JetBrains / IntelliJ terminal renders Codex CLI user-vs-assistant distinction too weak compared to Ghostty/iTerm
What feature would you like to see?
I want the Codex CLI UI to remain readable in JetBrains terminal emulators (IntelliJ IDEA and related JetBrains IDEs).
Today, the visual distinction between user input and Codex output is much weaker in JetBrains embedded terminals than it is in Ghostty or iTerm.
Current behavior
In JetBrains terminal:
- the separation between my prompts and Codex responses is hard to scan
- the left-margin markers are too subtle
- the color / contrast hierarchy is weak enough that long sessions visually blur together
In comparison, the same Codex CLI session is much easier to read in:
- Ghostty
- iTerm
What I would expect
Even if the terminal host differs, Codex should still present enough hierarchy that users can immediately distinguish:
- user prompt blocks
- assistant response blocks
- tool / progress lines
Suggested improvements
Any of these would help:
- Stronger default contrast between user and assistant blocks
- More distinct markers / separators in the left margin
- Optional high-contrast TUI mode for terminals with weaker rendering
- Detect JetBrains terminal environments and increase contrast / separation automatically
Why this matters
When scrolling through a long session in JetBrains terminal, it is much harder to tell who is speaking than in Ghostty or iTerm. The result is that Codex feels visually muddled even though the exact same CLI output is clear in other terminals.
Environment
- Codex CLI version: 0.114.0
- Terminal where it looks bad: JetBrains embedded terminal (IntelliJ IDEA / JetBrains IDEs)
- Terminals where it looks good: Ghostty, iTerm
- Platform: macOS
Additional context
There is already a broader issue about improving visual distinction between user input and AI output, but this report is specifically about the JetBrains embedded terminal experience being noticeably worse than Ghostty/iTerm for the same Codex CLI UI.
This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗