Your AI Programming Subscription Might Only Be Half Used

You spend $20 a month subscribing to AI programming tools — Qoder, Codex, Claude Code — but have you considered that these subscription credits can only be used within their own terminals? Want to use them in Cursor? No. In LobeChat? No. To call the API from your own script? Even less possible.

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

Until I installed this thing.

One Command Turns Your Subscription into an API

What qorder-proxy does is simple: translate the subscription credits of Qoder CLI into a local OpenAI / Anthropic compatible API.

Once installed, your 127.0.0.1:3000 becomes a fully functional OpenAI API endpoint. Cursor, Continue, LobeChat, ChatBox, 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-tier domestic models are no longer locked inside Qoder's terminal. They can work in any tool you like.

This is not a relay hub. No public deployment, no multi-user sharing, no commercial operation. It runs only on your local machine and consumes only your own subscription credits.

Why Do We Need "Translation"?

The problem lies in the protocol.

Qoder CLI has its own conversation protocol, message format, streaming response format, and tool call format. These formats differ from OpenAI's /v1/chat/completions and Anthropic's /v1/messages. It's like a person who only speaks Japanese and another who only speaks French sitting at the same table — they can see each other but cannot have a conversation.

qorder-proxy is the interpreter.

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

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

Install in 3 Minutes

I tried it, and indeed it took just 3 minutes.

# 1. Install Qoder CLI
npm install -g @qodercn-ai/qoderclicn

# 2. Clone and install the proxy
git clone https://github.com/lininn/qorder-proxy.git
cd qorder-proxy/qoder-proxy
npm install && npm link

# 3. Configure (Web UI, just fill in the token)
qorder-proxy --web

# 4. Start
qorder-proxy start

After starting, run curl http://127.0.0.1:3000/health, and if you see {"ok":true}, it's ready.

The web configuration UI was added in v2.0 — previously you had to configure via CLI commands; now you can fill everything in the browser: token, backend, port, three fields, save, and it auto-closes. It's ten times more comfortable than the last time I configured an nginx reverse proxy.

Which Models Does It Support?

Current model list for the CN backend (qoder.com.cn):

• qwen3.7-max — Tongyi Qianwen flagship, supports reasoning intensity adjustment (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 — Qianwen 3.6 series

There's also an auto mode — let Qoder choose the most suitable model itself.

What's it like using these models in Cursor? I tried Qwen3.7-Max with the Continue plugin; code completion and conversation both worked normally. DeepSeek V4 Pro for complex refactoring, speed comparable to using CLI directly — because under the hood it is indeed invoking the CLI.

Two Backends, Two Worlds

qorder-proxy supports switching between two backends:

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, independent of any third-party login flow. The Global backend uses OAuth, consistent with Qoder's official approach.

Security Boundary: It's Not a Relay Hub

This must be made clear.

qorder-proxy listens only on 127.0.0.1 by default, not 0.0.0.0, and does not expose to the public network. Configuration file permissions are 0600, readable only by the local user. Logs automatically redact Token, Cookie, and Authorization headers.

The README states firmly: No public deployment, no multi-user sharing, no API resale, no bypassing official billing risk control.

This is not boilerplate. The tool's design philosophy is "personal local use" — each request starts a new CLI subprocess, not designed for concurrency at all. If you try to use it as a public API gateway, performance will teach you a lesson.

Those public OpenAI relay hubs on the market — CloseAI, openai-forward, chatgptProxyAPI — do something else: public proxy forwarding, multi-user shared API keys, charged by volume or monthly. They are commercial services, not in the same race as qorder-proxy.

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

Limitations of Tool Calls

To be honest, it's not perfect.

Tool Use / Function Calling is implemented via Prompt instructions and text parsing, not native model capabilities. What does that mean? qorder-proxy inserts an instruction into the Prompt sent to CLI, asking the model to output tool calls in a specific format, then uses regex or JSON parsing to extract the results.

The drawbacks of this approach are obvious: if the model outputs non-standard JSON, or outright refuses to use the tool format, the response degrades to plain text. Additionally, requests with tools do not use streaming — because the complete JSON is needed for parsing.

But this is not a fatal flaw. Most developers use qorder-proxy for conversation and code generation, not heavy Agent workflows. For serious Agent work, directly using Qoder CLI itself is more appropriate.

Web Console: Small but Beautiful

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

• Dashboard — health status, Base URL, model count
• Models — model list (via /v1/models)
• Chat Test — simple non-streaming test
• Config — generate OpenAI / Anthropic / OpenCode configuration examples
• 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 rhythm — at least you'll know if you've been using too much today.

An Interesting Question

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

It does not provide tokens, does not share credits, does not bypass billing — but its mere existence indeed increases the value of a Qoder subscription. The $20 Qoder Pro, originally only usable in CLI, now turns into a whole local API ecosystem. For Qoder's official side, is this a good thing or a bad thing?

Users gain more value, but perhaps it also means higher resource consumption. If everyone connects their subscription credits to Cursor and uses them intensively, can Qoder's compute costs handle it?

Currently, Qoder has not expressed an opinion on this tool. The disclaimer in the README is very clear — if the official forbids it, please 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 site: https://qoder.com/zh

Get Qoder Pro for free: Use my referral link to register and get one month of Pro for free — use GLM-5.2 freely. → https://qoder.com.cn/referral?referral_code=SOTtWioqAyhveY6SPPYSfjMIuKm4CnLz