anything you can't take back. It won't buy, pay, send an email or deploy from a background turn. A job you set up yourself, like "every monday, email the team the numbers", runs with your permission.

Asking before anything that matters#

Anything that reaches another person, spends money or is hard to undo gets shown to you first: the draft, the total, the booking. pong goes ahead on your yes. Read-only and low-stakes things it just does.

The model#

Ask pong "what model are you running?" and it tells you the exact model and provider it's on right now. Builders can run pong on their own ChatGPT plan or any OpenAI-compatible endpoint from the developer console.

Speed#

A simple reply takes a few seconds, even from a sleeping computer. Browser jobs take minutes because pong is actually opening and walking the sites, but they run in the background, so you can put the phone down.