GET STARTED

Setup guide

Choose the app you use, follow only its section, and send a short test request before expanding your usage. This guide keeps endpoints, protocols, models and recovery steps separate so your existing setup stays safe.

HTTPS is enabled. Before entering an API key, confirm that your browser shows no certificate warning. Keep full keys on your own device and never paste them into chat, email or screenshots.

CHOOSE A ROUTE

Which app are you using?

Pick the closest route. You do not need to read every section.

RESPONSES · CODINGChatGPT / Codex desktopUse the official ChatGPT desktop shell for Codex tasks, project files and tools.Open Codex steps → MESSAGES · CODINGClaude Desktop CodeUse the Code interface for project-level coding while keeping the original Claude login intact.Open Claude steps → CHAT · Q&AChatbox AIUse a compatible OpenAI endpoint for everyday chat and model switching.Open Chatbox steps → TERMINAL · IDECodex CLIUse an isolated CODEX_HOME in a terminal, VS Code or PyCharm.Open CLI steps →

BEFORE YOU START

Prepare these three things

A saved API key

Open API keys, create a key for this app and save the full value immediately. A full key can be claimed only once. No account yet? Sign up and verify your email.

Enough balance for a test

Open the top-up page, add a small amount and send one short request. Then open API usage to verify the model, tokens and charge.

A model for your protocol

Copy a complete model ID from pricing. Chat clients and Python use Chat, Codex uses Responses, and Claude Code uses Messages.

Chat modelsgpt-6-astragpt-5.6-solgpt-5.6-terragpt-5.6-luna

Responses modelsgpt-6-astragpt-5.6-solgpt-5.6-terragpt-5.6-luna

Messages modelsclaude-opus-5-5claude-opus-5claude-opus-4-8claude-sonnet-5

RESPONSES · DESKTOP

Configure Codex in ChatGPT desktop

This is the separate Codex mode, not regular Chat. Regular Chat continues to use your OpenAI account; do not enter this site's key on an OpenAI sign-in page.

Choose your operating system

Create your own key

Open API keys and save your own full key. Never share it with an administrator or paste it into a public issue.

Run the macOS installer once

Quit ChatGPT completely, then run this command in Terminal. The installer creates an isolated FengyuToken Codex entry and does not replace the official configuration.

set -e
INSTALLER="$HOME/Downloads/install-fengyutoken-codex-macos.sh"
curl -fsSL 'https://fengyutoken.com/downloads/fengyutoken-codex-macos.sh' -o "$INSTALLER"
chmod 700 "$INSTALLER"
"$INSTALLER"

Launch and verify

Open the new FengyuToken Codex desktop entry. Choose a listed Responses model, send “only reply CONNECTED”, then verify the new record in API usage.

Install the official ChatGPT app

Install the current ChatGPT desktop app from Microsoft Store, open it once, then quit it completely. Create and save your own key on the API keys page.

Run the one-line installer

[Net.ServicePointManager]::SecurityProtocol=[Net.ServicePointManager]::SecurityProtocol -bor [Net.SecurityProtocolType]::Tls12; $i=Join-Path $env:TEMP "fengyutoken-codex-win.ps1"; Invoke-WebRequest -UseBasicParsing "https://fengyutoken.com/downloads/fengyutoken-codex-windows.ps1" -OutFile $i; powershell.exe -NoProfile -ExecutionPolicy Bypass -File $i

Verify before switching modes

Do not launch regular Chat/Work with this key. Quit the official app, open the new FengyuToken Codex entry, send a short request and verify the usage record.

There is no official Codex desktop guide for Linux yet. Use Codex CLI in VS Code, PyCharm or a terminal instead.

MESSAGES · DESKTOP

Configure Claude Desktop Code

Use the Code interface in the official Claude desktop app. This route keeps the original Claude login separate from the gateway configuration.

Use only a listed Messages model. Model discovery should remain disabled when the app offers that setting. A simplified connection check may fail even when a real Code request works; verify with a short request and check the usage log.

Install Claude and the CLI

Install Claude from its official download page, open it once, then install Claude Code in Terminal.

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

Configure the gateway

Open API keys, use your own key, set the site root as the gateway endpoint and choose a Messages model from the list below.

claude-opus-5-5claude-opus-5claude-opus-4-8claude-sonnet-5

Send a real Code request

Open a blank project, run a small request and confirm both the reply and a matching entry in API usage. A 402 indicates insufficient balance; a 503 indicates service/model availability and should not be fixed by repeatedly paying.

CHAT · DESKTOP

Connect Chatbox AI

Install Chatbox from its official download page, then add a custom OpenAI-compatible provider.

Endpointhttps://fengyutoken.com/v1

API keyUse a key created on your API keys page.

ProtocolOpenAI-compatible Chat

ModelChoose a model marked Chat on the pricing page.

Send one short message first. If it works, open API usage and confirm the model, input/output tokens and charge.

RESPONSES · CLI / IDE

Use Codex CLI in a terminal, VS Code or PyCharm

Keep this setup isolated from your official Codex account by using a separate CODEX_HOME.

Install Codex CLI

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

Use an isolated profile

export CODEX_HOME="$HOME/.codex-fengyu-cli"
export FENGYU_API_KEY="paste-your-key-in-this-terminal-only"
codex -m "gpt-6-astra"

Do not save a real key in this page, a public repository or a shared shell history. Clear the variables or close the terminal to return to your official account.

Verify the first request

Send “only reply CONNECTED”, then check API usage for the same model.

MESSAGES · CLI / IDE

Use Claude Code in a terminal or editor

Create a dedicated shell profile for this gateway. Keep the variables in the current terminal only when possible.

export ANTHROPIC_BASE_URL="https://fengyutoken.com"
export ANTHROPIC_AUTH_TOKEN="paste-your-key-in-this-terminal-only"
export ANTHROPIC_MODEL="claude-opus-5-5"
claude

Use /model to select another listed Messages model. When you finish, close the terminal or unset the variables so your normal Claude configuration remains unchanged.

DEVELOPER API

Call a Chat model from Python

Use the OpenAI-compatible Chat endpoint with a key created for your own project.

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_KEY",
    base_url="https://fengyutoken.com/v1",
)
response = client.chat.completions.create(
    model="gpt-6-astra",
    messages=[{"role": "user", "content": "only reply CONNECTED"}],
)
print(response.choices[0].message.content)

Keep the real key in an environment variable in production. Confirm the request and charge in API usage.

TROUBLESHOOTING

Common checks

401 / authenticationConfirm the key is active, complete and sent to the correct endpoint. Chat clients use https://fengyutoken.com/v1; Claude Code uses the site root.

402 / insufficient balanceTop up a small amount, then retry once. Check the account balance and the usage record rather than repeating the request.

404 / wrong routeConfirm the protocol matches the app: Chat, Responses or Messages. Do not mix the Chat endpoint with Claude Code.

503 / service unavailableRecord the time and model ID, wait and retry later. A simplified connection probe is not the same as a real request.

REFERENCE · SECURITY

Endpoints and key safety

Chat / Responseshttps://fengyutoken.com/v1For Chatbox, OpenAI-compatible clients, Codex desktop and Codex CLI.

Messageshttps://fengyutoken.comFor Claude Code and native Messages clients. Do not append /v1.

Create separate keys by project, revoke exposed keys immediately, and never put a full key in browser JavaScript, screenshots, public repositories or shared chat. Start with a small request and reconcile the result in the usage log.