AI Coding Tools: Installation and Setup

We primarily use Codex, while Claude Code and OpenCode are also available. This guide covers installation and connection to the AI4EC-Lab Token Plan.

Last checked: October 10, 2026.

Contents

• Before you start
• Codex
• Claude Code
• OpenCode
• First-use check

Before you start

The commands below are for macOS, Linux, or a Linux terminal in Windows Subsystem for Linux (WSL). Run them in your own terminal. Native Windows shells use different installer commands and environment syntax.

1. Obtain a Token Plan account and generate your API key in the service portal. External users can use the account application form and QR code / 账号申请表与二维码.
2. Open AI Hub and copy a model ID compatible with your chosen client. Replace every MODEL_ID_FROM_AI_HUB below with that exact name. Claude Code needs a Claude model; Codex needs a model served through the Responses API.
3. After installing your chosen tool, set your key in the terminal where you will launch it. The following prompt hides it while you type and keeps it out of the command text:

printf 'Token Plan API key: '
read -r -s TOKEN_PLAN_API_KEY
printf '\n'
export TOKEN_PLAN_API_KEY

This environment variable is shared by the examples below. Set it again in a new terminal, or configure it in your personal shell environment. Never put an actual key in a shared project configuration.

Codex

Install

Install the CLI with the official standalone installer, then open a new terminal and check the version:

curl -fsSL https://chatgpt.com/codex/install.sh | sh

codex --version

Source: official Codex CLI documentation.

Connect to Token Plan

Open or create ~/.codex/config.toml. Merge the following settings with any existing configuration. Place the top-level model and model_provider settings before TOML table headings:

model = "MODEL_ID_FROM_AI_HUB"
model_provider = "token_plan"

[model_providers.token_plan]
name = "AI4EC-Lab Token Plan"
base_url = "https://chenglab.xmu.edu.cn/llm/v1"
env_key = "TOKEN_PLAN_API_KEY"
wire_api = "responses"

env_key names the environment variable containing your key. Codex uses the Responses API, so the selected gateway model must support it. This configuration sends requests through the lab gateway using your Token Plan credential. Sources: custom providers and provider configuration reference.

Start

From the same terminal where you set the key:

cd /path/to/your/project
codex

Ask it to explain the project first. Use /status to inspect the active session and /model to inspect model selection. These commands are described in the official command reference.

Claude Code

Install

Use the native installer, then open a new terminal and check the version:

curl -fsSL https://claude.ai/install.sh | bash

claude --version

If you use Homebrew on macOS, this is an alternative installation method:

brew install --cask claude-code

Choose one installation method. Source: official Claude Code quickstart.

Connect to Token Plan

The lab’s API-key configuration uses the gateway root:

export ANTHROPIC_BASE_URL="https://chenglab.xmu.edu.cn/llm/"
export ANTHROPIC_API_KEY="$TOKEN_PLAN_API_KEY"

This sends the credential as x-api-key. If the gateway team specifies bearer-token authentication instead, use ANTHROPIC_AUTH_TOKEN in place of ANTHROPIC_API_KEY; do not leave conflicting credentials active. Source: Claude Code gateway configuration.

Start

Choose an available Claude model from AI Hub:

cd /path/to/your/project
claude --model "MODEL_ID_FROM_AI_HUB"

Run /status and confirm the gateway base URL and API-key credential source. Shell exports apply to clients launched from that terminal. For later sessions, set the variables again or add the gateway configuration to your personal shell settings.

OpenCode

Install

Use the official installer:

curl -fsSL https://opencode.ai/install | bash

Alternatively, if Node.js and npm are already installed:

npm install -g opencode-ai

Choose one method, open a new terminal, and check:

opencode --version

This section uses the opencode-ai client and its provider configuration format. Source: official OpenCode installation guide.

Connect to Token Plan

Open or create the global configuration at ~/.config/opencode/opencode.json. Merge this example with existing settings and replace both occurrences of MODEL_ID_FROM_AI_HUB:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "token_plan": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "AI4EC-Lab Token Plan",
      "options": {
        "baseURL": "https://chenglab.xmu.edu.cn/llm/v1",
        "apiKey": "{env:TOKEN_PLAN_API_KEY}"
      },
      "models": {
        "MODEL_ID_FROM_AI_HUB": {
          "name": "Token Plan model"
        }
      }
    }
  },
  "model": "token_plan/MODEL_ID_FROM_AI_HUB"
}

This example uses an OpenAI-compatible Chat Completions model. For a model served through Responses, change npm to @ai-sdk/openai. Match the protocol to the model offered by the gateway. The API key is read from the environment rather than stored in this file. Sources: custom provider setup and global configuration.

Start

cd /path/to/your/project
opencode

Run /models and select the model under AI4EC-Lab Token Plan. You can also store a credential interactively using /connect → Other → provider ID token_plan; if you choose that method, remove options.apiKey from the example so the stored credential is used.

First-use check

Start with a short prompt such as: “Explain the structure of this project without changing any files.” Check the selected model and gateway configuration, then verify usage in the Token Plan portal.

For troubleshooting, budget checks, and account support, see the Cheng Group wiki FAQ.

Back to Token Plan · Back to Wiki