Autonomous OpenAI Codex / GPT session runner. Same job as
claudeloop: never block on
a human, and never treat insufficient_quota / billing as a waitable
rate_limit_exceeded window.
Docs: https://adammatthewsteinberger.github.io/codexloop/
Status: M1–M5 on PyPI (0.1.0) — onion core, exec gateway, capacity probes,
control plane, generated REST CLI, optional app-server transport with exec
fallback. MIT-licensed; author Adam Matthew Steinberger.
pip install codexloop
codexloop --version
codexloop doctorFor local development:
pip install -e ".[dev]"codexloop run path/to/plan.md
codexloop watch --follow
codexloop stop| Document | Purpose |
|---|---|
| docs/plans/architecture-and-roadmap.md | Full design / transplant plan from claudeloop 0.5.4 |
| docs/plans/research-notes.md | Vendor SDK/CLI capacity + autonomy research |
| docs/plans/_shared-transplant-outline.md | Cross-product keep/swap + Global Constraints |
| docs/superpowers/plans/2026-08-13-codexloop-implementation.md | Bite-sized TDD implementation plan |
| Item | Value |
|---|---|
| PyPI / CLI | codexloop |
| Env prefix | CODEXLOOP_* |
| State dir | .codexloop/ |
| Done marker | CODEXLOOP_TASK_FULLY_COMPLETE |
MIT — see LICENSE.
Releases go TestPyPI → PyPI via Trusted Publishing. See docs/publishing.md.