选好方式、电脑和要用的工具，下面的步骤会跟着变，一步一步照着做就行。
Claude Code 命令行：在终端里让 Claude 读代码、改项目、跑命令。 Claude Code 插件：在 VS Code / Cursor 编辑器里用 Claude Code，要先装好编辑器。 Codex 命令行：在终端里让 GPT 等模型写代码、跑命令。 Codex 插件：在 VS Code / Cursor 编辑器里用 Codex，要先装好编辑器。 Codex 桌面 App：单独窗口的桌面程序。下载约 1 GB。只支持 Apple 芯片（M1 及以后）的 Mac，下载约 0.7 GB。Linux 没有桌面版。
Claude 桌面 App 本站不提供：安装包和首次启动要下载的内容太大，国内下不动。
一共 6 步。第 4 步会弹出一个配置窗口，在里面勾上「Claude Code 命令行」，点「开始配置」。第 4 步脚本让你输入编号时，输入 1。装软件走国内镜像，不需要梯子；改任何文件前都会先备份，随时能恢复。
手动配置要自己装软件、改配置文件。不熟悉的话，建议切到自动配置。
Linux 没有 Codex 桌面 App，可以用 Codex 命令行 或 Codex 插件。
登录本站，左边菜单点「API 密钥」→「添加密钥」，分组选「全部模型」，建好后点「复制」。
别用「Claude Code 专用」分组的密钥配 Codex：这个分组没有 GPT 模型。用错了也没关系，脚本会提示你再粘贴一个。
还没有账号：用 QQ 邮箱或 163 邮箱注册（验证码没收到看看垃圾箱）。账户里要有余额：在「钱包」里充值或兑换兑换码；余额为 0 也能配好，但用的时候会报"余额不足"。密钥以 sk- 开头，不要发给别人，泄露了就删掉重建。
| 分组 | 适合 |
|---|---|
| 全部模型 | 推荐。Claude Code、Codex 都能用，一个密钥就够 |
| OpenAI | 只用 GPT 模型 |
| Claude Code 专用 | 只在 Claude Code 里用 Claude |
先装好 VS Code（到官网 code.visualstudio.com 免费下载）或 Cursor，并打开过一次。没装的话脚本会提示你先去装。
要 64 位 Windows 10 2004 或更新，C 盘留出约 4 GB（下载约 1 GB，装好约 2 GB）。空间不够时脚本会先告诉你，不会白下。
要 Apple 芯片（M1 及以后）、macOS 13 或更新，磁盘留出约 3.5 GB。
按键盘上的 Win 键，输入 PowerShell，按回车。
按 Command + 空格，输入终端（或 Terminal），按回车。
按 Ctrl + Alt + T，或者在应用列表里打开"终端"。
会打开一个蓝色或黑色的窗口。普通的就行，不用"以管理员身份运行"。Windows 10、11 都可以。
要有 curl 和 python3。Ubuntu / Debian 上先运行 sudo apt install curl python3。
irm https://tibo-api.com/setup.ps1 | iexbash -c "$(curl -fsSL https://tibo-api.com/setup.sh)"点「复制」，回到终端窗口，按 Ctrl + V（或点一下鼠标右键）Command + VCtrl + Shift + V 粘贴，再按回车。
第一次运行如果提示"这台 Mac 还没装苹果的命令行工具"：按回车同意，在弹出的窗口里点「安装」，等它装完（几分钟）再运行一次命令。
过几秒会弹出配置窗口，接下来在窗口里操作（这个终端窗口先别关）。
这行命令会从本站下载一键配置脚本并运行。想先看看它做了什么：tibo-api.com/setup.ps1tibo-api.com/setup.sh。
过几秒会弹出「Tibo API 一键配置」窗口（任务栏里也能找到）。照着编号做：
把第 1 步复制的密钥粘贴到「密钥」框里（Ctrl + V），点「检查」。下面出现绿色的「密钥可用」就对了。
以前配过本站的，窗口一打开就自动检查现在用的密钥，不用再粘贴；想换密钥，粘贴新的就行。
如果写着"这个密钥没有 GPT 模型……Codex 用不了"，说明粘贴的是只有 Claude 的分组：换一个「全部模型」分组的密钥。
勾上「Claude Code 命令行」。
可以多勾几个，一次配好；推荐的已经勾好了。已经配好本站的会变灰，写着"已配好本站"。没装 VS Code 时插件那两项是灰的，旁边有官网下载链接：装好、打开过一次后，关掉窗口再运行一次命令。
默认模型可以不改。
想默认用别的模型，在下拉里选；想让 /model 列表只留几个常用的，点「改列表…」。Codex 的列表选「全部」时，本站上了新模型会自动出现。
点「开始配置」，等进度条走完。命令行工具下载不到 300 MB。桌面 App 约 1 GB，限速下载，要几分钟。插件会自动装进 VS Code / Cursor。
网络断了会自动从断点续上；中途关掉也没关系，再运行一次会接着下，不用重下。
Windows 会弹窗问"你要允许此应用对你的设备进行更改吗？"，点「是」（它带一个后台服务，微软规定要你同意才能装），然后等一两分钟。
做完后，每个工具旁边写着"刚配好 ✓"或"没做成"，并弹出总结，告诉你怎么开始用。
没做成的，原因写在下方的记录里，也可以看下面的「遇到问题」。
「Claude Code / Codex 现在配的是别家……怎么办」：以前配过别家服务、或登录过官方账号才会出现。下拉里选「改成本站」（原来的配置先备份）、「另加一个命令 tibo-claude / tibo-codex 走本站，原来的不变」（只选了命令行时才有）或「不动它」。Codex 的插件和桌面 App 跟命令行共用一份配置，选了它们就只能整体改成本站。
「命令行工具装到 [路径] [换个位置…]」：第一次装命令行工具时才出现。不用管它就装到默认位置；想装到别的盘，点「换个位置…」选一个文件夹。
窗口没弹出来：先看看任务栏。还是没有，就在终端里输入 $env:TIBO_SETUP_TEXT=1 回车，再运行一次第 3 步的命令，改用文字界面（一问一答）。
下面是终端里会看到的，蓝色是你要输入的，编号的说明在下面：
按回车开始。
以前用一键配置配过的电脑，这里会换成一个菜单，见第 6 步。
粘贴第 1 步复制的密钥，按回车。粘贴后屏幕上什么都不显示，这是正常的（为了不让别人看到），直接回车就行。看到「[好] 密钥可用」就对了。
如果提示"这个密钥没有 GPT 模型……Codex 用不了"，说明粘贴的是只有 Claude 的分组：按提示再粘贴一个「全部模型」分组的密钥。
输入 1，按回车。
想一次配好几个工具，就输入多个编号，比如 1 3。已经配好的工具不会再列出来。
等它装完。命令行工具下载不到 300 MB。桌面 App 约 1 GB0.7 GB，限速下载，要几分钟。插件会自动装进 VS Code / Cursor。
网络断了会自动从断点续上；中途关掉窗口也没关系，再运行一次会接着下，不用重下。
Windows 会弹窗问"你要允许此应用对你的设备进行更改吗？"，点「是」（它带一个后台服务，微软规定要你同意才能装），然后等一两分钟。
如果电脑上装着旧版的 ChatGPT（聊天）App，会问是否替换：新版 ChatGPT 同时有聊天和 Codex，旧的会移到废纸篓。
绿色 [好] 是做好了的，后面写着怎么开始用；红色 [x] 是没做成的，后面写着原因和怎么办（也可以看下面的「遇到问题」）。
只有以前配过别家服务、或登录过官方账号才会问。1 改成走本站；2 原来的 claude 不变，另加一个 tibo-claude 命令走本站（只选了命令行时才有这一项）；3 不动它。
只有以前配过别家服务、或登录过官方账号才会问。1 改成走本站；2 原来的 codex 不变，另加一个 tibo-codex 命令走本站（只选了命令行时才有这一项）；3 不动它。Codex 的插件和桌面 App 跟命令行共用一份配置，选了它们就只能整体改成本站。
直接回车，装到默认位置。想装到别的盘：输入盘符，比如 D（会装到 D:\Tibo），或者完整路径。想装到别处，输入完整路径。只在第一次装时问，以后升级装回同一个地方；电脑上原来就装了的不会问。
先装 Node.js：到 nodejs.org 下载 LTS 版安装。然后打开终端，运行：
npm install -g @anthropic-ai/claude-code --registry=https://registry.npmmirror.com打开终端：按 Win 键，输入 PowerShell，回车按 Command + 空格，输入"终端"，回车按 Ctrl + Alt + T。命令里的地址是国内 npm 镜像，不需要梯子。装完运行 claude --version，能看到版本号就好了。提示没有权限（EACCES）的，在命令前面加 sudo 再运行。
打开 VS Code（或 Cursor），点左边的「扩展」图标（Ctrl + Shift + XCommand + Shift + X），搜 Claude Code，安装发布者是 Anthropic 的那个。
还没装 VS Code：到官网 code.visualstudio.com 免费下载。插件 ID：anthropic.claude-code。
先装 Node.js：到 nodejs.org 下载 LTS 版安装。然后打开终端，运行：
npm install -g @openai/codex --registry=https://registry.npmmirror.com打开终端：按 Win 键，输入 PowerShell，回车按 Command + 空格，输入"终端"，回车按 Ctrl + Alt + T。命令里的地址是国内 npm 镜像，不需要梯子。装完运行 codex --version，能看到版本号就好了。提示没有权限（EACCES）的，在命令前面加 sudo 再运行。
打开 VS Code（或 Cursor），点左边的「扩展」图标（Ctrl + Shift + XCommand + Shift + X），搜 Codex，安装发布者是 OpenAI 的那个。
还没装 VS Code：到官网 code.visualstudio.com 免费下载。插件 ID：openai.chatgpt。
建议用自动配置来装：运行一键配置的命令，在弹出的窗口里只勾「Codex 桌面 App」，点「开始配置」到「3/4 选择要配置的」只输入 5。它从本站下载官方安装包（国内能下）、核对校验值后安装，还会顺手写好下一步的配置。
已经装好了的，直接做下一步。
打开（没有就新建）下面这个文件，写入下面的内容，把 sk-你的密钥 换成第 1 步复制的密钥：
C:\Users\你的用户名\.claude\settings.json~/.claude/settings.json
{
"env": {
"ANTHROPIC_BASE_URL": "https://tibo-api.com",
"ANTHROPIC_AUTH_TOKEN": "sk-你的密钥",
"ANTHROPIC_MODEL": "claude-opus-5-5",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "claude-opus-5-5",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-sonnet-5-5",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "claude-sonnet-5-5",
"CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1"
},
"modelPicker": {
"options": [
{ "model": "claude-opus-5-5", "label": "Claude Opus 5.5" },
{ "model": "claude-sonnet-5-5", "label": "Claude Sonnet 5.5" },
{ "model": "gpt-6.1-sol", "label": "GPT-6.1 Sol", "behavesAs": "claude-sonnet-5-5" },
{ "model": "kimi-k3", "label": "Kimi K3", "behavesAs": "claude-sonnet-5-5" },
{ "model": "glm-5.3", "label": "GLM-5.3", "behavesAs": "claude-sonnet-5-5" }
],
"replaceBuiltInOptions": true
}
}env 是地址、密钥和默认模型。想默认用 GPT，把 ANTHROPIC_MODEL 和 ANTHROPIC_DEFAULT_OPUS_MODEL 改成 gpt-6.1-sol。modelPicker 是 /model 列表。不是 Claude 的模型要带 "behavesAs": "claude-sonnet-5-5" 才会列出来，照着写就行；还能加 deepseek-v4-pro-0813、qwen3.8-max、MiniMax-M3、grok-4.7 等（Gemini 在 Claude Code 里用不了）。env、modelPicker 这两项，注意逗号；有 ANTHROPIC_API_KEY 的删掉（和 ANTHROPIC_AUTH_TOKEN 同时在会冲突）。在 VS Code 里按 Ctrl + Shift + PCommand + Shift + P，输入 Open User Settings (JSON)，回车，在最外层的大括号里加上（密钥同样换成自己的）：
"claudeCode.disableLoginPrompt": true,
"claudeCode.environmentVariables": [
{ "name": "ANTHROPIC_BASE_URL", "value": "https://tibo-api.com" },
{ "name": "ANTHROPIC_AUTH_TOKEN", "value": "sk-你的密钥" },
{ "name": "ANTHROPIC_MODEL", "value": "claude-opus-5-5" }
],用 Cursor 的，在 Cursor 里同样操作。前后有别的设置时注意逗号（最后一项后面不能有逗号）。
打开（没有就新建）下面这个文件，写入下面的内容，把 sk-你的密钥 换成第 1 步复制的密钥：
C:\Users\你的用户名\.codex\config.toml~/.codex/config.toml
model = "gpt-6.1-sol"
model_provider = "tibo"
[features]
api_key_model_discovery = true
[model_providers.tibo]
name = "Tibo API"
base_url = "https://tibo-api.com/v1"
wire_api = "responses"
experimental_bearer_token = "sk-你的密钥"
model_catalog_url = "https://tibo-api.com/dl/codex/models"[ ] 段之前）。原来有 model、model_provider、profile 的，删掉或在行首加 #。model_catalog_url 让 /model 列表从本站拿：按你的密钥给，有 GPT、Claude、kimi、glm 等，本站上了新模型会自动出现。[features] 下面那一行是给 Codex 桌面 App、VS Code 插件用的：少了它，模型下拉里只有「自定义」。文件里原来就有 [features] 段的，把这一行加进那一段，别再写第二个 [features]。新开一个终端窗口，进入你的项目文件夹，输入 claude 回车。
之前开着的终端找不到新装的命令，所以要新开。第一次会问是否信任这个文件夹，选 Yes。选了"另加命令"的，输入 tibo-claude。小技巧：在文件资源管理器里打开项目文件夹，在地址栏输入 powershell 回车，就会在这个文件夹里打开 PowerShell。
把 VS Code / Cursor 完全关掉再打开，点编辑器右上角的 Claude 图标开始对话。
不需要登录 Claude 账号。如果让你登录，说明配置还没生效：把编辑器完全关掉再开一次，还不行就检查上面两处配置。
新开一个终端窗口，进入你的项目文件夹，输入 codex 回车。
之前开着的终端找不到新装的命令，所以要新开。选了"另加命令"的，输入 tibo-codex。小技巧：在文件资源管理器里打开项目文件夹，在地址栏输入 powershell 回车，就会在这个文件夹里打开 PowerShell。
把 VS Code / Cursor 完全关掉再打开，在左侧栏点 Codex 图标。
如果出现登录界面，选用 API 密钥登录，粘贴本站密钥。
在开始菜单里在「应用程序」（或启动台）里打开 ChatGPT（Codex 桌面 App 装好后就叫这个名字）。
如果让你登录，选用 API 密钥，粘贴本站密钥。
换模型：在对话里输入 /model 回车，从列表里选。除了 Claude、GPT，还有 kimi、glm、deepseek、qwen、MiniMax、grok 等（你在 modelPicker 里写了的）。默认用 claude-opus-5-5。
换模型：输入 /model 回车，从列表里选。除了 GPT，还有 Claude、kimi、glm、deepseek、qwen、MiniMax、grok 等。默认用 gpt-6.1-sol。选好模型接着选思考强度（别家模型也能选，默认「中」）。
换模型：在输入框旁边的模型菜单里选，列表和 Codex 命令行是同一份：除了 GPT，还有 Claude、kimi、glm、deepseek 等。默认用 gpt-6.1-sol。菜单里还能调思考强度（别家模型也能调，默认「中」）。最高一档「Max」默认不显示，要用的话到 设置 → 配置 → 模型功能 → 可用推理强度 里勾上 Max。
Gemini 实测在 Claude Code、Codex 里用不了，所以没放进列表；在聊天客户端、自己的程序里照常能用。想换默认模型、或者列表里只留几个常用的，见下一步。想换默认模型，改配置文件里的 ANTHROPIC_MODELmodel。
新开一个终端，输入 tibo 回车（或者再运行一次第 3 步的命令），会再打开配置窗口：
/model 列表里放哪些，再点「开始配置」。选的会记住，换密钥也不丢。
加装别的勾上还没配的工具（已配好的是灰的），点「开始配置」，直接用已配置的密钥。
升级有新版时多一个「升级到最新」按钮，把本脚本装的命令行工具、桌面 App 升到最新。
什么都不改直接点「开始配置」：按现在的设置重写一遍配置，/model 列表也更新到本站最新的。
新开一个终端，输入 tibo 回车（或者再运行一次第 3 步的命令）。它先检查电脑，然后出一个菜单：
/model 列表里放哪些（直接回车 = 全部能用的）。选的会记住，换密钥也不丢。
3 装别的只列还没配的工具，已经配好的不会再出现；直接用已配置的密钥，不再问。
4 升级有新版时才出现，把本脚本装的命令行工具、桌面 App 升到最新。
改完要把正在开着的 Claude Code、Codex、编辑器关掉重开才生效。Codex 的列表由本站提供，本站上了新模型会自动出现；Claude Code 的列表写在它自己的设置里，本站上新后在窗口里直接点「开始配置」再选一次 2就能加进去。
2026 年 10 月 10 日之前配过的：/model 列表里只有 Claude、GPT，Codex 桌面 App、VS Code 插件的模型下拉里只有「自定义」，也还没有 tibo 命令。再运行一次第 3 步的命令，在窗口里直接点「开始配置」选 2，一路回车就更新好了。
$env:TIBO_SETUP_TEXT=1 回车，再运行一次命令，改用文字界面（一问一答）。anthropic.claude-code（Claude Code）或 openai.chatgpt（Codex）手动安装。配置脚本已经写好了。5。已经下好的安装包会直接用，不用重下。tibo（没有这个命令就再运行一次第 3 步的命令），在窗口里直接点「开始配置」选 2更新。tibo（没有这个命令就再运行一次第 3 步的命令），打开就会自动补上；然后把 App / VS Code 完全关掉再打开。tibo。R（配过本站的在菜单里选 R），选要恢复的那一次。恢复前的状态也会再存一份，恢复错了还能改回来。C:\Users\你的用户名\AppData\Local\Programs\Tibo（bin 里是 claude.exe，codex 里是 Codex），加进你自己的 PATH 环境变量~/.tibo（bin、codex），在 .zshrc / .bashrc 里加一段 # >>> tibo-api >>>。第一次装时可以换位置，换过的会记住。~/.claude/settings.json（Windows 是 C:\Users\你的用户名\.claude\settings.json）里的 env（地址、密钥、默认模型）和 modelPicker（/model 列表）两项，其他设置不动。settings.json 里加 claudeCode.environmentVariables 和 claudeCode.disableLoginPrompt，注释和其他设置都保留。~/.codex/config.toml 里加两段带 # >>> tibo-api 标记的内容；原来的 model、model_provider、profile 只是注释掉，不删。命令行、插件、桌面 App 都读这一份。其中 model_catalog_url 让 Codex 的 /model 列表从本站拿（按你的密钥给，上新模型自动有）；[features] 段里的 api_key_model_discovery = true 是 Codex 桌面 App、VS Code 插件的模型下拉要的（原来就有 [features] 段的，加进那一段）。tibo-claude（读 ~/.tibo/claude-settings.json）、tibo-codex，放在命令行程序的 bin 文件夹里。tibo 放在命令行程序的 bin 文件夹里，运行时从本站拿最新的一键配置脚本；在菜单里选过的默认模型和列表记在 ~/.tibo/models.json。~/.tibo/backup/日期-时间/。R 恢复配置；再删掉命令行程序的文件夹；在开始菜单搜"编辑账户的环境变量"，把 Path 里 Tibo 的两项删掉删掉 .zshrc / .bashrc 里 tibo-api 那一段。Codex 桌面 App：在「设置 → 应用」里卸载 ChatGPT把「应用程序」里的 ChatGPT 拖进废纸篓。settings.json，不是 settings.json.txt。config.toml 的路径和格式：model_provider = "tibo" 要在文件最上面、任何 [ ] 段之前；密钥要完整（sk- 开头）。config.toml 里要有 [features] 段的 api_key_model_discovery = true（见上面的配置）。改完把 App / VS Code 完全关掉再打开。modelPicker 里，并带 "behavesAs": "claude-sonnet-5-5"。Codex：[model_providers.tibo] 里要有 model_catalog_url 那一行。改完把工具关掉重开。CC Switch 是一个免费的桌面小工具（Windows / macOS / Linux），帮你管理 Claude Code、Codex 的接口配置，在几家服务之间一键切换。本站可以一键导入到 CC Switch，导入后供应商卡片上会直接显示你的账户余额，每 30 分钟自动刷新。
.msi，Mac 选 .dmg），先打开一次。claude-opus-5-5 等，Codex 选 gpt-6.1-sol 等），点「打开 CC Switch」。浏览器问要不要打开 CC Switch，点允许。tibo-api.com 查余额。点「导入」。「添加供应商」时，Claude 的地址填 https://tibo-api.com，Codex 的填 https://tibo-api.com/v1，密钥填 sk-...。
看余额：在这个供应商上点「配置用量查询」，模板选「通用模板」，Base URL 填 https://tibo-api.com（不带 /v1），API Key 填同一个密钥，打开「启用用量查询」后保存。单位那里显示 USD，其实是人民币元，数字就是你的余额；想显示成「元」，把脚本里的 "USD" 改成 "元"。
已经用「自动配置」（一行命令或 tibo）配好的电脑不需要 CC Switch。两个一起用时，CC Switch 切换供应商会改掉自动配置写好的 Claude Code 模型列表（/model 里可能只剩一个模型）；想恢复，运行 tibo 选「换模型」。
显示的是账户余额；如果这个密钥设了额度上限，显示的是它还能用多少（两者取小的那个）。
支持"OpenAI 兼容"或 Anthropic 接口的软件都能用，填这三项：
https://tibo-api.com/v1（OpenAI 格式的软件填这个）https://tibo-api.com（Anthropic 格式、Claude Code 填这个，不带 /v1）
密钥「API 密钥」页复制的 sk-...
模型名GPT：gpt-6.1-sol、gpt-6-astra；Claude：claude-opus-5-5、claude-sonnet-5-5 等；其他：gemini-3.8-flash、kimi-k3、glm-5.3、qwen3.8-max、deepseek-v4-pro-0813、MiniMax-M3、grok-4.7 等。全部模型和价格见「模型广场」。
精选分组的 Claude 模型只能在 Claude Code 里用：在聊天客户端、自己写的程序里调用会被拒绝。下面这些软件请用 GPT 或其他模型；一定要用 Claude 的，可以建一个经济分组「Claude（Kiro 号池）」的密钥，用名字带 -kiro 的模型（价格低，质量不保证）。
sk-...；API 地址填 https://tibo-api.com（软件会自己补 /v1；如果软件要求完整地址，就填 https://tibo-api.com/v1）。gpt-6.1-sol），打开开关，回到对话页选这个模型。from openai import OpenAI
client = OpenAI(base_url="https://tibo-api.com/v1", api_key="sk-你的密钥")
resp = client.chat.completions.create(
model="gpt-6.1-sol",
messages=[{"role": "user", "content": "你好"}],
)
print(resp.choices[0].message.content)流式输出、工具调用（function calling）、Responses 接口（client.responses.create）都支持，用法和官方一样。
sk- 开头），或者密钥已被删除、过期、禁用。到「API 密钥」页重新复制。/v1，Claude Code 不带。max_tokens）。每一次调用花了多少，在控制台的「使用日志」里都能看到。其他问题请加 QQ 群 1126821195，或发邮件到 support@tibo-api.com。
Pick how you want to set up, your computer and your tool. The steps below change to match; just follow them one by one.
Claude Code CLI: let Claude read code, edit your project and run commands in the terminal. Claude Code extension: use Claude Code inside VS Code / Cursor. Install the editor first. Codex CLI: let GPT and other models write code and run commands in the terminal. Codex extension: use Codex inside VS Code / Cursor. Install the editor first. Codex desktop app: a separate window app. About a 1 GB download.Apple-silicon Macs (M1 and later) only, about a 0.7 GB download.There is no Linux version.
We don't provide the Claude desktop app: its installer and first-run download are too large to download reliably in mainland China.
6 steps in all. In step 4 a setup window pops up: tick "Claude Code 命令行" and click "开始配置" (Start).In step 4, when the script asks for numbers, type 1. Software comes from mirrors in mainland China, so no VPN is needed; every file is backed up before it's changed, so you can always undo.
Manual setup means installing the software and editing config files yourself. If you're not comfortable with that, switch to one-click setup.
There is no Codex desktop app for Linux. Use the Codex CLI or the Codex extension instead.
Sign in to our site, click API Keys → Add key in the left menu, set the group to "All models" (全部模型), then click Copy.
Don't use a key from the "Claude Code 专用" group for Codex: that group has no GPT models. If you do, no harm done: the script tells you and asks for another key.
No account yet? Sign up with a QQ Mail or NetEase 163 Mail address (if the code doesn't arrive, check your spam folder). Your account needs a balance: top up or redeem a code under Wallet. With a zero balance the setup still finishes, but requests fail with "insufficient balance". Keys start with sk-. Never share your key; if it leaks, delete it and create a new one.
| Group | Use it for |
|---|---|
| All models (全部模型) | Recommended. Works in both Claude Code and Codex; one key is enough |
| OpenAI | GPT models only |
| Claude Code 专用 | Claude in Claude Code only |
Install VS Code first (free from code.visualstudio.com) or Cursor, and open it once. If it's missing, the script tells you to install it first.
Needs 64-bit Windows 10 version 2004 or later, and about 4 GB free on drive C (about 1 GB to download, about 2 GB once installed). If there isn't enough space, the script tells you before downloading anything.
Needs an Apple-silicon Mac (M1 or later) with macOS 13 or later, and about 3.5 GB of free disk space.
Press the Win key, type PowerShell and press Enter.
Press Command + Space, type Terminal and press Enter.
Press Ctrl + Alt + T, or open "Terminal" from your app list.
A blue or black window opens. The normal one is fine; you don't need "Run as administrator". Windows 10 and 11 both work.
You need curl and python3. On Ubuntu / Debian, run sudo apt install curl python3 first.
irm https://tibo-api.com/setup.ps1 | iexbash -c "$(curl -fsSL https://tibo-api.com/setup.sh)"Click Copy, go back to the terminal window, paste with Ctrl + V (or a right-click)Command + VCtrl + Shift + V, then press Enter.
The first time, if it says Apple's command line tools are not installed: press Enter to agree, click "Install" in the window that pops up, wait until it finishes (a few minutes), then run the command again.
After a few seconds a setup window pops up; the next step happens there (keep this terminal window open).
The command downloads the one-click setup script from our site and runs it. To see what it does first, open tibo-api.com/setup.ps1tibo-api.com/setup.sh.
After a few seconds a window called "Tibo API 一键配置" opens (you can also find it on the taskbar). Its text is in Chinese; follow the numbers:
"本站密钥" (your key): paste the key from step 1 into the box (Ctrl + V) and click "检查" (Check). Green text "密钥可用" below means the key works.
If you set up with us before, the window checks your current key as soon as it opens, so there's no need to paste it again; to change keys, paste the new one.
If it says "这个密钥没有 GPT 模型……Codex 用不了" (this key has no GPT models), you pasted a Claude-only key: use a key from the "All models" group.
"要配置的工具" (tools): tick "Claude Code 命令行".
You can tick several and set them up at once; the recommended ones are already ticked. Tools already set up with us are greyed out with "已配好本站". Without VS Code the two extension rows are greyed out, with a link to the download page: install it, open it once, then close the window and run the command again.
"模型" (models): you can leave the defaults.
Pick another default from the drop-down, or click "改列表…" (edit list) to keep only a few models in the /model list. If the Codex list is set to all ("全部"), new models we add show up automatically.
Click "开始配置" (Start) and wait for the progress bar to finish. The CLI tools are under 300 MB to download.The desktop app is about 1 GB and downloads at a limited speed, so it takes a few minutes.The extension is installed into VS Code / Cursor automatically.
If the network drops, it resumes where it stopped. If you close it halfway, run the command again and it continues; nothing is downloaded twice.
Windows asks "Do you want to allow this app to make changes to your device?". Click Yes (the app includes a background service, which Windows only installs with your consent), then wait a minute or two.
When it's done, each tool shows "刚配好 ✓" (done) or "没做成" (failed), and a summary pops up telling you how to start.
For anything that failed, the reason is in the log at the bottom; see also "Something went wrong" below.
"Claude Code / Codex 现在配的是别家……怎么办" (it's set up for another service; what now?): only shown if you used another service or signed in to an official account before. In the drop-down, pick "改成本站" (switch to us; the old config is backed up first), "另加一个命令 tibo-claude / tibo-codex 走本站，原来的不变" (add a separate command and keep the old one; offered only when you picked the CLI alone) or "不动它" (leave it alone). The Codex extension and desktop app share one config with the CLI, so if you pick them it can only be switched as a whole.
"命令行工具装到 [path] [换个位置…]" (where to install the CLI tools): only shown on the first install. Leave it for the default location, or click "换个位置…" to pick a folder on another drive.
No window? Check the taskbar first. If it's not there, type $env:TIBO_SETUP_TEXT=1 in the terminal, press Enter, and run the step-3 command again to use the text interface (questions and answers) instead.
This is what the terminal shows (the script's messages are in Chinese). Blue is what you type; the numbers are explained below:
"Press Enter to start." Press Enter.
If you set up this computer with the script before, you get a menu here instead; see step 6.
"Paste your key." Paste the key from step 1 and press Enter. Nothing shows on screen when you paste. That's normal (it keeps your key hidden); just press Enter. "[好] 密钥可用" means the key works.
If it says "这个密钥没有 GPT 模型……Codex 用不了" (this key has no GPT models), you pasted a Claude-only key: paste a key from the "All models" group when asked.
"Type the numbers." Type 1 and press Enter.
To set up several tools at once, type several numbers, such as 1 3. Tools that are already set up are no longer listed.
"Installing and configuring." Wait for it to finish. The CLI tools are under 300 MB to download.The desktop app is about 1 GB0.7 GB and downloads at a limited speed, so it takes a few minutes.The extension is installed into VS Code / Cursor automatically.
If the network drops, it resumes where it stopped. If you close the window halfway, run the command again and it continues; nothing is downloaded twice.
Windows asks "Do you want to allow this app to make changes to your device?". Click Yes (the app includes a background service, which Windows only installs with your consent), then wait a minute or two.
If the old ChatGPT (chat) app is installed, it asks whether to replace it: the new ChatGPT has both chat and Codex, and the old one goes to the Trash.
"Done." Green [好] lines are finished, with how to start using each one; red [x] lines didn't work, with the reason and what to do (see also "Something went wrong" below).
Only asked if you used another service or signed in to an official account before. 1 switches it to us; 2 keeps your claude as it is and adds a separate tibo-claude command that uses us (offered only when you picked the CLI alone); 3 leaves it alone.
Only asked if you used another service or signed in to an official account before. 1 switches it to us; 2 keeps your codex as it is and adds a separate tibo-codex command that uses us (offered only when you picked the CLI alone); 3 leaves it alone. The Codex extension and desktop app share one config with the CLI, so if you pick them it can only be switched as a whole.
"Where to install the CLI tools?" Press Enter for the default location. To use another drive, type a drive letter such as D (installs to D:\Tibo) or a full path.To use another place, type a full path. It's asked only on the first install; later updates go to the same place. Tools that were already installed are not asked about.
Install Node.js first: download the LTS version from nodejs.org. Then open a terminal and run:
npm install -g @anthropic-ai/claude-code --registry=https://registry.npmmirror.comTo open a terminal: press Win, type PowerShell, press Enterpress Command + Space, type Terminal, press Enterpress Ctrl + Alt + T. The address in the command is the npm mirror in mainland China, so no VPN is needed. When it finishes, run claude --version; if it prints a version number, you're done. If it says permission denied (EACCES), put sudo in front of the command and run it again.
Open VS Code (or Cursor), click the Extensions icon on the left (Ctrl + Shift + XCommand + Shift + X), search for Claude Code and install the one published by Anthropic.
No VS Code yet? Download it free from code.visualstudio.com. Extension ID: anthropic.claude-code.
Install Node.js first: download the LTS version from nodejs.org. Then open a terminal and run:
npm install -g @openai/codex --registry=https://registry.npmmirror.comTo open a terminal: press Win, type PowerShell, press Enterpress Command + Space, type Terminal, press Enterpress Ctrl + Alt + T. The address in the command is the npm mirror in mainland China, so no VPN is needed. When it finishes, run codex --version; if it prints a version number, you're done. If it says permission denied (EACCES), put sudo in front of the command and run it again.
Open VS Code (or Cursor), click the Extensions icon on the left (Ctrl + Shift + XCommand + Shift + X), search for Codex and install the one published by OpenAI.
No VS Code yet? Download it free from code.visualstudio.com. Extension ID: openai.chatgpt.
We suggest installing it with one-click setup: run the setup command and tick only "Codex 桌面 App" in the window that pops up, then click "开始配置" (Start)type only 5 at "3/4 选择要配置的". It downloads the official installer from our site (works in mainland China), checks the checksum, installs it, and writes the config from the next step for you.
Already installed? Go straight to the next step.
Open (or create) the file below and put in the following, replacing sk-your-key with the key from step 1:
C:\Users\your-user-name\.claude\settings.json~/.claude/settings.json
{
"env": {
"ANTHROPIC_BASE_URL": "https://tibo-api.com",
"ANTHROPIC_AUTH_TOKEN": "sk-your-key",
"ANTHROPIC_MODEL": "claude-opus-5-5",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "claude-opus-5-5",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-sonnet-5-5",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "claude-sonnet-5-5",
"CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1"
},
"modelPicker": {
"options": [
{ "model": "claude-opus-5-5", "label": "Claude Opus 5.5" },
{ "model": "claude-sonnet-5-5", "label": "Claude Sonnet 5.5" },
{ "model": "gpt-6.1-sol", "label": "GPT-6.1 Sol", "behavesAs": "claude-sonnet-5-5" },
{ "model": "kimi-k3", "label": "Kimi K3", "behavesAs": "claude-sonnet-5-5" },
{ "model": "glm-5.3", "label": "GLM-5.3", "behavesAs": "claude-sonnet-5-5" }
],
"replaceBuiltInOptions": true
}
}env holds the address, key and default model. To default to GPT, change ANTHROPIC_MODEL and ANTHROPIC_DEFAULT_OPUS_MODEL to gpt-6.1-sol.modelPicker is the /model list. Models that aren't Claude need "behavesAs": "claude-sonnet-5-5" to be listed; just copy it. You can also add deepseek-v4-pro-0813, qwen3.8-max, MiniMax-M3, grok-4.7 and more (Gemini doesn't work in Claude Code).env and modelPicker entries, and mind the commas. Remove ANTHROPIC_API_KEY if it's there (it conflicts with ANTHROPIC_AUTH_TOKEN).In VS Code, press Ctrl + Shift + PCommand + Shift + P, type Open User Settings (JSON), press Enter, and add this inside the outermost braces (again with your own key):
"claudeCode.disableLoginPrompt": true,
"claudeCode.environmentVariables": [
{ "name": "ANTHROPIC_BASE_URL", "value": "https://tibo-api.com" },
{ "name": "ANTHROPIC_AUTH_TOKEN", "value": "sk-your-key" },
{ "name": "ANTHROPIC_MODEL", "value": "claude-opus-5-5" }
],On Cursor, do the same in Cursor. Mind the commas around other settings (no comma after the last entry).
Open (or create) the file below and put in the following, replacing sk-your-key with the key from step 1:
C:\Users\your-user-name\.codex\config.toml~/.codex/config.toml
model = "gpt-6.1-sol"
model_provider = "tibo"
[features]
api_key_model_discovery = true
[model_providers.tibo]
name = "Tibo API"
base_url = "https://tibo-api.com/v1"
wire_api = "responses"
experimental_bearer_token = "sk-your-key"
model_catalog_url = "https://tibo-api.com/dl/codex/models"[ ] section). If the file already has model, model_provider or profile, delete those lines or put # in front of them.model_catalog_url makes the /model list come from our site: it's built for your key, includes GPT, Claude, kimi, glm and more, and new models show up automatically.[features] is for the Codex desktop app and VS Code extension: without it, their model menu only shows "Custom". If the file already has a [features] section, add the line to that section instead of writing a second [features].Open a new terminal window, go to your project folder, type claude and press Enter.
Windows that were already open can't find the new command, so open a new one. The first time it asks whether you trust the folder; choose Yes. If you chose the separate command, type tibo-claude. Tip: open your project folder in File Explorer, type powershell in the address bar and press Enter to open PowerShell right in that folder.
Fully quit and reopen VS Code / Cursor, then click the Claude icon at the top right of the editor.
No Claude account sign-in is needed. If it asks you to sign in, the settings haven't taken effect yet: fully quit the editor and open it again, and if that doesn't help, check the two configs above.
Open a new terminal window, go to your project folder, type codex and press Enter.
Windows that were already open can't find the new command, so open a new one. If you chose the separate command, type tibo-codex. Tip: open your project folder in File Explorer, type powershell in the address bar and press Enter to open PowerShell right in that folder.
Fully quit and reopen VS Code / Cursor, then click the Codex icon in the left sidebar.
If a sign-in screen appears, choose to sign in with an API key and paste your key from our site.
Open ChatGPT from the Start menufrom Applications (or Launchpad). That's the name of the Codex desktop app once installed.
If it asks you to sign in, choose API key and paste your key from our site.
Switching models: type /model in the chat and pick from the list. Besides Claude and GPT it has kimi, glm, deepseek, qwen, MiniMax, grok and more (whatever you put in modelPicker). The default is claude-opus-5-5.
Switching models: type /model and pick from the list. Besides GPT it has Claude, kimi, glm, deepseek, qwen, MiniMax, grok and more. The default is gpt-6.1-sol. After picking a model you choose the reasoning effort (other vendors' models too; the default is medium).
Switching models: use the model menu next to the input box. It's the same list as the Codex CLI: besides GPT it has Claude, kimi, glm, deepseek and more. The default is gpt-6.1-sol. The same menu sets the reasoning effort (other vendors' models too; the default is medium). The top level, Max, is hidden by default: turn it on under Settings → Configuration → Model features → Available reasoning efforts.
In our tests Gemini doesn't work in Claude Code or Codex, so it's left out of the list; it works as usual in chat apps and your own programs. To change the default model or keep only a few models in the list, see the next step. To change the default model, edit ANTHROPIC_MODELmodel in the config file.
Open a new terminal and type tibo (or run the step-3 command again). The setup window opens again:
/model list, then click Start. Your choices are remembered, even when you change keys.
Add toolsTick the tools not set up yet (the ones already done are greyed out) and click Start; it reuses your configured key.
UpgradeWhen there's a newer version, an "升级到最新" (upgrade) button appears; it upgrades the CLI tools and desktop app installed by this script.
Change nothingJust click Start: it rewrites the config with your current settings and refreshes the /model list to our latest.
Open a new terminal and type tibo (or run the step-3 command again). It checks your computer, then shows a menu (in Chinese):
/model list (Enter = all usable ones). Your choices are remembered, even when you change keys.
3 Add toolsLists only the tools not set up yet; ones already done don't appear again. It reuses your configured key without asking.
4 UpgradeAppears only when there's a newer version; upgrades the CLI tools and desktop app installed by this script.
Fully quit and reopen any running Claude Code, Codex or editor for changes to take effect. The Codex list comes from our site, so new models show up automatically; the Claude Code list lives in its own settings, so click Start in the windowchoose 2 again to add models we add later.
Set up before 10 Oct 2026? Your /model list only has Claude and GPT, the model menu in the Codex desktop app and VS Code extension only shows "Custom", and there's no tibo command yet. Run the step-3 command again and just click Start in the windowchoose 2, pressing Enter through the questions.
$env:TIBO_SETUP_TEXT=1 in the terminal, press Enter, and run the command again to use the text interface.anthropic.claude-code (Claude Code) or openai.chatgpt (Codex) and install it by hand. The config is already written.5 at "3/4 选择要配置的". The installer you already downloaded is reused.tibo (or the step-3 command if tibo doesn't exist) and click Start in the windowchoose 2.tibo (or the step-3 command if tibo doesn't exist); it adds the line as soon as it starts. Then fully quit and reopen the app / VS Code.tibo.R at the start (or pick R in the menu if you set up before), then pick the run to restore. The current state is saved again first, so you can undo the restore too.C:\Users\your-user-name\AppData\Local\Programs\Tibo (claude.exe in bin, Codex in codex), added to your own PATH~/.tibo (bin, codex), with a # >>> tibo-api >>> block added to .zshrc / .bashrc. You can choose another place on the first install, and it is remembered.env (address, key, default model) and modelPicker (the /model list) entries in ~/.claude/settings.json (on Windows, C:\Users\your-user-name\.claude\settings.json); other settings are untouched.claudeCode.environmentVariables and claudeCode.disableLoginPrompt to the user settings.json, keeping comments and other settings.# >>> tibo-api to ~/.codex/config.toml; your old model, model_provider and profile lines are commented out, not deleted. The CLI, extension and desktop app all read this one file. Its model_catalog_url makes Codex take its /model list from our site (per key; new models appear automatically), and api_key_model_discovery = true in the [features] section is what the model menu in the Codex desktop app and VS Code extension needs (added to your existing [features] section if there is one).tibo-claude (reads ~/.tibo/claude-settings.json) and tibo-codex, in the CLI tools' bin folder.tibo sits in the CLI tools' bin folder and fetches the latest setup script from our site each time; the defaults and lists you pick in the menu are saved in ~/.tibo/models.json.~/.tibo/backup/date-time/.R to restore your config; delete the CLI tools folder; search the Start menu for "Edit environment variables for your account" and remove the two Tibo entries from Pathremove the tibo-api block from .zshrc / .bashrc. Codex desktop app: uninstall ChatGPT in Settings → Appsdrag ChatGPT from Applications to the Trash.settings.json, not settings.json.txt.config.toml: model_provider = "tibo" must be at the top, before any [ ] section, and the key must be complete (starting with sk-).config.toml needs api_key_model_discovery = true in its [features] section (see the config above). Then fully quit and reopen the app / VS Code.modelPicker with "behavesAs": "claude-sonnet-5-5". Codex: [model_providers.tibo] needs the model_catalog_url line. Then quit and reopen the tool.CC Switch is a free desktop tool (Windows / macOS / Linux) that manages the API settings of Claude Code and Codex and switches between services in one click. You can import this site into CC Switch in one click; the provider card then shows your account balance, refreshed every 30 minutes.
.msi for Windows, .dmg for Mac), install it and open it once.claude-opus-5-5 for Claude, gpt-6.1-sol for Codex), then click "Open CC Switch". Allow the browser to open CC Switch.tibo-api.com for your balance with your key. Click "Import".Under "Add Provider", use https://tibo-api.com as the address for Claude and https://tibo-api.com/v1 for Codex, and your sk-... key.
Balance: click "Configure usage query" on the provider, choose the "General" template, set Base URL to https://tibo-api.com (without /v1) and API Key to the same key, turn the query on and save. The unit reads USD but the number is your balance in CNY (yuan); to show 元, change "USD" to "元" in the script.
If you set up with the one-click setup (the one-line command or tibo), you don't need CC Switch. Using both, switching providers in CC Switch replaces the Claude Code model list the setup wrote (/model may show a single model); run tibo and choose "change models" to restore it.
The figure is your account balance; if the key has a spending limit, it's what the key can still spend (whichever is lower).
Any software that supports "OpenAI-compatible" or Anthropic APIs works. Fill in these three things:
https://tibo-api.com/v1 (for OpenAI-style software)https://tibo-api.com (for Anthropic-style software and Claude Code, without /v1)
KeyThe sk-... copied from the API Keys page
Model nameGPT: gpt-6.1-sol, gpt-6-astra; Claude: claude-opus-5-5, claude-sonnet-5-5, etc.; others: gemini-3.8-flash, kimi-k3, glm-5.3, qwen3.8-max, deepseek-v4-pro-0813, MiniMax-M3, grok-4.7, etc. See Model Square for all models and prices.
Claude models work in Claude Code only: our upstreams accept Claude Code requests only, so calling Claude from chat apps or your own programs is refused. Use GPT or the other models in the apps below.
sk-... and the API address to https://tibo-api.com (the app adds /v1 itself; if it asks for the full address, use https://tibo-api.com/v1).gpt-6.1-sol) to the model list, switch them on, and pick the model on the chat page.from openai import OpenAI
client = OpenAI(base_url="https://tibo-api.com/v1", api_key="sk-your-key")
resp = client.chat.completions.create(
model="gpt-6.1-sol",
messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)Streaming, tool calls (function calling) and the Responses API (client.responses.create) are all supported and work the same as the official API.
sk-), or it was deleted, expired or disabled. Copy it again from the API Keys page./v1; Claude Code does not.max_tokens).You can see what every call cost under Usage Logs in the console. For anything else, join our QQ group 1126821195 or email support@tibo-api.com.