shuiyuan-rag-service

To keep a FastAPI RAG service persistent, maintain a daemon process that preloads embedding and reranking models in memory. This avoids repeated cold starts and provides shared HTTP search endpoints for connected clients.

Shuiyuan Community search becomes slow or unavailable when the RAG service needs cold starts or models are unloaded. Reloading BGEM3 embedding and BGE reranker models into a persistent FastAPI service restores retrieval availability.

Yes, a persistent FastAPI RAG service allows sharing embedding resources across multiple clients. By keeping BGEM3 embedding and BGE reranker models loaded in memory, connected clients can access the HTTP search endpoint simultaneously.

Limitations of a persistent RAG service include the memory overhead required to keep BGEM3 and BGE models loaded continuously. This approach is best suited for local campus knowledge retrieval scenarios requiring consistent availability rather than occasional queries.

Yes, the Shuiyuan RAG service supports troubleshooting and reload workflows. It provides operational diagnostics, health monitoring, and service management commands to restart the FastAPI daemon and restore retrieval capabilities.