Continuum for Codex

Continuum is an open-source control layer for long-running Codex CLI projects.

Use it when you find yourself repeatedly nudging the same Codex project forward with the same prompt, for example `Proceed`. Do not use it when the work still needs frequent human decisions, prompt reframing, or review after every small step.

The name refers to continuity across many Codex passes.

GitHub · First release · Contact

What It Does

Continuum can run one or more autonomous Codex projects simultaneously. It starts each project with `codex exec`, resumes it with `codex exec resume --last`, keeps project-local progress notes, records runtime state, and provides start, stop, restart, and monitoring controls.

How It Works

For each project, you define a real long-term goal and enable the project once. Continuum then keeps the project going in the same Codex thread until the model reports that it has achieved its goal, or ran into a blocker.

Continuum is project-specific by design. Projects that are not explicitly enabled remain normal interactive Codex projects.

Operational Notes

It can be used in both macOS and Linux environments. On macOS, launches automatically run `caffeinate` while a worker is active so the system does not idle-sleep during unattended runs.

Credits and quota handling are reactive rather than predictive. Temporary rate limits are retried, and surfaced hard quota exhaustion stops the worker cleanly.

SwiftBar support is available for menu bar status and controls here.

Get Started

The repository includes the runner, CLI, project enabler, sample Codex configuration, and setup guides. The GitHub README is the best starting point.

Independent project. Not affiliated with OpenAI.