免费领glm5.2?还可以用在codex,claude上?Get glm5.2 for free? It can also be used on codex and claude?


AI free token/AI免费积分

你每个月花 20 美元订阅 AI 编程工具——Qoder 、 Codex 、 Claude Code——但你想过没有,这些订阅额度只能在自家终端里用? Cursor 想用?不行。 LobeChat 想用?不行。自己写的脚本想调 API ?更不行。

 

你的 20 美元,买了一个只能用一种姿势的 VIP 卡。

 

直到我装了这个东西。

 

一行命令,把订阅变成 API

 

qorder-proxy 做的事很简单:把 Qoder CLI 的订阅额度,翻译成你本机的 OpenAI / Anthropic 兼容 API 。

 

装好之后,你的 127.0.0.1:3000 就是一个完整的 OpenAI API 端点。 Cursor 、 Continue 、 LobeChat 、 ChatBox 、甚至你自己写的 Python 脚本——任何支持 OpenAI 协议的客户端,都能直接调 Qoder 订阅里的模型。

 

Qwen3.7-Max 、 DeepSeek V4 Pro 、 GLM-5.2 、 Kimi K2.6——这些国产顶级模型,不再被锁在 Qoder 终端里。它们可以在你喜欢的任何工具里工作。

 

这不是中转站。没有公网部署,没有多人共享,没有商业运营。它只在你本机运行,只吃你自己的订阅额度。

 

为什么需要"翻译"?

 

问题出在协议上。

 

Qoder CLI 有自己的对话协议、消息格式、流式响应格式和工具调用格式。这些格式和 OpenAI 的 /v1/chat/completions 不一样,和 Anthropic 的 /v1/messages 也不一样。就像一个只会说日语的人和一个只会说法语的人坐在同一张桌子上——他们能看见彼此,但聊不起来。

 

qorder-proxy 就是那个翻译官。

 

它启动一个本地 HTTP 服务,接收 OpenAI 或 Anthropic 格式的请求,翻译成 Qoder CLI 能理解的格式,调用 CLI 完成推理,再把结果翻译回客户端期望的格式返回。

 

整个过程中,你的 Token 从来不离开你的机器。

 

3 分钟装好

 

我试了一遍,确实 3 分钟够了。

 

npm install -g @qodercn-ai/qoderclicn

git clone https://github.com/lininn/qorder-proxy.git
cd qorder-proxy/qoder-proxy
npm install && npm link

qorder-proxy --web

qorder-proxy start

 

启动后 curl http://127.0.0.1:3000/health,看到 {"ok":true} 就能用了。

 

Web 配置界面是 v2.0 新加的——之前只能用 CLI 命令行配,现在浏览器一填就完事, Token 、后端、端口三项,保存后自动关闭。比我上次配 nginx 反代舒服十倍。

 

它支持什么模型

 

CN 后端( qoder.com.cn )当前支持的模型列表:

 

•qwen3.7-max —— 通义千问旗舰,支持推理强度调节( low/medium/high/max 四档)•deepseek-v4-pro / deepseek-v4-flash —— DeepSeek V4 双版本•glm-5.1 / glm-5.2 —— 智谱 GLM 最新版•kimi-k2.6 —— 月之暗面 Kimi 最新•qwen3.6-plus / qwen3.6-flash —— 千问 3.6 系列

 

还有个 auto 模式——让 Qoder 自己选最合适的模型。

 

这些模型在 Cursor 里用是什么体验?我试了 Qwen3.7-Max 配 Continue 插件,代码补全和对话都正常走。 DeepSeek V4 Pro 做复杂重构,速度和直接用 CLI 差不多——因为底层确实就是在调 CLI 。

 

两个后端,两个世界

 

qorder-proxy 支持双后端切换:

 

  CN 后端 Global 后端
CLI 命令 qoderclicn qodercli
认证方式 Personal Access Token OAuth 登录
服务地址 qoder.com.cn qoder.com

 

一行命令切换:qorder-proxy config set backend global,然后 qorder-proxy restart。

 

CN 后端用 PAT 认证,在 qoder.com.cn/account/integrations 创建,不依赖任何第三方登录流程。 Global 后端走 OAuth ,和 Qoder 官方一致。

 

安全边界:它不是中转站

 

这点必须说清楚。

 

qorder-proxy 默认只监听 127.0.0.1,不绑 0.0.0.0,不暴露公网。配置文件权限 0600 ,只有本机用户可读。日志自动脱敏 Token 、 Cookie 、 Authorization 头。

 

README 里写得很硬:禁止公网部署、禁止多人共享、禁止转售 API 、禁止绕过官方计费风控。

 

这不是套话。这个工具的设计逻辑就是"个人本机使用"——每次请求启动一个新的 CLI 子进程,根本不是为并发设计的。你要是拿它当公网 API 网关,性能会教你做人。

 

市面上那些 OpenAI 中转站——CloseAI 、 openai-forward 、 chatgptProxyAPI——做的是另一件事:公网代理转发,多人共享 API Key ,按量或按月收费。那是商业服务,跟 qorder-proxy 不是同一个赛道。

 

qorder-proxy 更像是一个协议适配器——你说法语,它说日语,它帮你翻译,翻译完就完事了。

 

工具调用的局限

 

说实话,不完美。

 

工具调用( Tool Use / Function Calling )是通过 Prompt 指令 + 文本解析实现的,不是模型原生能力。什么意思?就是 qorder-proxy 会在发给 CLI 的 Prompt 里塞一段指令,让模型按特定格式输出工具调用,然后再用正则或 JSON 解析把结果捞出来。

 

这个方案的缺点很明显:如果模型输出不规范的 JSON ,或者干脆拒绝使用工具格式,响应会降级为纯文本。另外,带工具的请求不走流式——因为需要完整 JSON 才能解析。

 

但这不算是致命伤。大部分开发者用 qorder-proxy 的场景是对话和代码生成,不是重度 Agent 工作流。真要跑 Agent ,直接用 Qoder CLI 本身更合适。

 

Web 控制台:小而美

 

启动后访问 http://127.0.0.1:3000/ui,有个本地 Web 控制台:

 

•Dashboard —— 健康状态、 Base URL 、模型数量•Models —— 模型列表(调 /v1/models)•Chat Test —— 简单非流式测试•Config —— 生成 OpenAI / Anthropic / OpenCode 配置示例•Usage —— 本地用量统计( token 级别,估算值)

 

Usage 页面的数据是本地估算,不代表 Qoder 官方账单。但用来监测自己的消耗节奏够用了——至少能知道今天是不是用太多了。

 

一个有意思的问题

 

qorder-proxy 诞生在一个灰色地带的边缘。

 

它不提供 Token 、不共享额度、不绕过计费——但它的存在本身,确实让 Qoder 订阅的价值变大了。 20 美元的 Qoder Pro ,原来只能在 CLI 里用,现在变成了一整套本地 API 生态。这对 Qoder 官方来说,是好事还是坏事?

 

用户得到了更多价值,但也许也意味着更高的资源消耗。如果每个人都把订阅额度接到 Cursor 里高强度使用, Qoder 的算力成本会不会扛不住?

 

目前 Qoder 没有对这个工具表态。 README 里的免责声明写得很清楚——如果官方不允许,请立即停止使用。

 

这句话,既是边界,也是悬在头上的问号。

 


 

相关链接:
- qorder-proxy GitHub : https://github.com/lininn/qorder-proxy
- Qoder CN PAT 创建: https://qoder.com.cn/account/integrations
- Qoder 官网: https://qoder.com/zh

 

免费领 Qoder Pro:用我的邀请链接注册,直接送一个月 Pro——GLM-5.2 随便用。→ https://qoder.com.cn/referral?referral_code=SOTtWioqAyhveY6SPPYSfjMIuKm4CnLz

 

You spend $20 a month on AI coding tools like Qoder, Codex, and Claude Code— but have you ever wondered that these subscriptions only work in their own terminals? Want to use Cursor? No. Want to use LobeChat? No. Want to call the API from your own scripts? Definitely not.

Your $20 buys a VIP card that only works in one posture.

Until I installed this thing.

One command turns your subscription into an API

qorder-proxy does something simple: it translates your Qoder CLI subscription quota into an OpenAI / Anthropic-compatible API on your local machine.

After installation, 127.0.0.1:3000 becomes a complete OpenAI API endpoint. Cursor, Continue, LobeChat, ChatBox, or even your own Python scripts—any client that supports the OpenAI protocol can directly call models from your Qoder subscription.

Qwen3.7-Max, DeepSeek V4 Pro, GLM-5.2, Kimi K2.6—these top domestic models are no longer locked inside the Qoder terminal. They can work in any tool you like.

This isn't a relay station. No public deployment, no multi-user sharing, no commercial operation. It only runs on your local machine and only consumes your own subscription quota.

Why do we need "translation"?

The problem lies in the protocols.

Qoder CLI has its own conversation protocol, message format, streaming response format, and tool-calling format. These formats differ from OpenAI's /v1/chat/completions and also from Anthropic's /v1/messages. It's like a person who only speaks Japanese sitting at a table with a person who only speaks French—they can see each other, but they can't communicate.

qorder-proxy is the interpreter.

It starts a local HTTP service that receives requests in OpenAI or Anthropic format, translates them into a format Qoder CLI understands, calls the CLI to complete inference, then translates the result back into the format the client expects and returns it.

Throughout the entire process, your Token never leaves your machine.

Installed in 3 minutes

I tried it myself, and indeed 3 minutes is enough.

After starting, curl http://127.0.0.1:3000/health and see {"ok":true}—it's ready to use.

The web configuration interface is a new addition in v2.0—previously you could only configure via CLI commands, now just fill in a browser form. Token, backend, and port—save and it auto-closes. Ten times more comfortable than configuring nginx reverse proxy last time.

What models does it support

CN backend (qoder.com.cn) currently supports the following models:

• qwen3.7-max — Tongyi Qianwen flagship, supports adjustable reasoning intensity (low/medium/high/max four levels) • deepseek-v4-pro / deepseek-v4-flash — DeepSeek V4 dual versions • glm-5.1 / glm-5.2 — Zhipu GLM latest versions • kimi-k2.6 — Moonshot Kimi latest • qwen3.6-plus / qwen3.6-flash — Qwen 3.6 series

There's also an auto mode—letting Qoder pick the most suitable model on its own.

What's it like using these models in Cursor? I tried Qwen3.7-Max with the Continue plugin, and both code completion and conversations worked normally. DeepSeek V4 Pro for complex refactoring runs at about the same speed as using the CLI directly—because under the hood, it really is calling the CLI.

Two backends, two worlds

qorder-proxy supports dual-backend switching:

Switch with one command: qorder-proxy config set backend global, then qorder-proxy restart.

The CN backend uses PAT authentication, created at qoder.com.cn/account/integrations, without relying on any third-party login flow. The Global backend uses OAuth, consistent with Qoder's official approach.

Security boundary: it's not a relay station

This point must be made clear.

qorder-proxy listens only on 127.0.0.1 by default, does not bind 0.0.0.0, and does not expose to the public internet. Config file permissions are 0600, readable only by the local user. Logs automatically redact Tokens, Cookies, and Authorization headers.

The README states firmly: no public deployment, no multi-user sharing, no reselling APIs, no circumventing official billing risk controls.

This isn't boilerplate. The tool's design logic is "personal local use only"—each request spawns a new CLI subprocess. It was never designed for concurrency. If you try to use it as a public API gateway, performance will humble you.

Those OpenAI relay stations on the market—CloseAI, openai-forward, chatgptProxyAPI—are doing something completely different: public proxy forwarding, multi-user shared API Keys, charged by usage or monthly subscription. That's a commercial service, not in the same league as qorder-proxy.

qorder-proxy is more like a protocol adapter—you speak French, it speaks Japanese; it translates for you, and that's it.

Tool-calling limitations

To be honest, it's not perfect.

Tool calling (Tool Use / Function Calling) is implemented via Prompt instructions plus text parsing, not native model capability. What does this mean? qorder-proxy injects a block of instructions into the Prompt sent to the CLI, asking the model to output tool calls in a specific format, then uses regex or JSON parsing to extract the results.

The downside of this approach is obvious: if the model outputs non-standard JSON, or outright refuses to use the tool format, the response degrades to plain text. Also, requests with tools don't go through streaming—because complete JSON is needed for parsing.

But this isn't a fatal flaw. Most developers using qorder-proxy are in dialogue and code-generation scenarios, not heavy Agent workflows. If you really need to run Agents, using Qoder CLI directly is more appropriate.

Web console: small but elegant

After starting, visit http://127.0.0.1:3000/ui to access a local web console:

• Dashboard — health status, Base URL, model count • Models — model list (calls /v1/models) • Chat Test — simple non-streaming test • Config — generate OpenAI / Anthropic / OpenCode configuration samples • Usage — local usage statistics (token-level, estimated values)

The data on the Usage page is locally estimated and does not represent Qoder's official billing. But it's sufficient for monitoring your own consumption pace—at least you'll know if you've used too much today.

An interesting question

qorder-proxy was born on the edge of a gray area.

It doesn't provide Tokens, doesn't share quotas, doesn't circumvent billing— but its very existence does increase the value of a Qoder subscription. A $20 Qoder Pro subscription that previously only worked in the CLI is now a complete local API ecosystem. Is this a good thing or a bad thing for Qoder officially?

Users get more value, but it may also mean higher resource consumption. If everyone connects their subscription quota to Cursor for intensive use, can Qoder's compute costs hold up?

Currently, Qoder hasn't made any statement about this tool. The disclaimer in the README is clear—if the official does not permit it, stop using it immediately.

This sentence is both a boundary and a question mark hanging overhead.

Related links: - qorder-proxy GitHub: https://github.com/lininn/qorder-proxy - Qoder CN PAT creation: https://qoder.com.cn/account/integrations - Qoder official website: https://qoder.com/zh

Get Qoder Pro for free: Register using my referral link and get one month of Pro for free—use GLM-5.2 as much as you want. → https://qoder.com.cn/referral?referral_code=SOTtWioqAyhveY6SPPYSfjMIuKm4CnLz

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry