KeelenUse case — Kimi K3
Kimi K3 made the coding plan flat. Keelen runs it flat-out.
A Kimi Code membership buys a flat quota of serious agentic coding — the hard part is turning those model hours into merged work you can trust. Keelen does exactly that: connect a Kimi Code membership key (or a Moonshot Open Platform API key) and an autonomous loop plans your roadmap into tasks, implements them in isolated single-use VMs on Kimi K3 via Claude Code, verifies every change through five merge gates, and ships tested pull requests. Zero token markup — usage stays on your membership quota.
Keelen is bring-your-own-key across four engines — Kimi K3 (a Kimi Code membership or Moonshot key), Anthropic Claude (subscription login or API key), OpenAI Codex (ChatGPT login or API key), and Zhipu GLM 5.2 (a Z.ai API key) — chosen per project. The Indie plan is $29/month: 200 iterations, 3 projects, 2 concurrent runs, and one scheduled project on the 24/7 loop.
- One pasted Kimi key turns Kimi K3 into a scheduled dev team on your repos.
- Kimi Code membership recommended: the loop runs on your flat membership quota — no per-token bill, no markup.
- Verification does the trusting: red-first tests, an adversarial diff review, your own suite, and green CI before anything merges.
- Not Kimi-only: switch a project to Claude, Codex, or GLM anytime without losing your roadmap.
The membership engine
A Kimi Code membership is a coding plan, not a meter: a flat quota of Kimi K3 agent use with no per-token bill. That shape is exactly what a standing loop wants. Where a membership's value usually dies — idle hours — Keelen schedules work instead: free-form requests become prioritized roadmap items, items become dev-ready tasks with acceptance criteria, and tasks become pull requests, each implemented in a fresh single-use VM on your key. The flatter the quota, the more of the long tail you can afford to hand over: test debt, refactors, dependency bumps, papercuts.
Trust the gates, not the vibes
The case for running any model unattended — membership or metered — is the same: verification. New tests are applied without the implementation and must fail; an independent adversarial review reads the diff with no context from the run that wrote it; your real test command runs on a clean checkout; nothing merges over a red required CI check; and gated auto-merge waits out a review window you control. A change that can't prove itself doesn't land, whichever engine wrote it.
A night on the loop
The scheduler picks your top task, boots an isolated VM, clones the repo, and works the task on Kimi K3 through Claude Code: failing tests first, then the implementation, then your suite on a clean checkout. Each finished task is its own pull request with acceptance criteria. Decisions that belong to you come back as one plain-English question in the Needs-you queue — the loop doesn't improvise product answers. In the morning you review a merge queue, not a chat log.
Honest limits
Provider weather is handled for you: when Kimi reports a usage limit or an overload window, runs back off and retry on a budget — retries never count against your iterations, and the loop resumes on its own. Ambiguous requests become clarifying questions rather than guesses. And your test suite bounds the autonomy: a repo with no tests onboards in an honest no-test posture until the loop's own PRs add one.
Recommended plan
Indie — $29 / month
FAQ
What Kimi credential does Keelen need?
One of two keys, pasted once. Recommended: a Kimi Code membership key from kimi.com — the loop then runs on your membership's flat coding quota with no per-token charges. A Moonshot Open Platform API key works too; usage then bills per token to your Moonshot account. Either way the key is encrypted at rest, injected only into the single-use VM that runs your iteration, and deleted the moment you revoke it.
Is Kimi K3 good enough to run unattended?
Keelen doesn't ask you to trust the model — it makes every change prove itself. New tests must fail before the implementation and pass after, an adversarial review reads the diff independently, your own test suite runs on a clean checkout, and nothing merges over red CI. When a task is beyond the engine, it comes back to you as a question instead of a bad merge.
Is Keelen Kimi-only?
No — Keelen is bring-your-own-key across four engines: Moonshot Kimi K3, Anthropic Claude (Opus and Sonnet, via Claude Code), OpenAI Codex (GPT-5.6), and Zhipu GLM 5.2. Engines are per project: run Kimi on the side project and Claude on the client repo, side by side in one workspace.
What does Keelen cost on top of my Kimi plan?
A flat platform fee with zero token markup: Indie is $29/month for 200 iterations, Operator is $79/month for 1,000, and Agency is $299/month for 5,000. On a Kimi Code membership key your usage stays on the membership quota; on a Moonshot API key it bills to Moonshot at cost. Most tasks ship in one to three iterations.