codexfast

What the tool does

Codexfast is a macOS runtime launcher for OpenAI's Codex app that re enables custom API features which are present in the app but gated behind local checks. The README describes it as applying temporary runtime patches for the current session while leaving the original app.asar, Info.plist, app bundle and app signature untouched. The features it restores include a fast settings control in the settings panel, a composer slash command named /fast, a speed submenu in the composer, compatibility with the GPT-5.5 and GPT-5.6 model catalog for custom API users, and a disable automatic updates switch in settings. The README states that it patches only the local gates still needed for a verified build and does not add a backend service or call a private OpenAI API. The project is published on npm as the codexfast package and the simplest invocation is npx codexfast launch.

How the patching works

According to the README, the Codex app already contains the fast, slash command, speed and updater UI paths in its packaged frontend bundle, so codexfast only patches the local gates that remain. The launch command starts Codex with a local Chrome DevTools Protocol endpoint, attaches through the browser level CDP target before the renderer JavaScript runs, intercepts matching renderer JavaScript responses for that session, and applies narrow patch rules in memory. The README advises keeping the codexfast launch process running while using Codex, because settings and patched feature chunks can load lazily and the runtime interceptor must stay attached after the first window appears. The disable automatic updates switch is stored in the Codex desktop configuration as desktop.disableAutomaticUpdates, and codexfast injects a process local main process hook that discovers updater and desktop settings modules by source signature across the build files, then reads the latest configuration before each update check.

Which builds are verified

The README lists a long set of verified local builds, with the latest being ChatGPT.app and Codex.app build 6892, version 26.818.31338, which keeps the runtime automatic update hook and settings schema but does not ship the renderer disable automatic updates row. It also verifies many earlier builds going back through 2025, and it points to a feature scope document for the exact feature boundaries per build. There is a separate OAuth login mode for custom API that lets a user sign in with any ChatGPT account and point model requests at a custom provider while keeping official authentication, which avoids the capability restrictions of pure API login. The README includes a detailed config.toml example showing the custom provider settings, including a local base url and a dummy bearer token placeholder. The repository carries 59 stars and 4 forks with 4 open issues at the time of this writing, the default branch is main, and the license is MIT.

Editorial conclusion

Codexfast is an MIT licensed macOS tool on the main branch with 59 stars that launches Codex via npx codexfast launch, patches only in memory for the current session, and lists build 6892 as its latest verified local build.