How it works

A small recorder, a desktop app, and later, a ping on your phone.

The recorder writes down what your coding agents do on your computer. The desktop app reads that record on the same computer and shows it three ways. The phone ping is planned; it isn’t built yet.

  1. The recorder

    It runs on your computer and notes the programs your coding agents start: when each one starts and stops, which chat started it, the folders it works in, the names of files it creates or opens, and the destinations it connects to. It writes that to a record on your machine.

    Destinations are addresses. To put names on them, the recorder looks up a short, fixed list of well-known host names; an address that isn’t on the list stays an address.

  2. The desktop app

    It reads the record where it is and shows three views. It doesn’t upload the record, and it doesn’t stop programs for you.

  3. Planned

    A ping on your phone

    When something needs a look, your phone gets “needs your attention” and the name of a category, like “left running”. That’s the whole message: no program names, files, folders or addresses. The details stay on the computer. It isn’t built yet.

The three views

  • Right now

    Which agent chats are open, and the programs each one is running on your machine, as they run.

    Doesn’t showyour conversation with the agent. It shows what ran, not what was said.

  • Left running

    What is still running after a chat ended, like a dev server holding a port or a background job that never stopped.

    Doesn’t showwhether a program should be stopped. It shows what is still running; that call stays yours.

  • Where they connected

    The destinations the agents’ programs reached from this machine, named where the recorder can name them, with a “new” mark on ones it hasn’t seen before.

    Doesn’t showthe contents of those connections. It records where, not the data that went back and forth.

Rooted RealityIllustrative · not real data

Right now

2 chats · 3 programs
Claude Code~/code/shop-api· open 42 min

Looks like: running tests.

node· listening on :517341 min
node38 min
Claude Code~/code/hooks· open 9 min

Looks like: editing and testing.

python2 min

Doesn’t show: your conversation with the agent.

Left running

2 programs
node · holding :3000
running 2 h 34 min · its chat ended 12:24
from chat 0b9d in ~/code/shop-api
python · holding :8000
running 1 h 40 min · its chat ended 13:05
from chat 5e10 in ~/code/docs

Doesn’t show: whether it should be stopped.

Where they connected

today · 4 destinations
node· registry.npmjs.org×38
git· github.com×4
python· api.github.comnew×2
python· 151.101.64.223no name for this address

Doesn’t show: the contents of those connections.

What it records, and what it never records

Records

  • The programs your coding agents start
  • When each one starts and stops, and which chat started it
  • The folders they work in
  • The names of files created, changed or removed while a chat is open
  • The destinations they connect to: the address, and a name where the recorder can put one

Never records

  • Your conversations with the agent
  • What's inside your files
  • The contents of network traffic

It keeps names and places, never what’s inside them.

What crosses your computer’s edge, and what doesn’t: Trust.

What it works with

Claude Code today; Codex next. Mac first; Windows next.

Join the waitlist.

The private beta is forming now, Mac first. Free during the beta; a one-time Pro purchase later.

Join the waitlist