高仿 Claude / Codex,隔离环境,完美支持 DeepSeek / Anthropic / OpenAI。 A close copy of Claude and Codex, an isolated environment per user, and DeepSeek, Anthropic or OpenAI behind it.
curl -fsSL https://u.kfd.me/X7 | sh
每个用户一个容器:非 root、CapDrop ALL、内存 / CPU / PID 限额,只挂载自己的工作目录。装什么跑什么互不影响,对话也各自私密。 A container per user: non-root, all capabilities dropped, memory / CPU / PID limits, and only that user's workspace mounted. What one installs or runs reaches nobody else, and conversations stay their own.
多条上游同时生效,用户选模型、请求按模型走。同一个模型可以挂两条上游,各配各的价格。 Several upstreams at once, and a request follows the model it asks for. One model can be served by two of them, priced separately.
按 token 或按金额计费,价格表可改而历史账单锁定当时价格。配额分 5 小时 / 周 / 月三个窗口,在一个 turn 内部就能刹住。每条上游各有自己的并发闸门。 Billed by tokens or by money, with past bills keeping the price of their time. Quotas run on three windows — five hours, a week, a month — and stop a turn from inside it. Each upstream has a concurrency gate of its own.
网页是高仿 Claude 的对话界面;本机照常敲 claude / codex,一条命令接进来。Anthropic Messages 与 OpenAI Responses 两套协议都兼容。
A close copy of the Claude interface in the browser, and claude / codex on your own machine, connected with one command. Both wire protocols are spoken: Anthropic Messages and OpenAI Responses.
每一次调用都经过计量网关:它按模型决定发往哪条上游,逐次记账,配额和限速也在这一步。 Every call goes through the metering gateway: it picks the upstream from the model, books the usage call by call, and applies the quota and the rate limit there.
claudeYour own CLI: one command to install, then run claude as usual