AI CRM: local intelligence, inside the conversation

01 /The problem

Client conversations needed support beyond office hours. A useful agent had to work inside the existing WhatsApp and Laravel CRM flow, while keeping recurring model costs under control.

02 /What shipped

The WhatsApp API connects client messages to the Laravel application. Laravel orchestrates the CRM workflow and calls a local model through Ollama to generate replies.

Using local inference removes the per-message charge from a paid LLM API. The agent stays part of the production CRM rather than becoming a separate tool for staff to manage.

03 /My responsibility & architecture

My contribution was the Laravel integration and AI automation connecting WhatsApp conversations, CRM workflows and local inference. Laravel coordinates messages and model calls while Ollama supplies the local language model. This keeps the automation in the existing application workflow.

04 /Operational tradeoffs

Local inference removes a paid API charge per message, but still needs compute and operational support. It shifts the engineering concern toward the availability and capacity of the model host. That distinction matters when discussing cost: the system avoids paid LLM API usage, rather than making inference free.

05 /Result

The agent handled 68% of client chats without paid LLM APIs, with replies available 24/7. The result combines the application backend, a messaging integration, and local inference in one operational workflow.