使用教程

快速开始

ClawBases 提供 OpenAI、Anthropic、Gemini 兼容的 API 接入。本页说明如何充值、创建 API 密钥，并在 Claude Code、Codex CLI 等客户端中完成配置。

打开控制台钱包充值 API 密钥

账户和 API 密钥

首次使用时，先准备余额，再按客户端用途创建 API 密钥。Claude、OpenAI、Gemini、图像等分组尽量分开使用，后续排查账单和权限会更清楚。

准备余额。打开钱包，选择固定金额或填写自定义金额，然后按页面提示完成支付。
打开 API 密钥页。从控制台左侧进入 API 密钥页面，确认已有密钥和分组。
创建密钥。填写名称，选择对应分组，设置过期时间和额度。完整密钥只在创建后显示一次。
配置客户端。把密钥写入环境变量或本地配置文件，不要提交到公开仓库，也不要出现在截图或日志里。

ClawBases API 密钥列表界面 — 列表中可以查看密钥名称、启用状态、分组和额度。

ClawBases 创建 API 密钥界面 — 创建密钥时先选分组，再按需要设置过期时间和额度。

Claude Code

Claude Code 使用 Anthropic 原生协议，Base URL 不需要 /v1。建议为 Claude 单独创建一个 Claude 分组的 API 密钥。

Anthropic Base URL：https://clawbases.com。建议通过 ANTHROPIC_AUTH_TOKEN 读取密钥，不要写进项目代码。

环境准备

Claude Code 需要 Node.js 18 或更高版本。已经安装过 Node.js 的用户，可以直接跳到安装 Claude Code。

平台	推荐方式	验证命令
macOS	`brew update && brew install node`，也可以从 `nodejs.org` 下载 LTS 安装包。	`node --version && npm --version`
Windows	从 `nodejs.org` 下载 LTS 版本，或用 `choco install nodejs` / `scoop install nodejs`。	`node --version; npm --version`
Linux	Ubuntu / Debian 推荐 NodeSource；也可以用系统包管理器安装 `nodejs` 和 `npm`。	`node --version && npm --version`

安装 Claude Code

macOS Native

brew install --cask claude-code

# 或使用官方脚本
curl -fsSL https://claude.ai/install.sh | bash

Windows Native

# PowerShell
irm https://claude.ai/install.ps1 | iex

# CMD
curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd

Linux Native

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

# Alpine Linux 如遇 ripgrep 依赖问题
apk add libgcc libstdc++ ripgrep
export USE_BUILTIN_RIPGREP=0

NPM 备用安装

npm install -g @anthropic-ai/claude-code
claude --version
claude doctor

连接 ClawBases

macOS / Linux 临时环境变量

export ANTHROPIC_BASE_URL="https://clawbases.com"
export ANTHROPIC_AUTH_TOKEN="sk-clawbases-..."

claude

macOS / Linux 永久写入

echo 'export ANTHROPIC_BASE_URL="https://clawbases.com"' >> ~/.zshrc
echo 'export ANTHROPIC_AUTH_TOKEN="sk-clawbases-..."' >> ~/.zshrc
source ~/.zshrc

Windows PowerShell

$env:ANTHROPIC_BASE_URL = "https://clawbases.com"
$env:ANTHROPIC_AUTH_TOKEN = "sk-clawbases-..."

[System.Environment]::SetEnvironmentVariable(
  "ANTHROPIC_BASE_URL",
  "https://clawbases.com",
  [System.EnvironmentVariableTarget]::User
)
[System.Environment]::SetEnvironmentVariable(
  "ANTHROPIC_AUTH_TOKEN",
  "sk-clawbases-...",
  [System.EnvironmentVariableTarget]::User
)

~/.claude/settings.json

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://clawbases.com",
    "ANTHROPIC_AUTH_TOKEN": "sk-clawbases-...",
    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1"
  }
}

VS Code 扩展

# 安装 Claude Code for VS Code 后创建：
# macOS / Linux: ~/.claude/config.json
# Windows: C:\Users\你的用户名\.claude\config.json

{
  "primaryApiKey": "any-value"
}

验证、启动和更新

echo $ANTHROPIC_BASE_URL
echo $ANTHROPIC_AUTH_TOKEN

cd /path/to/your/project
claude

npm install -g @anthropic-ai/claude-code

Codex CLI

Codex CLI 使用 OpenAI 兼容协议，Base URL 必须包含 /v1。建议使用 OpenAI 分组的 API 密钥，并通过环境变量读取。

OpenAI 兼容 Base URL：https://clawbases.com/v1。客户端里如果有 Base URL 字段，请填写这个完整地址。

环境准备

Codex CLI 需要 Node.js 和 npm。安装前先确认终端能识别 node 和 npm。

平台	推荐方式	验证命令
macOS	`brew update && brew install node`，或从 `nodejs.org` 下载 LTS 安装包。	`node --version && npm --version`
Windows	安装 Node.js LTS，或使用 `choco install nodejs`、`scoop install nodejs`。	`node --version; npm --version`
Linux	使用 NodeSource、系统包管理器，或团队统一的 Node 版本管理工具。	`node --version && npm --version`

安装 Codex CLI

npm i -g @openai/codex

# 国内网络如果安装较慢，可以临时指定镜像
npm i -g @openai/codex --registry=https://registry.npmmirror.com

验证安装

codex --version

cd /path/to/your/project
codex

配置 ClawBases

~/.codex/config.toml

model_provider = "rb"
model = "gpt-5.5"
model_reasoning_effort = "high"
disable_response_storage = true
sandbox_mode = "workspace-write"

[features]
plan_tool = true
apply_patch_freeform = true
view_image_tool = true
web_search_request = true
unified_exec = false
streamable_shell = false
rmcp_client = true

[model_providers.rb]
name = "rb"
base_url = "https://clawbases.com/v1"
wire_api = "responses"
env_key = "CLAWBASES_API_KEY"
requires_openai_auth = true

[sandbox_workspace_write]
network_access = true

macOS / Linux 环境变量

export CLAWBASES_API_KEY="sk-clawbases-..."

# 想永久生效就写入 shell 配置
echo 'export CLAWBASES_API_KEY="sk-clawbases-..."' >> ~/.zshrc
source ~/.zshrc

Windows PowerShell

[System.Environment]::SetEnvironmentVariable(
  "CLAWBASES_API_KEY",
  "sk-clawbases-...",
  [System.EnvironmentVariableTarget]::User
)

~/.codex/auth.json

{
  "OPENAI_API_KEY": "sk-clawbases-..."
}

VS Code 扩展

# 安装 Codex - OpenAI's coding agent 后，
# 确认本机已有 ~/.codex/config.toml、
# ~/.codex/auth.json 和 CLAWBASES_API_KEY。

codex

启动和更新

cd /path/to/your/project
codex

npm i -g @openai/codex

其他客户端

Cursor、Cline、Chatbox、Cherry Studio、Gemini CLI、Dify、N8N 等工具通常只需要 Base URL、模型名和 API 密钥。先确认客户端使用的是 Anthropic、OpenAI 兼容还是 Gemini 相关协议，再填对应端点。

用途	Base URL	密钥分组
Claude / Anthropic 原生协议	`https://clawbases.com`	Claude 或 Anthropic 相关分组
OpenAI 兼容协议	`https://clawbases.com/v1`	OpenAI / GPT / Codex 相关分组
Gemini 模型 / Gemini CLI	按客户端使用的 Gemini 或 OpenAI 兼容地址填写	`gemini-lite` 分组，当前公告倍率 0.8
图像生成	`https://clawbases.com`	`media` 分组；`gpt-image-2` 按实际用量计费，每张图片最低 0.08 元
图片与视频任务	`https://clawbases.com`	`media` 分组；同步视频调用 `/v1/video/generations`，异步任务调用 `/v1/videos`
Nana Banana 同步图片	`https://clawbases.com`	`media` 分组，调用 `/v1/images/generations`

图片与视频任务接口

Nana Banana、VEO、video-fast、video-pro 和 video-standard 模型使用 media 分组创建的 API Key。Nana Banana 的 *_sync 模型使用同步图片接口；nana-banana-2 和 nana-banana-pro 使用异步图片任务；视频模型默认建议使用同步视频入口，按需也可以使用异步任务入口。

同步图片：POST https://clawbases.com/v1/images/generations。同步视频：POST https://clawbases.com/v1/video/generations。异步任务：POST https://clawbases.com/v1/videos；查询任务：GET https://clawbases.com/v1/videos/{task_id}。异步调用按 4 倍扣费；表格中的价格为模型广场基础价，最终扣费以控制台账单为准。

模型	用途	基础价格	关键参数
`nana-banana-2`	图片生成与编辑	¥0.05 / 请求	`duration=4`；`images` 最多 9 张；`resolution` 支持 1K 或 2K
`nana-banana-pro`	图片生成与编辑	¥0.06 / 请求	`duration=4`；`images` 最多 9 张；`resolution` 支持 1K 或 2K
`nana-banana-2_sync`	同步图片生成与编辑	¥0.20 / 请求	`/v1/images/generations`；支持 `images` 或 `image` 参考图
`nana-banana-pro_sync`	同步图片生成与编辑	¥0.24 / 请求	`/v1/images/generations`；支持 `images` 或 `image` 参考图
`veo-3-1`	8 秒视频生成	¥0.16 / 请求	`duration=8`；支持文生视频、首尾帧、参考图
`veo-omni-flash`	10 秒多参考图视频	¥0.16 / 请求	`duration=10`；`Ingredients_images` 最多 6 张
`veo-omni-flash-video-edit`	参考视频编辑	¥0.16 / 请求	`video_url` 必填；可带 `Ingredients_images`；`duration=10`
`video-fast-480p`	480p Fast 视频	¥0.70 / 请求	`duration`：4-15 秒；支持图片、视频、音频参考素材
`video-fast-720p`	720p Fast 视频	¥0.80 / 请求	`duration`：4-15 秒；支持图片、视频、音频参考素材
`video-pro-480p`	480p Pro 视频	¥0.72 / 请求	`duration`：4-15 秒；支持图片、视频、音频参考素材
`video-pro-720p`	720p Pro 视频	¥1.10 / 请求	`duration`：4-15 秒；支持图片、视频、音频参考素材
`video-pro-1080p`	1080p Pro 视频	¥2.50 / 请求	`duration`：4-15 秒；支持图片、视频、音频参考素材
`video-standard-720p`	720p Standard 视频	¥0.72 / 请求	`duration=15`；支持图片、视频、音频参考素材

同步 Nana Banana 图片

同步生成图片

curl -sS --max-time 600 -X POST "https://clawbases.com/v1/images/generations" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nana-banana-2_sync",
    "prompt": "生成一张干净的产品海报，白色背景，高级商业摄影质感",
    "size": "1024x1024",
    "n": 1,
    "images": [
      "https://your-cdn.com/reference.jpg"
    ]
  }'

同步返回格式

{
  "created": 1714200180,
  "data": [
    {
      "url": "https://.../generated/image.jpg"
    }
  ]
}

同步视频调用

同步视频生成

curl -sS --max-time 600 -X POST "https://clawbases.com/v1/video/generations" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "video-fast-720p",
    "prompt": "一只玻璃纸鹤在桌面旋转，柔和侧光，微距镜头",
    "duration": 4,
    "aspect_ratio": "1:1"
  }'

同步视频返回

{
  "id": "task_xxx",
  "task_id": "task_xxx",
  "object": "video",
  "model": "video-fast-720p",
  "status": "in_progress"
}

同步视频入口返回任务对象后，继续用查询接口获取最终 video_url。异步任务入口适合需要后台排队、批量提交或明确接受 4 倍扣费的场景。

提交异步任务

Nana Banana 图片任务

curl -sS -X POST "https://clawbases.com/v1/videos" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nana-banana-2",
    "prompt": "生成一张干净的产品海报，白色背景，高级商业摄影质感",
    "duration": 4,
    "resolution": "2k",
    "images": [
      "https://your-cdn.com/reference.jpg"
    ]
  }'

VEO 3.1 首尾帧视频

curl -sS -X POST "https://clawbases.com/v1/videos" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "veo-3-1",
    "prompt": "让主体从第一帧自然运动到最后一帧，电影感光线，镜头平稳",
    "duration": 8,
    "aspect_ratio": "16:9",
    "images": [
      "https://your-cdn.com/first-frame.jpg",
      "https://your-cdn.com/last-frame.jpg"
    ]
  }'

video-pro 多素材参考

curl -sS -X POST "https://clawbases.com/v1/videos" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "video-pro-1080p",
    "prompt": "高质量产品广告片，镜头运动顺滑，商业质感",
    "duration": 6,
    "aspect_ratio": "16:9",
    "image_url": "https://your-cdn.com/main-reference.jpg",
    "extra_images": ["https://your-cdn.com/ref-1.jpg"],
    "extra_videos": ["https://your-cdn.com/ref-video.mp4"],
    "extra_audios": ["https://your-cdn.com/ref-audio.mp3"]
  }'

视频编辑

curl -sS -X POST "https://clawbases.com/v1/videos" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "veo-omni-flash-video-edit",
    "prompt": "让参考视频更明亮、更电影感，保留主体动作",
    "duration": 10,
    "aspect_ratio": "16:9",
    "video_url": "https://your-cdn.com/reference-video.mp4",
    "Ingredients_images": [
      "https://your-cdn.com/product.jpg"
    ]
  }'

查询任务和读取结果

轮询任务状态

curl -sS "https://clawbases.com/v1/videos/TASK_ID" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY"

常见响应字段

{
  "id": "task_385412",
  "task_id": "task_385412",
  "object": "video",
  "model": "video-pro-1080p",
  "status": "completed",
  "task_status": "completed",
  "progress": 100,
  "image_url": null,
  "video_url": "https://.../generated/video.mp4",
  "url": "https://.../generated/video.mp4",
  "result_urls": ["https://.../generated/video.mp4"]
}

任务没有立刻返回图片或视频

这是正常的。/v1/videos 是异步任务接口，提交成功后先返回 task_id，需要用查询接口轮询。状态为 queued、processing 或 in_progress 时继续等待；状态为 completed 时读取 video_url、image_url、url 或 result_urls。

同步和异步怎么选

gpt-image-2、nana-banana-2_sync 和 nana-banana-pro_sync 使用 /v1/images/generations 或 /v1/images/edits。视频模型同步调用使用 /v1/video/generations。异步 Nana Banana、VEO、video-fast、video-pro 和 video-standard 任务使用 /v1/videos，异步调用按 4 倍扣费。

提示模型或分组不可用

请确认 API Key 属于 media 分组，并确认请求地址是对应接口：同步图片使用 /v1/images/generations，同步视频使用 /v1/video/generations，异步任务使用 /v1/videos。不要把这些模型发到 /v1/chat/completions 或 /v1/responses。

gpt-image-2 图片接口

图片生成和编辑使用专用 Base URL https://clawbases.com。请使用 media 分组创建的 API Key；普通文本分组 Key 即使能鉴权，也可能没有图片渠道。当前图像接口接入 Azure 图像线路。gpt-image-2 按实际用量计费，并按每张图片最低 0.08 元结算；如实际用量高于最低价，则按实际用量扣费。最终扣费以控制台账单为准。

生成图片用 JSON 请求 /v1/images/generations；编辑图片用 multipart/form-data 请求 /v1/images/edits。两个接口的模型名都是 gpt-image-2。

生成图片

curl -sS --max-time 600 -X POST "https://clawbases.com/v1/images/generations" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "一只橘猫坐在赛博朋克城市的霓虹灯下，电影感，细节丰富",
    "size": "1024x1024",
    "quality": "high",
    "output_format": "png",
    "n": 1
  }' \
  | jq -r '.data[0].b64_json // empty' | base64 --decode > output.png

编辑图片

curl -sS --max-time 600 -X POST "https://clawbases.com/v1/images/edits" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -F "model=gpt-image-2" \
  -F "image[]=@/path/to/source.png" \
  -F "prompt=把图片里的背景改成东京夜景，保留人物长相、发型、表情和姿势" \
  -F "size=1024x1024" \
  -F "quality=high" \
  -F "output_format=png" \
  -F "n=1" \
  | jq -r '.data[0].b64_json // empty' | base64 --decode > output.png

多张参考图

curl -sS --max-time 600 -X POST "https://clawbases.com/v1/images/edits" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -F "model=gpt-image-2" \
  -F "image[]=@/path/to/1.png" \
  -F "image[]=@/path/to/2.png" \
  -F "prompt=参考两张图生成同一人物的新场景，保持身份一致" \
  -F "size=1024x1024" \
  -F "quality=high" \
  -F "output_format=png" \
  -F "n=1"

返回处理

# 成功响应主要字段：
# data[0].b64_json        base64 图片内容（gpt-image-2 默认返回）
# data[0].revised_prompt  模型实际使用的提示词
# usage                   token 用量统计

jq -r '.data[0].b64_json // empty' response.json | base64 --decode > output.png

返回没有 `data[0].b64_json`

gpt-image-2 成功时默认在 data[0].b64_json 返回 base64 图片，不返回 url。如果返回的是 error 对象，按其中的 message 排查；如果是 502、server_error 等临时错误，稍后重试即可。

提示模型或分组不可用

通常是 Key 不属于 media 分组，或请求发送到了通用文本端点。请重新创建 media 分组 Key，并确认 URL 是 https://clawbases.com。

连接中断 / 超时（curl 56、schannel closed abruptly）

图片接口是同步返回，quality=high 的请求通常要等 1–3 分钟，更大尺寸更久。等待期间如果客户端或中间代理超时，就会出现 curl: (56) schannel: server closed abruptly 这类连接中断。请加大超时并自动重试，例如 --max-time 600 --retry 3 --retry-all-errors；首次联调建议先用 1024x1024 跑通，再按需提高尺寸。

Windows 下如何运行

PowerShell 里的 curl 是 Invoke-WebRequest 的别名，请改用 curl.exe，并把上面的命令写成单行（去掉行尾的 \）。Windows 自带没有 jq/base64，可先用 -o response.json 保存返回，再用 PowerShell 解码：powershell -Command "$j=Get-Content response.json | ConvertFrom-Json; [IO.File]::WriteAllBytes('out.png', [Convert]::FromBase64String($j.data[0].b64_json))"。

常用命令

Claude Code 和 Codex CLI 的斜杠命令会随版本变化，下面这些最常用，适合日常排查、清理上下文和控制成本。

命令	用途	建议场景
`/help`	查看当前客户端支持的命令和快捷键。	刚安装或升级后先看一遍。
`/clear`	清空当前会话上下文。	换任务、上下文混乱、模型开始沿用旧信息时使用。
`/compact`	压缩当前上下文，保留关键进度。	长任务跑久了，但还想延续当前工作状态。
`/cost`	查看当前会话的用量或费用信息。	排查余额消耗、评估一次任务成本。
`/model`	查看或切换模型。	需要在速度、成本和推理能力之间切换时使用。

如果某个命令不存在，以当前客户端内的 /help 为准。不同版本的 Claude Code 和 Codex CLI 命令名称可能略有差异。

故障排查

安装失败或下载很慢

先确认网络能访问对应安装源，Windows 终端建议用管理员权限重试。npm 安装慢时可以临时加 --registry=https://registry.npmmirror.com。

命令未找到

确认 Node.js、npm 和全局安装目录已加入 PATH。macOS / Linux 可检查 npm config get prefix，Windows 需要重新打开 PowerShell 或 VS Code。

密钥无效或 401

确认密钥复制完整、没有多余空格，并且未被禁用或过期。完整密钥只在创建后显示一次，丢失后需要重新创建。

端点不匹配

Claude Code 使用 https://clawbases.com；OpenAI 兼容客户端使用 https://clawbases.com/v1。多数接入问题都出在这里。

环境变量没有生效

临时环境变量只对当前终端有效。写入 ~/.zshrc、~/.bashrc 或 Windows 用户环境变量后，需要重新打开终端和编辑器。

无法连接 API

先确认 Base URL 填写正确，再检查代理、防火墙和 DNS。macOS / Linux 可用 curl -I https://clawbases.com，Windows 可用 Test-NetConnection clawbases.com -Port 443。

余额或分组不对

确认账户余额充足，密钥分组和模型类型一致。图像模型需要使用 media 分组和图像端点。

图像接口不通

图像生成不要使用通用 OpenAI 兼容端点，请使用 https://clawbases.com，并确认密钥属于 media 分组。

需要接入协助

把客户端名称、Base URL、模型名和报错信息发给我们即可。请不要发送完整 API 密钥。

Telegram @clawbases support@clawbases.com

扫码添加官方客服

Usage guide

Getting started

ClawBases provides OpenAI-, Anthropic-, and Gemini-compatible API access. This page covers adding balance, creating API keys, and configuring clients such as Claude Code and Codex CLI.

Open Console Add Funds API Keys

Account and API keys

Start with a usable balance, then create API keys for each protocol or model group. Keeping Claude, OpenAI, Gemini, and image keys separate makes billing and permission checks easier.

Add balance. Open Wallet, choose a preset amount or enter a custom amount, then complete payment.
Open API Keys. Go to the API Keys page from the console sidebar and review existing groups.
Create a key. Enter a name, choose a group, and set quota or expiration. The full key is shown once.
Use it in clients. Put the key in environment variables or config files, never in public repos, screenshots, or logs.

ClawBases Add Funds screen — Select an amount and payment method in Wallet.

ClawBases API Keys list — The API Keys list shows name, status, group, and quota.

ClawBases Create API Key drawer — Choose the correct group when creating a key, then set expiration and quota as needed.

Claude Code

Claude Code uses the native Anthropic protocol, so the endpoint does not include /v1. Create a Claude group key for this setup.

Anthropic endpoint: https://clawbases.com. Store the key in ANTHROPIC_AUTH_TOKEN instead of project code.

Prerequisites

Claude Code needs Node.js 18 or newer. If Node.js is already installed, you can skip straight to installing Claude Code.

Platform	Recommended setup	Verify
macOS	Use `brew update && brew install node`, or install the LTS package from `nodejs.org`.	`node --version && npm --version`
Windows	Install Node.js LTS from `nodejs.org`, or use `choco install nodejs` / `scoop install nodejs`.	`node --version; npm --version`
Linux	Use NodeSource on Ubuntu / Debian, or install `nodejs` and `npm` from your package manager.	`node --version && npm --version`

Install Claude Code

macOS Native

brew install --cask claude-code

# Or use the official install script
curl -fsSL https://claude.ai/install.sh | bash

Windows Native

# PowerShell
irm https://claude.ai/install.ps1 | iex

# CMD
curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd

Linux Native

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

# Alpine Linux if ripgrep dependencies fail
apk add libgcc libstdc++ ripgrep
export USE_BUILTIN_RIPGREP=0

NPM fallback

npm install -g @anthropic-ai/claude-code
claude --version
claude doctor

Connect ClawBases

macOS / Linux temporary env

export ANTHROPIC_BASE_URL="https://clawbases.com"
export ANTHROPIC_AUTH_TOKEN="sk-clawbases-..."

claude

macOS / Linux persistent env

echo 'export ANTHROPIC_BASE_URL="https://clawbases.com"' >> ~/.zshrc
echo 'export ANTHROPIC_AUTH_TOKEN="sk-clawbases-..."' >> ~/.zshrc
source ~/.zshrc

Windows PowerShell

$env:ANTHROPIC_BASE_URL = "https://clawbases.com"
$env:ANTHROPIC_AUTH_TOKEN = "sk-clawbases-..."

[System.Environment]::SetEnvironmentVariable(
  "ANTHROPIC_BASE_URL",
  "https://clawbases.com",
  [System.EnvironmentVariableTarget]::User
)
[System.Environment]::SetEnvironmentVariable(
  "ANTHROPIC_AUTH_TOKEN",
  "sk-clawbases-...",
  [System.EnvironmentVariableTarget]::User
)

~/.claude/settings.json

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://clawbases.com",
    "ANTHROPIC_AUTH_TOKEN": "sk-clawbases-...",
    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1"
  }
}

VS Code extension

# After installing Claude Code for VS Code, create:
# macOS / Linux: ~/.claude/config.json
# Windows: C:\Users\your-name\.claude\config.json

{
  "primaryApiKey": "any-value"
}

Verify, start, update

echo $ANTHROPIC_BASE_URL
echo $ANTHROPIC_AUTH_TOKEN

cd /path/to/your/project
claude

npm install -g @anthropic-ai/claude-code

Codex CLI

Codex CLI uses an OpenAI compatible protocol, so the endpoint must include /v1. Create an OpenAI group key and load it from an environment variable.

OpenAI compatible endpoint: https://clawbases.com/v1. If your client has a Base URL field, include the full /v1 path.

Prerequisites

Codex CLI needs Node.js and npm. Before installing, make sure your terminal can run node and npm.

Platform	Recommended setup	Verify
macOS	Use `brew update && brew install node`, or install the LTS package from `nodejs.org`.	`node --version && npm --version`
Windows	Install Node.js LTS, or use `choco install nodejs` / `scoop install nodejs`.	`node --version; npm --version`
Linux	Use NodeSource, your package manager, or your team's preferred Node version manager.	`node --version && npm --version`

Install Codex CLI

npm i -g @openai/codex

# If install is slow, use a temporary mirror
npm i -g @openai/codex --registry=https://registry.npmmirror.com

Verify installation

codex --version

cd /path/to/your/project
codex

Configure ClawBases

~/.codex/config.toml

model_provider = "rb"
model = "gpt-5.5"
model_reasoning_effort = "high"
disable_response_storage = true
sandbox_mode = "workspace-write"

[features]
plan_tool = true
apply_patch_freeform = true
view_image_tool = true
web_search_request = true
unified_exec = false
streamable_shell = false
rmcp_client = true

[model_providers.rb]
name = "rb"
base_url = "https://clawbases.com/v1"
wire_api = "responses"
env_key = "CLAWBASES_API_KEY"
requires_openai_auth = true

[sandbox_workspace_write]
network_access = true

macOS / Linux env

export CLAWBASES_API_KEY="sk-clawbases-..."

# Persist it for future terminals
echo 'export CLAWBASES_API_KEY="sk-clawbases-..."' >> ~/.zshrc
source ~/.zshrc

Windows PowerShell

[System.Environment]::SetEnvironmentVariable(
  "CLAWBASES_API_KEY",
  "sk-clawbases-...",
  [System.EnvironmentVariableTarget]::User
)

~/.codex/auth.json

{
  "OPENAI_API_KEY": "sk-clawbases-..."
}

VS Code extension

# After installing Codex - OpenAI's coding agent,
# confirm ~/.codex/config.toml, ~/.codex/auth.json,
# and CLAWBASES_API_KEY are configured on this machine.

codex

Start and update

cd /path/to/your/project
codex

npm i -g @openai/codex

Other clients

Cursor, Cline, Chatbox, Cherry Studio, Gemini CLI, Dify, N8N, and similar tools usually need only a Base URL, model name, and API key. The endpoint changes by protocol.

Use case	Base URL	Key group
Claude / native Anthropic	`https://clawbases.com`	Claude or Anthropic related group
OpenAI compatible	`https://clawbases.com/v1`	OpenAI / GPT / Codex related group
Gemini models / Gemini CLI	Use the Gemini or OpenAI compatible Base URL required by your client	`gemini-lite` group, announced at 0.8 multiplier
Image generation	`https://clawbases.com`	`media` group; `gpt-image-2` is billed by actual usage with a CNY 0.08 minimum per image
Image and video tasks	`https://clawbases.com`	`media` group; use `/v1/video/generations` for sync video and `/v1/videos` for async tasks
Nana Banana synchronous images	`https://clawbases.com`	`media` group, call `/v1/images/generations`

Image and video task API

Nana Banana, VEO, video-fast, video-pro, and video-standard models use API keys from the media group. Nana Banana *_sync models use the synchronous image endpoint; nana-banana-2 and nana-banana-pro use async image tasks. Video models should use the sync video endpoint by default, and can use the async task endpoint when needed.

Synchronous image call: POST https://clawbases.com/v1/images/generations. Synchronous video call: POST https://clawbases.com/v1/video/generations. Async task call: POST https://clawbases.com/v1/videos; query with GET https://clawbases.com/v1/videos/{task_id}. Async calls cost x4; table prices are the model catalog base prices, and the console bill is authoritative.

Model	Use case	Base price	Key parameters
`nana-banana-2`	Image generation / editing	¥0.05 / request	`duration=4`; up to 9 `images`; `resolution` supports 1K or 2K
`nana-banana-pro`	Image generation / editing	¥0.06 / request	`duration=4`; up to 9 `images`; `resolution` supports 1K or 2K
`nana-banana-2_sync`	Synchronous image generation and editing	¥0.20 / request	`/v1/images/generations`; supports `images` or `image` references
`nana-banana-pro_sync`	Synchronous image generation and editing	¥0.24 / request	`/v1/images/generations`; supports `images` or `image` references
`veo-3-1`	8-second video generation	¥0.16 / request	`duration=8`; supports text-to-video, first/last frames, and reference images
`veo-omni-flash`	10-second multi-reference video	¥0.16 / request	`duration=10`; `Ingredients_images` supports up to 6 images
`veo-omni-flash-video-edit`	Reference video editing	¥0.16 / request	`video_url` is required; optional `Ingredients_images`; `duration=10`
`video-fast-480p`	480p Fast video	¥0.70 / request	`duration`: 4-15 seconds; supports image, video, and audio references
`video-fast-720p`	720p Fast video	¥0.80 / request	`duration`: 4-15 seconds; supports image, video, and audio references
`video-pro-480p`	480p Pro video	¥0.72 / request	`duration`: 4-15 seconds; supports image, video, and audio references
`video-pro-720p`	720p Pro video	¥1.10 / request	`duration`: 4-15 seconds; supports image, video, and audio references
`video-pro-1080p`	1080p Pro video	¥2.50 / request	`duration`: 4-15 seconds; supports image, video, and audio references
`video-standard-720p`	720p Standard video	¥0.72 / request	`duration=15`; supports image, video, and audio references

Synchronous Nana Banana images

Generate synchronously

curl -sS --max-time 600 -X POST "https://clawbases.com/v1/images/generations" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nana-banana-2_sync",
    "prompt": "Create a clean product poster on a white background with premium commercial photography lighting",
    "size": "1024x1024",
    "n": 1,
    "images": [
      "https://your-cdn.com/reference.jpg"
    ]
  }'

Synchronous response shape

{
  "created": 1714200180,
  "data": [
    {
      "url": "https://.../generated/image.jpg"
    }
  ]
}

Synchronous video call

Generate video synchronously

curl -sS --max-time 600 -X POST "https://clawbases.com/v1/video/generations" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "video-fast-720p",
    "prompt": "A glass paper crane rotating on a tabletop, soft side light, macro shot",
    "duration": 4,
    "aspect_ratio": "1:1"
  }'

Synchronous video response

{
  "id": "task_xxx",
  "task_id": "task_xxx",
  "object": "video",
  "model": "video-fast-720p",
  "status": "in_progress"
}

After the sync video endpoint returns a task object, keep using the query endpoint to obtain the final video_url. The async task endpoint is useful when you need background queueing, batch submission, or explicitly accept cost x4.

Create an async task

Nana Banana image task

curl -sS -X POST "https://clawbases.com/v1/videos" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nana-banana-2",
    "prompt": "Create a clean product poster on a white background with premium commercial photography lighting",
    "duration": 4,
    "resolution": "2k",
    "images": [
      "https://your-cdn.com/reference.jpg"
    ]
  }'

VEO 3.1 first/last frame video

curl -sS -X POST "https://clawbases.com/v1/videos" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "veo-3-1",
    "prompt": "Animate naturally from the first frame to the last frame with cinematic lighting and stable camera motion",
    "duration": 8,
    "aspect_ratio": "16:9",
    "images": [
      "https://your-cdn.com/first-frame.jpg",
      "https://your-cdn.com/last-frame.jpg"
    ]
  }'

video-pro with references

curl -sS -X POST "https://clawbases.com/v1/videos" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "video-pro-1080p",
    "prompt": "High quality product commercial, smooth camera movement, premium visual style",
    "duration": 6,
    "aspect_ratio": "16:9",
    "image_url": "https://your-cdn.com/main-reference.jpg",
    "extra_images": ["https://your-cdn.com/ref-1.jpg"],
    "extra_videos": ["https://your-cdn.com/ref-video.mp4"],
    "extra_audios": ["https://your-cdn.com/ref-audio.mp3"]
  }'

Video edit

curl -sS -X POST "https://clawbases.com/v1/videos" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "veo-omni-flash-video-edit",
    "prompt": "Make the reference video brighter and more cinematic while preserving the subject motion",
    "duration": 10,
    "aspect_ratio": "16:9",
    "video_url": "https://your-cdn.com/reference-video.mp4",
    "Ingredients_images": [
      "https://your-cdn.com/product.jpg"
    ]
  }'

Poll and read the result

Poll task status

curl -sS "https://clawbases.com/v1/videos/TASK_ID" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY"

Common response fields

{
  "id": "task_385412",
  "task_id": "task_385412",
  "object": "video",
  "model": "video-pro-1080p",
  "status": "completed",
  "task_status": "completed",
  "progress": 100,
  "image_url": null,
  "video_url": "https://.../generated/video.mp4",
  "url": "https://.../generated/video.mp4",
  "result_urls": ["https://.../generated/video.mp4"]
}

The task does not immediately return media

That is expected. /v1/videos is an async task API. A successful create request returns a task_id; keep polling while the status is queued, processing, or in_progress. When it is completed, read video_url, image_url, url, or result_urls.

When should I use sync or async?

gpt-image-2, nana-banana-2_sync, and nana-banana-pro_sync use /v1/images/generations or /v1/images/edits. Video models use /v1/video/generations for sync calls. Async Nana Banana, VEO, video-fast, video-pro, and video-standard tasks use /v1/videos, and async calls cost x4.

Model or group not available

Confirm that the API key belongs to the media group and that the request URL matches the interface: sync images use /v1/images/generations, sync videos use /v1/video/generations, and async tasks use /v1/videos. Do not send these models to /v1/chat/completions or /v1/responses.

gpt-image-2 Image API

Image generation and editing use the dedicated Base URL https://clawbases.com. Use an API key created in the media group; a text-model key may authenticate but still have no image route available. The image route uses Azure image endpoints. gpt-image-2 is billed by actual usage with a minimum charge of CNY 0.08 per generated or edited image. If actual usage exceeds the minimum, actual usage billing applies. The console bill is authoritative.

Use JSON with /v1/images/generations for text-to-image generation. Use multipart/form-data with /v1/images/edits for image editing. Both endpoints use the model name gpt-image-2.

Generate an image

curl -sS --max-time 600 -X POST "https://clawbases.com/v1/images/generations" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "An orange cat under neon lights in a cyberpunk city, cinematic, richly detailed",
    "size": "1024x1024",
    "quality": "high",
    "output_format": "png",
    "n": 1
  }' \
  | jq -r '.data[0].b64_json // empty' | base64 --decode > output.png

Edit an image

curl -sS --max-time 600 -X POST "https://clawbases.com/v1/images/edits" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -F "model=gpt-image-2" \
  -F "image[]=@/path/to/source.png" \
  -F "prompt=Change the background to Tokyo at night while preserving the person's face, hair, expression, and pose" \
  -F "size=1024x1024" \
  -F "quality=high" \
  -F "output_format=png" \
  -F "n=1" \
  | jq -r '.data[0].b64_json // empty' | base64 --decode > output.png

Multiple reference images

curl -sS --max-time 600 -X POST "https://clawbases.com/v1/images/edits" \
  -H "Authorization: Bearer $CLAWBASES_API_KEY" \
  -F "model=gpt-image-2" \
  -F "image[]=@/path/to/1.png" \
  -F "image[]=@/path/to/2.png" \
  -F "prompt=Use both references to create a new scene with the same person" \
  -F "size=1024x1024" \
  -F "quality=high" \
  -F "output_format=png" \
  -F "n=1"

Read the response

# Main fields in a successful response:
# data[0].b64_json        base64 image bytes (gpt-image-2 default)
# data[0].revised_prompt  the prompt the model actually used
# usage                   token usage stats

jq -r '.data[0].b64_json // empty' response.json | base64 --decode > output.png

No `data[0].b64_json` in the response

On success, gpt-image-2 returns the image as base64 in data[0].b64_json — it does not return a url. If you get an error object instead, follow its message; for transient 502 / server_error responses, simply retry.

Model or group not available

This usually means the key is not in the media group, or the request was sent to a general text endpoint. Create a media-group key and confirm the URL is https://clawbases.com.

Connection dropped / timeout (curl 56, schannel closed abruptly)

The image endpoints respond synchronously; a quality=high request typically takes 1–3 minutes, and larger sizes take longer. If the client or an intermediate proxy times out during the wait, you see errors like curl: (56) schannel: server closed abruptly. Raise the timeout and auto-retry, e.g. --max-time 600 --retry 3 --retry-all-errors, and start with 1024x1024 before scaling up.

Running on Windows

In PowerShell, curl is an alias for Invoke-WebRequest — use curl.exe and put the command on a single line (drop the trailing \). Windows has no built-in jq/base64, so save the response with -o response.json and decode it in PowerShell: powershell -Command "$j=Get-Content response.json | ConvertFrom-Json; [IO.File]::WriteAllBytes('out.png', [Convert]::FromBase64String($j.data[0].b64_json))".

Commands

Slash commands vary by Claude Code and Codex CLI version, but these are the ones most useful for day-to-day setup, context cleanup, and cost checks.

Command	What it does	When to use it
`/help`	Shows supported commands and shortcuts for the current client.	Run it after a fresh install or upgrade.
`/clear`	Clears the current conversation context.	Use it when switching tasks or when old context starts leaking into the answer.
`/compact`	Compresses the current context while keeping the important progress.	Useful for long coding sessions you want to continue.
`/cost`	Shows session usage or cost details when the client supports it.	Use it to estimate spend or investigate balance usage.
`/model`	Shows or changes the active model.	Use it when switching between speed, cost, and reasoning quality.

If a command is missing, trust the current client's /help output. Command names can change slightly between releases.

Troubleshooting

Install fails or downloads slowly

Check that your network can reach the install source. On Windows, retry from an administrator terminal. For slow npm installs, add --registry=https://registry.npmmirror.com temporarily.

Command not found

Make sure Node.js, npm, and your global npm bin directory are on PATH. On macOS / Linux, check npm config get prefix. On Windows, reopen PowerShell or VS Code after changing environment variables.

Invalid key or 401

Check that the key was copied completely, has no extra spaces, and is not disabled or expired. The full key is only shown once after creation.

Wrong endpoint

Claude Code uses https://clawbases.com. OpenAI compatible clients use https://clawbases.com/v1. This is the most common setup difference.

Environment variables did not apply

Temporary environment variables only apply to the current terminal. After editing ~/.zshrc, ~/.bashrc, or Windows user environment variables, reopen your terminal and editor.

Cannot connect to the API

Check the Base URL first, then review proxy, firewall, and DNS settings. On macOS / Linux, try curl -I https://clawbases.com. On Windows, try Test-NetConnection clawbases.com -Port 443.

Balance or group mismatch

Confirm that your account has balance and that the key group matches the model. Image models need the image group and image endpoint.

Image API does not work

Image generation should use https://clawbases.com, not the general OpenAI compatible endpoint. Also confirm that the key belongs to the media group.

Need setup help

Send the client name, Base URL, model name, and error message to support. Please do not send the full API key.

Telegram @clawbases support@clawbases.com

ClawBases Enterprise WeChat contact QR code

Scan with Enterprise WeChat for official support

账户和 API 密钥

Claude Code

环境准备

安装 Claude Code

macOS Native

Windows Native

Linux Native

NPM 备用安装

连接 ClawBases

macOS / Linux 临时环境变量

macOS / Linux 永久写入

Windows PowerShell

~/.claude/settings.json

VS Code 扩展

验证、启动和更新

Codex CLI

环境准备

安装 Codex CLI

安装 Codex CLI

验证安装

配置 ClawBases

~/.codex/config.toml

macOS / Linux 环境变量

Windows PowerShell

~/.codex/auth.json

VS Code 扩展

启动和更新

其他客户端

图片与视频任务接口

同步 Nana Banana 图片

同步生成图片

同步返回格式

同步视频调用

同步视频生成

同步视频返回

提交异步任务

Nana Banana 图片任务

VEO 3.1 首尾帧视频

video-pro 多素材参考

视频编辑

查询任务和读取结果

轮询任务状态

常见响应字段

任务没有立刻返回图片或视频

同步和异步怎么选

提示模型或分组不可用

gpt-image-2 图片接口

生成图片

编辑图片

多张参考图

返回处理

返回没有 data[0].b64_json

提示模型或分组不可用

连接中断 / 超时（curl 56、schannel closed abruptly）

Windows 下如何运行

常用命令

故障排查

安装失败或下载很慢

命令未找到

密钥无效或 401

端点不匹配

环境变量没有生效

无法连接 API

余额或分组不对

图像接口不通

需要接入协助

Account and API keys

Claude Code

Prerequisites

Install Claude Code

macOS Native

Windows Native

Linux Native

NPM fallback

Connect ClawBases

macOS / Linux temporary env

macOS / Linux persistent env

Windows PowerShell

~/.claude/settings.json

VS Code extension

返回没有 `data[0].b64_json`

No `data[0].b64_json` in the response