OpenAI's Codex CLI Passes 117k Stars — a Terminal Agent Written in Rust
Codex is trending again on GitHub. Nearly 9,800 commits into a Rust-based local coding agent, the language choice tells you what OpenAI is optimising for.
OpenAI's Codex repository is back near the top of GitHub's daily trending list, with roughly 2,000 stars added in a day against a total above 117,000 and 17,900 forks.
The description is deliberately unglamorous: a lightweight coding agent that runs in your terminal. It is available for macOS, Linux and Windows, connects to ChatGPT plans, and integrates with IDEs including VS Code.
Rust is the decision worth reading
The repository is organised around codex-rs alongside codex-cli, and the project is built with Bazel. For a company whose public surface is overwhelmingly Python, choosing Rust for the local agent is a statement about what this tool is for.
A terminal coding agent is a long-running local process that shells out constantly, watches files, and has to stay responsive while a model streams tokens. Startup time and memory footprint are user-visible in a way they are not for a server-side service. Rust addresses exactly that, at the cost of a slower contributor ramp.
What the repository structure shows
Nearly 9,800 commits sit behind it. The tree includes an sdk directory, a docs directory, a patches directory, and a .devcontainer — the shape of a project expecting to be embedded and extended rather than only installed.
There is also an AGENTS.md at the root, which has quietly become a convention for telling coding agents how to behave inside a given repository. A coding agent shipping its own AGENTS.md is a reasonable place for that convention to get reinforced.
The category is consolidating
Terminal-first coding agents have gone from novelty to a crowded category in about two years. The differentiators now are not whether the agent can edit files — they all can — but process model, permissioning, how sandboxing works, and how well the thing behaves when a task runs for an hour.
Watching where the commits land in a repository this active is a better signal about that competition than any launch announcement.
Source: openai/codex — GitHub