Openwork

OpenWork

OpenWork 是一个面向组织部署的浏览器智能体平台。 它把聊天工作台、管理后台、产品网关和 OpenSandbox agent runtime 组合成一套可落地的系统：部署完成后，团队只需要创建账号并配置模型，用户就可以直接在 Web 应用里完成对话、工具调用展示、Monaco 工作区编辑、终端和文件交互。

OpenWork is a browser-based agent platform for teams that want a deployable, web-first AI workspace. It combines a user-facing chat workspace, an admin console, and a backend gateway so teams can deploy once, configure models centrally, and let users work directly in the browser.

OpenWork 是什么

OpenWork 面向这样一类场景：

• 希望把智能体能力部署到自有环境，而不是只依赖托管产品
• 希望统一模型配置、搜索配置、权限和运行时策略
• 希望用户直接在浏览器里完成对话、工具执行展示和文件交互
• 希望通过 OpenSandbox 接入隔离运行时、openwork CLI 和 CLI agent

从架构上看，OpenWork 不是“单个聊天应用”，而是一套由前端、后端和部署流程共同组成的平台仓库。

What OpenWork Is

OpenWork is designed for teams that want more than a thin chat UI:

• a browser-based AI workspace for end users
• centralized admin control for models and platform settings
• a backend gateway that owns auth, billing/quota, logging, and routing
• an OpenSandbox runtime layer for isolated CLI agent execution and project workspaces

The repo is intentionally structured as a platform codebase rather than a single product surface.

Core Capabilities

• Browser-based chat workspace, Monaco runtime workbench, file operations, terminal, and tool execution display
• Centralized admin console for model, search, and platform configuration
• service as the product gateway for auth, billing/quota, logs, model dispatch, OpenSandbox runtime routing, and PTY/workspace API proxying
• Generic platform search integration routed through the backend
• openwork-agent-runtime image with bridge, Claude Code instructions, and runtime openwork CLI

Architecture

当前主流程如下：

chat frontend
  -> service /api/openwork/*
     -> auth, billing/quota, logs, group state
     -> model provider API
     -> OpenSandbox sandbox
        -> openwork-agent-runtime
           -> openwork-agent-bridge
              -> Claude Code

几个关键边界：

• service/ 仍然是产品网关，并负责 OpenSandbox runtime 定位、bridge 连接、P