Install
curl -fsSL https://raw.githubusercontent.com/brandonwie/codex-hud/main/install.sh | bash
OpenAI Codex CLI status line
The brandonwie/codex-hud repo adds a compact footer for terminal Codex sessions: model, effort, git state, context budget, rolling usage, and pace without leaving the prompt.
curl -fsSL https://raw.githubusercontent.com/brandonwie/codex-hud/main/install.sh | bash
Native Rust renderer
The default status line is rendered by codex-hud, a single self-contained Rust binary (edition 2021, MIT): minimal dependencies, no unsafe code, and a size-optimized build of about 574 KB. No Node runtime sits on the rendering path.
~19 ms saved per paint, about 1.34× faster. The win is Node's ~20 ms interpreter cold-start, which the native binary skips. The rest is identical git and log work, so end to end the gap stays modest and honest.
Median of 40 runs (npm run measure:rust 40), Apple M4 Pro, 2026-06-18. Indicative, single machine.
Config playground
Changing these controls updates the status line preview and generated codex-hud.toml configuration.
Install path
curl -fsSL https://raw.githubusercontent.com/brandonwie/codex-hud/main/install.sh | bash
curl -fsSL https://raw.githubusercontent.com/brandonwie/codex-hud/main/install.sh | bash
codex-hud --line --color
codex-hud --init-config
Search-ready by default
Targets the practical phrase set around OpenAI Codex CLI, status line, context usage, and terminal HUD workflows.
The status line is rendered by codex-hud, a native Rust binary: a single self-contained executable with no Node runtime on the rendering path.
No framework payload, no remote fonts, no client dependency chain, and only one small local script for the live preview.
FAQ
No. The project stages its patched launcher under user-local paths and keeps stock Codex install paths untouched.
Codex HUD reads optional codex-hud.toml files from user, project, and explicit config paths, then falls back to defaults.
Model, reasoning effort, project, branch, context usage, rolling usage windows, token totals, and pace formatting.
The default renderer is codex-hud, a small native Rust binary: minimal dependencies, no unsafe code, and a size-optimized build of about 574 KB. The Rust binary's own startup and render is effectively instant; end-to-end status-line latency is bounded mainly by the git commands it runs.
Yes. Patched mode binds model, reasoning effort, service tier, context, and token totals to each session. A fresh session can briefly show no effort plus Ctx:? and Tkn:? until its rollout opens. The 5h and 7d limits remain account-wide, so shared movement is expected and is not session bleed.
Launch Codex with codex -m <model> -c 'model_reasoning_effort="<effort>"'. Codex's /model flow can persist the selected model and effort to global ~/.codex/config.toml.