How it works in Codex
Run codex as always. Since Codex 0.157 it draws full screen and makes the selection itself:
drag over text, and peekme reads the selection Codex highlights, also over SSH. Then press Alt+P.
Inside tmux, Codex leaves the mouse to the terminal when tmux has the mouse off, which is tmux's
default. A selection is then your terminal's own, and over SSH peekme cannot read it. Add
set -g mouse on to ~/.tmux.conf and restart Codex. peekme tells you this the first time you
press Alt+P there.
Which model explains
No API key is needed. peekme asks Codex's own app-server, with your Codex login, on a temporary
thread that is not saved, so nothing appears in your history. Codex runs a shared server in the
background, and peekme uses it, so it adds only a few MB of memory. Without that server peekme
starts its own codex app-server, which takes about 250 MB.
A peek uses the model your account lists as fast. /model in Codex shows the names. A typed question (Alt+Shift+P) uses the
thread's model and gets a copy of the whole thread. To change either model:
export PEEKME_MODEL=<model> # peeks, default your account's fast model
export PEEKME_CODEX_ASK_MODEL=<model> # questions, default the thread's model
peekme and /side
Codex's own /side (also /btw) starts a side chat that stays out of the main transcript. It is
good for a question about the work. peekme is for the words in front of you: you select them
instead of typing them, the answer opens right under them, a box can open inside a box, and the
answer stays on the screen as an underline. Use both.
A closer comparison
What gets peekme
codex, codex "prompt", codex resume and codex fork open with peekme. codex exec and
anything with output going to a pipe or a file run Codex directly. command codex starts plain
Codex once.
Tested with Codex 0.161 on Linux and macOS, also inside tmux and over SSH.
peekme also works in Claude Code, GitHub Copilot CLI, Kiro CLI, with the same keys.