WrenAI with Ollama: setup, SQL queries and limitations

In this Italian-language walkthrough, Dimitri Bellini reviews WrenAI for querying databases in natural language and turning results into charts. His self-hosted setup uses Ollama with Qwen 3, which he recalls as a 32-billion-parameter model. He finds its answers acceptable, while stressing that results depend on the model.

The setup discussion covers Docker and a GPU for local inference, with cloud providers as an alternative. Bellini uses the Linux launcher and describes adapting an Ollama configuration file and environment variables before starting the Docker Compose stack. He shows the installed interface rather than a complete installation walkthrough.

Using a PostgreSQL sample database about DVD rentals, he explains how to connect a data source, inspect table relationships and add missing relationships manually. Questions about revenue by film category produce results, SQL that users can inspect, and charts they can pin to a dashboard or export as SVG or PNG.

A live chart request runs slowly on his virtual machine with RTX 8000 graphics cards. Bellini points to storage and inference hardware as possible causes; the demonstration does not establish a general performance benchmark. His main criticism concerns the open source edition he tests: it accepts one data source, and changing it requires resetting the project. He argues that this restriction limits its usefulness for everyday database analysis.