KeelenUse case — Claude Max
Your Claude Max plan sleeps when you do. Put it on the night shift.
If you pay for Claude Max, you own an AI that can code around the clock — and probably use it four hours a day. Keelen runs the other twenty: an autonomous dev team that holds your roadmap, implements tasks in isolated single-use VMs on the Claude subscription you already pay for, verifies every change through five gates, and opens tested pull requests overnight. Zero token markup — the loop runs on the plan you already own.
The hook is Claude-specific; Keelen isn't. It's bring-your-own-key across four engines — a Claude Pro or Max login (or Anthropic API key), a Codex login from your ChatGPT subscription (or OpenAI API key), a GLM (Z.ai) API key, or a Kimi K3 key (Kimi Code membership or Moonshot) — chosen per project. The Operator plan is $79/month: 1,000 iterations, 10 projects, 5 concurrent runs, and 5 scheduled projects on the 24/7 loop.
- Your idle subscription hours become an overnight dev team — no new token bill, no markup.
- Wake up to tested pull requests with acceptance criteria, not a chat transcript to replay.
- Five gates before anything merges: red-first tests, adversarial review, your own suite, green CI, and a review window you control.
- Not Claude-only: the same loop runs on Codex (ChatGPT), GLM (Z.ai), or Kimi (Moonshot) credentials — switch engines without losing your roadmap.
The idle-subscription math
A Max plan is priced for heavy use, and most of it goes unused: you prompt while you're at the keyboard, and the plan idles the other twenty hours. Keelen inverts the ratio. You queue intent — free-form roadmap requests in plain English — and the loop spends your idle hours turning the top of the queue into shipped work. Keelen's fee is flat and the expensive input is the subscription you already own; at full Operator use the platform fee works out to about $0.08 per run.
What a night on the loop looks like
The scheduler picks the top task from your roadmap, boots a fresh single-use VM, clones your repo, and works the task with Claude Code: tests written first and proven to fail, then the implementation, then your own test suite on a clean checkout. Each finished task ships as its own pull request with acceptance criteria. When something needs a decision only you can make, the loop doesn't improvise — it files one plain-English question in your Needs-you queue and moves to the next task. Your morning review is a merge queue, not an archaeology dig.
Five gates before it touches main
Every change proves itself before merge: a red-first proof (new tests fail before the implementation and pass after — tests born green are rejected), an adversarial review by an independent model 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. Prefer eyes on everything? Set manual merge or branch-only per project and the loop leaves the button to you.
Honest limits
Keelen asks instead of guessing: ambiguous requests come back as clarifying questions, not creative interpretations. Your test suite is the ceiling on autonomy — a repo with no tests onboards with an honest no-test posture until the loop's own PRs add a suite. And your plan's limits are respected: when Claude reports a usage window, runs back off and retry on a budget, and those retries never count against your iterations.
Recommended plan
Operator — $79 / month
FAQ
Will Keelen burn through my Claude Max usage limits?
It uses whatever capacity you choose to give it, and you hold the dials: concurrent runs are capped by your plan tier, scheduling is per project, and you can pause one project or everything at any time. If Claude reports a usage limit mid-run, the loop backs off and retries on a budget — those retries never count against your Keelen iterations.
Does Keelen see or store my Claude credentials?
Your token is encrypted at rest under independently rotated keys and injected only into the isolated single-use VM that runs your iteration — the machine is destroyed when the run ends, and credentials on disk die with it. Revoke access anytime and the token is deleted immediately. Keelen never resells tokens; usage bills to your own Anthropic account.
Do I connect a Claude subscription login or an API key?
Either. The recommended default is a subscription token minted with Claude Code's setup-token command — the loop then runs on your Claude Pro or Max plan with no per-token API charges. An Anthropic API key works too; usage then bills to your API account at cost.
Is Keelen Claude-only?
No. Keelen is bring-your-own-key across four engines: Anthropic Claude (Opus and Sonnet, via Claude Code), OpenAI Codex (GPT-5.6 — sign in with your ChatGPT subscription or paste an API key), 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.