Codex++ Hands-On: Complete Guide to Connecting Chinese LLMs Without a ChatGPT Account

Introduction

For developers in China, using OpenAI's Codex has always come with pain points like difficult account registration and unstable network connections. OpenAI Codex is OpenAI's AI programming assistant platform, specifically optimized for code generation based on GPT models. It supports the entire development workflow including code writing, debugging, and refactoring, and boasts a rich plugin ecosystem. However, since OpenAI's services are not directly accessible in mainland China, and account registration requires overseas phone number verification, the barrier to entry for Chinese developers has remained high.

Recently, an open-source project called Codex++ offers an elegant solution — connect directly to Chinese LLMs without a ChatGPT account, while still being able to use Computer Use and various plugin features.

This article is based on a hands-on review shared by a Bilibili content creator, compiling the complete configuration process to help you get started quickly.

What is Codex++: A Bridge for Connecting Chinese Models to Codex

Codex++ is an open-source management tool project whose core value lies in serving as a "bridge" — adapting Chinese LLM API interfaces to work within the Codex usage framework. From a technical perspective, Codex++ is essentially an API Adapter. Since the Codex client expects to communicate with a backend that conforms to the OpenAI API specification, while Chinese LLMs each have their own API formats and calling methods, Codex++ performs protocol translation in between — translating requests from the client into formats that Chinese models can understand, and converting the models' responses back into the format the client expects. This design pattern is known as an "API compatibility layer" in the open-source community, similar to the design philosophy of projects like OneAPI and LiteLLM.

This means:

• No ChatGPT account needed: Completely bypasses OpenAI's account system
• Supports Chinese models: You only need an API Key from a Chinese model provider
• Full functionality: Advanced features like Computer Use and various plugins all work normally

Computer Use is a capability that allows AI models to directly control a computer's desktop environment, first introduced by Anthropic in their Claude model. This technology allows AI to interact with graphical interfaces just like a human — through mouse clicks, keyboard input, screenshot recognition, and more — to complete complex tasks like installing software, operating browsers, and filling out forms. Integrating Computer Use into the Codex ecosystem means AI can not only write code but also directly execute operations in the development environment, significantly improving automation.

For developers who cannot reliably access OpenAI services, this is undoubtedly a practical alternative.

Preparation Before Configuration

Before starting the configuration, you need to prepare the following two things:

1. Codex++ open-source project: Obtain and install from the project repository
2. Chinese model API Key: Get an API key from your chosen Chinese LLM service provider

Currently, mainstream Chinese LLM API providers include: DeepSeek (known for its cost-effective reasoning models, especially DeepSeek-R1 which excels at code generation); Tongyi Qianwen (Alibaba Cloud, offering multiple model specifications from general conversation to code-specialized); and Zhipu AI's GLM series (with unique advantages in Chinese comprehension and tool calling). These providers all offer token-based billing for their API services, have low registration barriers, support Chinese phone numbers and payment methods, and typically offer free credits for new users to try.

Once these two items are ready, you can proceed to the formal configuration process.

Detailed Configuration Process for Connecting Chinese Models via Codex++

Step 1: Add Provider Configuration

1. Open the Codex++ management tool interface
2. Find and click the "Provider Configuration" option
3. Click the "Add" button in the right panel

Step 2: Fill in Model API Information

After entering the add page, fill in the following items step by step:

• Name: Customize an easily identifiable name (e.g., "DeepSeek" or "Tongyi Qianwen")
• Access Method: Select "Pure API"
• Enable Target Function: Check to enable
• Base URL: Enter the target model's API base address
• API Key: Enter the key obtained from your model service provider
• Type Selection: First select the "Chat" type

Regarding the Base URL, it is the root endpoint address of the API service, and all API requests use it as a prefix. For example, DeepSeek's Base URL is typically https://api.deepseek.com/v1, and Tongyi Qianwen's is https://dashscope.aliyuncs.com/compatible-mode/v1. Different providers have different Base URL formats, but providers following the OpenAI-compatible format usually end with /v1, which allows Codex++ to switch between different model backends by simply replacing the Base URL without modifying other configurations.

Step 3: Pull Available Models and Save

After filling in the information, click the "Pull from Upstream" button, and the system will automatically fetch the list of available models from that provider. This step actually sends a request to the provider's /models endpoint to retrieve information about all models accessible with the current API Key. After confirming everything is correct, click "Save" to complete the Chinese model configuration.

Step 4: Enable Configuration and Restart Service

1. Return to the provider list page
2. Find the provider you just configured and click "Use"
3. Click "Restart Codex++" in the upper right corner to apply the configuration

The restart operation is necessary because Codex++ needs to reload the configuration file and establish a connection pool with the new provider, ensuring subsequent requests are correctly routed to the target model service.

Step 5: Log In and Start Using

After restarting, you need to log into Codex++:

• Users with an account: Log in directly with your existing account
• Users without an account: Click "Other Methods" and enter your API Key to complete login

After successful login, your configured Chinese models will appear in the available model list, and all plugin features will be unlocked simultaneously.

Usage Experience and Notes

Core Advantages

Based on hands-on testing, the Codex++ solution has several clear advantages:

• Low configuration barrier: The entire process requires no VPN, with purely graphical operations
• No feature limitations: Advanced features like Computer Use can all be called normally
• Flexible switching: Supports configuring multiple providers with the ability to switch between different models at any time

Usage Notes

• API Key costs are charged by the corresponding Chinese model service provider — please understand the billing rules before use. Generally speaking, Chinese model API pricing is far lower than OpenAI's. For example, DeepSeek's input token price is about one-tenth of GPT-4's, but price differences between model specifications (such as standard vs. reasoning-enhanced versions) can be significant. It's recommended to choose appropriate models based on task complexity to control costs
• Different Chinese models have varying capabilities — choose the appropriate model based on your actual tasks. For code generation tasks, DeepSeek-Coder or Tongyi Qianwen's code-specialized versions are recommended; for general conversation and document processing, GLM-4 or Qianwen's general version are good choices; for complex programming problems requiring deep reasoning, reasoning-enhanced models like DeepSeek-R1 are suggested
• As an open-source project, it's recommended to follow its update announcements for the latest features and fixes. Also note that since Codex++ is a third-party adaptation tool, when OpenAI updates the Codex client protocol, you may need to wait for Codex++ to sync its updates before continuing normal use

Conclusion

Codex++ provides Chinese developers with a low-barrier path to using the Codex ecosystem. Through simple provider configuration, you can seamlessly integrate Chinese LLMs like DeepSeek and Tongyi Qianwen, enjoying full plugin and Computer Use functionality support. This approach of achieving cross-platform compatibility through an API adaptation layer demonstrates the open-source community's creativity in breaking down technical barriers. If you've been struggling with ChatGPT account issues and unable to experience Codex's powerful capabilities, give this solution a try.

Key Takeaways