How to Connect DeepSeek to Codex: A Complete Guide (No ChatGPT Account Needed)

Why Replace Codex's Default Model with a Domestic LLM

OpenAI's Codex is one of the most talked-about AI agent tools for coding and automation. It's worth noting that Codex belongs to the "AI Agent" category, which is fundamentally different from conversational AI tools like ChatGPT. Traditional conversational AI is essentially "single-turn or multi-turn text exchange" — the user asks a question, the model outputs an answer, and the whole process is passively responsive. AI Agents, by contrast, introduce an "autonomous decision-making loop": they can independently break down complex tasks, continuously invoke system tools, control desktop interfaces, and execute multi-step operations without human intervention. This architecture is commonly referred to in academia as the ReAct (Reasoning + Acting) framework. Codex can automatically decompose a natural language instruction like "help me organize my Downloads folder and generate a report" into a series of concrete, executable steps — opening the folder, reading the file list, sorting by category, generating a report, and saving the output. This ability to "let AI do the work for you" is the core value that sets Codex apart from ordinary chatbots.

However, Codex is not user-friendly for people in China — the official login requires a ChatGPT Plus account, which means most users face a triple barrier: network access, account registration, and payment.

Fortunately, the community has found a viable workaround: using a third-party management tool to modify how Codex authenticates, and plugging in domestic large models like DeepSeek or Doubao as the backend. This allows users without a ChatGPT account to fully experience Codex's capabilities. This article is based on a hands-on tutorial from Bilibili and walks you through the complete setup process so you can get the "AI that controls your computer without writing code" workflow running within 2 hours.

In short, there are two paths:

• Path 1 (Easiest): You have a ChatGPT Plus account — just download Codex and log in directly.
• Path 2 (Domestic Alternative): No ChatGPT account — use the Codex++ management tool to connect DeepSeek or another domestic model.

This article focuses on Path 2, which is the realistic choice for the vast majority of users in China.

Step 1: Install Codex

Installing Codex itself is straightforward — no different from installing any regular software.

Works on Both Windows and Mac

Open your system's app store, search for Codex, and click "Free Download" to install. Mac users follow the exact same process in the App Store.

Important reminder: After installation, do NOT open Codex directly!

Launching it directly will prompt you to log in with a ChatGPT Plus account, and without one, you'll be stuck on the login screen with no way forward. The correct approach is to first complete the "model swap" configuration, then launch Codex through the management tool.

Step 2: Install the Codex++ Management Tool

Codex natively only supports ChatGPT login, which is why the community developed third-party tools. The Codex++ management tool has a clear purpose: it doesn't modify any of Codex's core functionality — it only changes the login mechanism, allowing you to configure a domestic large model as the backend.

This kind of "backend swap" is possible thanks to an important technical prerequisite: mainstream domestic models like DeepSeek and Doubao widely offer API interfaces that are compatible with the OpenAI API format. When OpenAI released the GPT series, it defined a standardized HTTP interface spec centered on the /v1/chat/completions endpoint, which accepts JSON requests in a specific format and returns structured responses. Because of OpenAI's first-mover advantage, this interface format has effectively become an industry standard — virtually all major domestic LLM providers have proactively implemented compatibility with this protocol to reduce migration costs for developers. In other words, these models "speak the same language." Codex++ only needs to redirect requests at the login stage from OpenAI's servers to a domestic model's servers — no changes to Codex itself are required. This open API ecosystem is also the foundation that allows domestic LLMs to integrate quickly into various third-party tools.

Download and Install

Users with access to international networks can download the latest release directly from the GitHub Releases page. If you can't access GitHub from within China, the tutorial author has pre-uploaded installation packages to a file-sharing platform (separate versions for Windows and Mac).

Install by following the default steps and clicking "Next" through the prompts. After installation, three apps will appear on your system:

• Codex (the main app)
• Codex++
• Codex++ 管理工具 (Management Tool — this is the one you need)

Be sure to search for and select Codex++ 管理工具 from the Start Menu — don't confuse it with the others. Once opened, the tool has many features, but we only need one: Model Provider Configuration.

Step 3: Configure DeepSeek as the Backend Model

This is the core of the entire process. The tutorial uses DeepSeek V4 as the example — it stands out for its excellent performance and extremely low cost, making it ideal for learning purposes.

Adding a Provider

In the management tool, click "Model Provider / Provider Configuration" on the left sidebar. Initially, only one default entry exists. Click "Add Provider" and fill in the following:

• Name: Any name you like — only visible to you, e.g., "My DeepSeek"
• Access Mode: Must select Pure API Access
• Base URL: Enter api.deepseek.com

"Pure API Access" is selected because Codex needs to access DeepSeek services through an API interface. The Base URL (base request address) tells Codex where to send requests — think of it as the "address" of the LLM service; all API calls will be directed to specific endpoints under this address. It's recommended to copy and paste this to avoid typos.

Obtaining and Entering the API Key

An API Key is essentially a "Bearer Token" authentication mechanism — possessing that string is treated as proof of a legitimate user; the server verifies the key itself, not the caller's identity. This is the sole credential for your DeepSeek account — a randomly generated encrypted string used to prove to the server that you are the one making the call. A few important notes: don't write the key in plaintext and push it to a public code repository, and it's a good idea to set a spending limit on the DeepSeek platform to prevent unexpected charges if the key is misused. In the use case described in this article, since the key is stored in the local management tool's configuration file, the security risk is manageable as long as you don't actively share your computer or configuration files.

Every time Codex executes a task through DeepSeek, it consumes a certain number of Tokens — the basic unit of measurement for text processing in large models. In Chinese contexts, one Chinese character typically corresponds to 1.5–2 tokens. API billing is usually split into "input tokens" and "output tokens" priced separately, with output tokens typically costing more (since generation requires more compute than comprehension). DeepSeek's billing is based on token usage, which is why you need to top up your account before using it. Compared to OpenAI's pricing, DeepSeek's token rates are extremely low — one of the key reasons it's so popular for learning and development in China.

To get your API Key:

1. Go to the DeepSeek official website and navigate to the "API Open Platform."
2. Click "API keys" on the left sidebar, then click "Create."
3. A key string will appear — this key is shown only once, copy it immediately (paste it into Notepad or a messaging app). Once you close the window, you cannot view it again — you'll need to create a new one.

Paste the copied key into the corresponding field in the management tool.

Fetching Models and Saving the Configuration

Leave the protocol option at its default, then click "Fetch from Upstream." This step tests the connection using the URL and key you entered — if it successfully retrieves DeepSeek's current list of available models (e.g., DeepSeek V4 Pro), the configuration is working.

After a successful fetch, there's no need to select a model on this screen. Scroll up and click "Save." Once saved, "My DeepSeek" will appear in the provider list. Hover over it and click "Use" — when the status shows "In Use," the configuration is complete.

Step 4: Top Up Your DeepSeek Account

Codex is just the interface shell — the real intelligence comes from the backend LLM. Whichever model you use, you need to fund that provider's account.

Go to the DeepSeek API platform's top-up page. Starting with just ¥10 is enough to get going. DeepSeek's token consumption is very low, and ¥10 is more than sufficient for learning purposes — it covers the majority of course content.

As a reference point, the tutorial author runs tasks "from morning to night, even while the computer is on overnight," and still only spends about ¥100 in a month. The reason behind this is that DeepSeek uses a MoE (Mixture of Experts) architecture — this architecture only activates a small portion of the model's parameters (typically 1/8 to 1/4 of the total) when processing each token, rather than activating all parameters like a traditional Dense model. This dramatically reduces the computational cost per inference while maintaining the model's overall capacity. This architectural innovation is directly reflected in API pricing, making DeepSeek's per-token cost as low as 1/10th (or less) of comparable GPT-4 products. For typical learners, actual consumption is far below heavy-use scenarios — ¥10 is more than enough to get started.

Step 5: Launch Codex via the Management Tool and Log In

Once configured, do not click the Codex icon to launch it directly. Instead, click "Restart Codex++" in the upper right corner of the Codex++ management tool — this is what triggers the modified login flow.

The tool will automatically launch the installed Codex application. On the login screen:

1. Do NOT select "Sign in with ChatGPT"
2. Select "Sign in with another method"
3. When prompted to enter an "OpenAI API key," paste your previously saved DeepSeek API Key and click "Continue"

The reason a DeepSeek key works in the "OpenAI API key" field is precisely because of the API protocol compatibility mentioned earlier — Codex sees request formats identical to OpenAI's and has no idea that the actual calls are going to DeepSeek's servers. This transparent substitution at the protocol level is the underlying logic that makes the entire solution work.

The first login may take a moment. Once you're inside the main interface, check the model selector in the lower right corner, confirm your configured model is available, and switch to DeepSeek V4 Pro (for better performance). At this point, you have a fully functional AI agent powered by a domestic large model that can control your computer.

What's Coming: Official Native Third-Party Model Support

It's worth keeping an eye on this: the Codex official team is reportedly planning to support native integration of select third-party large models. This trend aligns with the broader evolution of the AI tools ecosystem — more and more AI applications are moving from "locked to a single model" toward a Multi-Model Pluggable Architecture. This design decouples the "application capability layer" from the "intelligence backend layer": enterprise users handling sensitive data can opt for privately deployed models; cost-sensitive scenarios can switch to the lowest-cost inference model; tasks requiring maximum capability can switch to the current best-in-class flagship model. Similar designs already exist in tools like Cursor, Continue, and Open WebUI, where users can freely choose their backend model based on task requirements, cost budget, or data privacy needs. If Codex's official multi-model support launches, it would mean domestic models like DeepSeek and Doubao gain the same official support status as OpenAI, and the configuration experience would be significantly simplified — marking OpenAI's shift from a "closed ecosystem" to a "platform ecosystem."

Until that feature officially launches, connecting via a management tool remains the most reliable current approach. Once the official feature goes live, the configuration process will become even simpler — worth staying tuned.

Summary: Five Steps to Connect Codex with DeepSeek

The entire integration process can be summarized as: Install Codex → Install the management tool → Configure the API → Top up → Launch and log in via the management tool. The core logic is: Codex is the tool shell with AI Agent capabilities (using the ReAct framework for autonomous task decomposition and execution); DeepSeek is the backend LLM that actually provides the intelligence (using MoE architecture for low-cost, high-quality inference); and the Codex++ management tool leverages OpenAI API protocol compatibility between the two to handle the login bridge, seamlessly replacing the original OpenAI backend with a domestic model.

For users in China without a ChatGPT account, this solution requires as little as ¥10 to get started, delivering an AI agent capable of controlling your computer — outstanding value for money.

Key Takeaways