Compai团队发布了一款名为CRM的开源项目，采用agent-first架构，将AI智能体作为产品的核心，而传统的CRM数据库则被定位为智能体的记录工具[1]。该系统采用MIT许可证开源[1]。
该系统能够在独立部署环境下自主运行，智能体可以自行决定下一步操作、自动预约后续任务、控制研究预算，无需人工交互干预[1]。系统遵循严格的数据完整性规则，不对个人信息进行推测，工具仅报告观察到的数据，强证据会被写入记录，弱证据则转化为建议[1]。
在技术实现方面，该项目采用Turborepo单体库架构，使用Bun和Vercel进行部署，后端基于NestJS API，前端采用Next.js应用，数据存储使用Postgres数据库[1]。系统采用单租户内部设计，支持Google登录、环境变量白名单管理，所有用户可查看全部内容[1]。此外，该系统支持可选的外部数据源集成，包括LinkedIn、Web搜索、公司品牌数据等API[1]。
The Compai team has released an open-source CRM project that fundamentally reframes customer relationship management around autonomous AI agents rather than traditional software features [1]. In this architecture, the AI agent serves as the core product, while the CRM database functions primarily as a record-keeping tool for the agent's activities and insights [1].
The system operates independently on its own deployment and schedule, making autonomous decisions about next steps, automatically scheduling follow-ups, and managing research budgets without requiring human intervention [1]. Data integrity is enforced through strict rules that prohibit speculation about individuals; all observations must be evidence-based, with strong evidence recorded as facts and weaker observations converted into suggestions [1]. The platform employs a single-tenant internal design featuring Google login authentication, environment variable whitelisting, and full data transparency across all users [1].
The technical implementation uses Turborepo as a monorepo framework, Bun as the runtime, and Vercel as the hosting platform, with NestJS powering the API layer and Next.js handling the front-end application, backed by a PostgreSQL database [1]. Users can optionally integrate external data sources including LinkedIn, web search capabilities, and company branding data through available APIs [1]. The project is released under the MIT open-source license [1].