Process tables, listening sockets, /proc or ps, the container engine if one answers, and the transcript stores your coding agents already write. All of it on this machine, all of it as you.

Nothing. There is no server, no telemetry, no update check and no network code beyond two local operations, both worth being exact about:

• it connects outward to ports on this machine to see what answers;
• it binds a candidate port for a moment to confirm it is free, then closes it.

Neither calls listen(). portlist has no port of its own, so a tool for watching what is listening never adds to the list.

Views 7 reads what Claude Code, Codex, Copilot CLI, VS Code, Cursor and Gemini write locally. Prompts stay on the machine: there is nowhere for them to go. Account details are narrower than the file they come from - the plan and the organisation, never the address or the account id, and values are scrubbed as well as field names, because an organisation name can itself contain an address.

It never stops anything on its own. portlist kill and portlist cleanup stop what you say yes to, one listener at a time, after showing what it is; the leftover guess decides what to ask about, never what to end, because a tool that both guesses which process is abandoned and ends it will eventually end the wrong one. Every stop re-checks with a fresh scan that the process is still on that port, refuses pid 0, pid 1, portlist itself and anything filed as part of the system, and stops a container, a Homebrew service, a launchd job or a systemd unit through its manager rather than by signal. Standard input that is not a terminal is never asked a question: without --yes, nothing is stopped.

Open an issue at https://github.com/Mr-hunt-007/portlist/issues. If it is sensitive, say so in the issue without the detail and it can move somewhere private from there.