crm: Open-source CRM recorded directly by AI agents

While existing CRMs are tools for humans to input data, this project is a system where AI agents independently research and record customer information. Agents operate on their own schedules in isolated deployment environments and continue working in the background even when the browser is closed. Rather than offering simple chat functionality, the CRM itself serves as the agent's notebook.

The agent adheres to a principle of no guessing. It does not allow the model to assign its own confidence levels and records only evidence that has been actually observed. Strong evidence is saved directly to the database, while weak evidence remains as suggestions for human judgment. This design prevents incorrect information from permanently persisting in customer records.

It works by analyzing its own email threads and signature blocks without requiring any external API keys. Adding LinkedIn or Brand Data keys enables the collection of a broader range of information. The sandbox environment blocks network access, fundamentally preventing any path through which customer email content could leak externally.

Built on eve, Vercel's durable agent framework, it uses a Next.js, NestJS, and Postgres stack. Designed with a single-tenant architecture, it eliminates unnecessary organization management logic. Developers can observe in real time the steps the agent takes to derive information and intervene directly when necessary.