openclaw-webmodel

Free web-based AI models for OpenClaw — ChatGPT, Grok, and more, without API keys.

Use ChatGPT and Grok directly through their web interfaces. No API keys needed, no per-token billing.

English | 中文

English

What is this?

openclaw-webmodel is an OpenClaw lifecycle plugin that lets your agent use web-based AI models (ChatGPT, Grok) via browser automation. Instead of paying for API access, it controls your logged-in browser sessions through Playwright CDP.

Features

• ChatGPT integration — /chatgpt <message> for one-shot or /chatgpt to lock mode
• Grok integration — /grok <message> for chat, auto-detects image generation prompts
• Lock mode — After /chatgpt or /grok, all subsequent messages go through that model automatically
• Auto session reset — Switching models automatically creates a clean session
• Dual-mode routing — Discord and TUI/HTTP paths are fully isolated, no interference
• Multi-instance support — Each gateway gets its own relay port (gateway port + 1000), no conflicts
• Relay provider — Built-in micro HTTP server acts as an OpenAI-compatible proxy for TUI mode
• Image generation — Grok Imagine and ChatGPT DALL-E, with auto-detection and local image saving
• Upgrade CLI — openclaw-webmodel upgrade to detect, remove old, and install new version

How it works

TUI/HTTP mode:

User: /grok hello
  → Plugin calls Grok via browser automation
  → Returns response directly
  → Locks Grok mode + switches default model to relay + resets session

User: (next message)
  → before_model_resolve hook → switches to relay provider
  → before_prompt_build hook → forwards to Grok via browser
  → Relay HTTP server serves the response back to gateway

Discord mode:

User: /grok hello
  → Plugin calls Grok via browser automation
  → Returns response directly + locks Grok mode

User: (next message)
  → before_dispatch hook → detects Discord (event.accountId present)
  → Calls Grok via browser → sends response via CLI to Discord channel
  → Returns { handled: true } → skips entire agent pipeline

Commands

| Command | Description | |---------|-------------| | /chatgpt <msg> | Send message to ChatGPT, auto-lock mode | | /chatgpt | Toggle ChatGPT lock mode | | /chatgpt off | Unlock, return to default model | | /grok <msg> | Send message to Grok, auto-lock mode | | /grok | Toggle Grok lock mode | | /grok off | Unlock, return to default model | | /chatgpt-auth | Save ChatGPT cookies from Chrome | | /grok-auth | Save Grok cookies from Chrome |

Setup

1. Install the plugin:

openclaw plugins install openclaw-webmodel
openclaw gateway restart

That's it. The plugin auto-configures everything on first startup:

• Adds webmodel-relay provider to models.providers (port = gateway port + 1000)
• Adds auth profile webmodel-relay:default
• Adds itself to plugins.allow and plugins.entries
• If openclaw-superpowers is installed, adds deferTo cooperation

2. Authenticate with web models:

# Start Chrome with debugging, log into chatgpt.com/grok.com, then:
/chatgpt-auth
/grok-auth

Upgrade

# Via CLI (recommended)
openclaw-webmodel upgrade

# Or manually
openclaw plugins install openclaw-webmodel

The upgrade command automatically detects old installations, removes them, and installs the latest version.

Architecture

• Browser automation — Playwright CDP connects to your logged-in Chrome session
• Relay HTTP server — Runs on gateway_port + 1000, serves OpenAI-compatible SSE streaming responses
• Dual-mode routing — Discord uses before_dispatch + CLI delivery; TUI uses before_model_resolve + relay server
• Lock file — .webmodel-lock in per-instance OPENCLAW_HOME/.openclaw/ directory
• Config model switch — On lock/unlock, updates both agents.defaults.model and agents.list[].model for session routing
• Plugin cooperation — Works with openclaw-superpowers via deferTo mechanism

中文

这是什么？

openclaw-webmodel 是一个 OpenClaw 生命周期插件，让你的 Agent 可以免费使用 ChatGPT、Grok 等网页版 AI 模型。通过浏览器自动化控制你已登录的浏览器会话，无需 API 密钥，无按量计费。

功能

• ChatGPT 集成 — /chatgpt <消息> 单次调用或 /chatgpt 锁定模式
• Grok 集成 — /grok <消息> 聊天，自动检测画图请求
• 锁定模式 — 执行 /chatgpt 或 /grok 后，后续所有消息自动走对应模型
• 自动会话重置 — 切换模型时自动创建干净会话
• 双模式路由 — Discord 和 TUI/HTTP 路径完全隔离，互不干扰
• 多实例支持 — 每个网关独立 relay 端口（网关端口 + 1000），无冲突
• 中继提供商 — 内置微型 HTTP 服务器作为 OpenAI 兼容代理（TUI 模式）
• 图片生成 — Grok Imagine 和 ChatGPT DALL-E，自动检测并本地保存图片
• 升级 CLI — openclaw-webmodel upgrade 一键检测旧版、删除、升级

命令

| 命令 | 说明 | |------|------| | /chatgpt <消息> | 发送消息给 ChatGPT，自动锁定模式 | | /chatgpt | 切换 ChatGPT 锁定 | | /chatgpt off | 解锁，恢复默认模型 | | /grok <消息> | 发送消息给 Grok，自动锁定模式 | | /grok | 切换 Grok 锁定 | | /grok off | 解锁，恢复默认模型 | | /chatgpt-auth | 从 Chrome 保存 ChatGPT 登录凭据 | | /grok-auth | 从 Chrome 保存 Grok 登录凭据 |

安装

openclaw plugins install openclaw-webmodel
openclaw gateway restart

插件首次启动时自动配置所有必要项（relay provider、auth profile、插件注册）。然后用 /chatgpt-auth 和 /grok-auth 保存浏览器登录状态即可使用。

升级

openclaw-webmodel upgrade

自动检测旧版安装、删除旧文件、安装最新版本。

License

MIT