Most AI CRMs bolt a chat box onto a form
Classic CRM is a database with data entry. AI CRM often means a sidebar chatbot on top of the same seat tax. The work — finding what is true about a company or person, and writing it down carefully — still falls on humans who should be selling or building.
Agentic-first: CRM as the agent's notebook
trycompai/crm (MIT, TypeScript, ~4k stars and rising) builds the other way: a durable research agent runs on its own schedule, decides what to look at next, spends a research budget, and stops when the budget is gone. The API stays dumb on purpose — events go on a queue; the agent decides meaning. Strong evidence updates the record. Weak evidence becomes a human suggestion. The rule: nothing about a person is guessed.
Why ownership matters even more here
An agent that researches your pipeline should not live only inside a vendor cloud you rent forever. Self-hosted, single-tenant CRM with Postgres you control fits operators who want IP and data on their side of the table. We deploy and shape open stacks like this when they fit — phone-first ops systems for the floor, agentic CRM for the relationship graph — under the same day-one ownership rule.
Simple first still wins
Do not replace every tool at once. If sales research and CRM hygiene are the leak, an agentic open CRM may be the first owned system. If the floor is the leak, start with ordering or kitting. Compose the stack; do not buy a religion.
Curious whether an agentic open CRM belongs in your stack? Reach out — we listen first, then pick foundations you can own.