Reference Datasets on pgvector (extension for PostgreSQL)

Originally, the application relied on an in-memory file to store and provide relevant data to the AI provider for Suggestion Assistant. This approach required loading the entire RAG dataset into memory for every user request, which can lead to slow response times.

The in-memory solution also has a hard limit of 140,000 records. Beyond this point, the system is unable to generate or process datasets, restricting its ability to handle larger client volumes.

From v2025.2, the application has the option to host RAG datasets in a pgvector backed PostgreSQL database. This enables direct querying of large datasets without the need to load everything into memory.

Using a pgvector database enables storage and retrieval of millions of embeddings, leveraging PostgreSQL’s persistence and indexing capabilities to eliminate memory-bound limitations. This approach supports dynamic updates to datasets without requiring a full reload, while significantly improving end-user experience by delivering faster and more efficient suggestions compared to the in-memory solution.

Please contact Evotix representative to enable this option.