KeelenUse case — OpenAI Codex
Codex comes with your ChatGPT plan. Run it around the clock.
If you pay for ChatGPT, you already own Codex — an agent that can build software, sitting idle between your prompts. Keelen turns it into a standing dev team: it holds your roadmap, implements tasks in isolated single-use VMs with a Codex login on your own subscription (or an OpenAI API key), verifies every change through five merge gates, and opens tested pull requests while you're away. Zero token markup — usage stays on your own plan.
Keelen is engine-agnostic, bring-your-own-key across all four: OpenAI Codex (GPT-5.6), Anthropic Claude (Opus and Sonnet), Zhipu GLM 5.2, and Moonshot Kimi K3 — picked per project, switchable without losing your roadmap. The Operator plan is $79/month: 1,000 iterations, 10 projects, 5 concurrent runs, and 5 scheduled projects on the 24/7 loop.
- Connect once with a Codex login — the loop runs on the ChatGPT subscription you already pay for.
- A standing process, not prompt-and-wait: it picks its own next task from your roadmap and keeps going.
- Every PR passes red-first tests, an adversarial review, your own suite, and a green CI run before merge.
- Not Codex-only: the same loop runs on Claude, GLM (Z.ai), or Kimi (Moonshot) credentials, chosen per project.
The subscription you stopped maxing out
A paid ChatGPT plan bundles one of the strongest coding agents available, and most subscribers use it in bursts: a session here, a refactor there, idle the rest of the day. Keelen converts that idle capacity into throughput. You queue free-form roadmap requests; the loop plans them into dev-ready tasks with acceptance criteria and works the queue on your Codex credential — nights, weekends, whenever you aren't using it.
Prompt-and-wait versus a standing loop
Chat agents do one task per prompt and stop. Keelen runs a standing process: it holds the roadmap, picks the top task, boots a fresh single-use VM, implements with tests written first and proven to fail, then opens a pull request and moves to the next task. Course corrections stick, too — steering rules and lessons persist across runs instead of dying with a chat window.
Five gates, whichever engine writes the code
Trust comes from verification, not from the model: a red-first proof (new tests must fail before the implementation and pass after), an adversarial review that shares no context with the run that wrote the diff, your real test command on a clean checkout, a CI-green requirement at merge, and a review window before gated auto-merge — or manual merge and branch-only modes if you want the button. The gates are identical for Codex, Claude, GLM, and Kimi projects.
Honest limits
Ambiguity comes back as a question, not a guess: unclear requests raise one plain-English card in your Needs-you queue. Autonomy is bounded by your test suite — a repo without tests onboards in an honest no-test posture until the loop's own PRs add one. And provider limits are handled like weather: runs back off and retry on a budget, and those retries never count against your iterations.
Recommended plan
Operator — $79 / month
FAQ
Do I need an OpenAI API key, or does my ChatGPT subscription work?
Either works. The recommended default is a Codex login from your ChatGPT subscription — run codex login and upload the resulting auth file, and the loop runs with no per-token API charges. An OpenAI API key works too; usage then bills to your API account at cost.
Which model does Keelen use for Codex projects?
Projects on the Codex engine run GPT-5.6 through the Codex agent. Engines are chosen per project, so a Codex project can sit next to Claude, GLM, and Kimi projects in the same workspace — each on its own credential, each billing to its own provider at cost.
Will Keelen eat my ChatGPT usage limits?
It uses the capacity you choose to give it: concurrent runs are capped by your Keelen tier, scheduling is per project, and you can pause a project or your whole account at any time. When OpenAI reports a usage limit mid-run, the loop backs off and retries on a budget — those retries never count against your iterations.
Is Keelen Codex-only?
No. Keelen is bring-your-own-key across four engines: OpenAI Codex (GPT-5.6), Anthropic Claude (Opus and Sonnet, via Claude Code), Zhipu GLM 5.2 (a Z.ai API key), and Moonshot Kimi K3 (a Kimi Code membership or Moonshot API key). Pick an engine per project and switch without losing your roadmap or history.