Arcade AI

Arcade 把自己定位成给企业 AI Agent 用的 actions 运行层：Agent 要调用 Gmail、Slack、Google Drive 这类第三方服务时，鉴权、权限范围与用户授权都由 Arcade 处理，开发者只写工具定义，不用为每个 SaaS 重做一遍 OAuth。它按用户粒度管权限，Agent 只能拿到被授权的那部分数据，审计日志能追到哪次调用用了谁的凭据。文档里有 Python 与 TypeScript 的接入方式和预置工具清单。 ⭐ GitHub: ArcadeAI/arcade-ai — 1.1k stars 🔗 文档: https://docs.arcade.dev/

Arcade positions itself as an actions runtime for enterprise AI agents: when an agent calls Gmail, Slack or Google Drive, Arcade handles authentication, scopes and per-user consent so the developer only writes the tool definition instead of rebuilding OAuth for every SaaS. Permissions are managed per end user, an agent sees only the data that user granted, and audit logs trace which call used whose credentials. The docs cover Python and TypeScript integration plus the prebuilt tool catalogue. ⭐ GitHub: ArcadeAI/arcade-ai — 1.1k stars 🔗 Docs: https://docs.arcade.dev/