The reference corpus is local. User queries are stored locally in SQLite. Retrieval traces, latency traces, answer records, and evaluation results are stored locally.
Ollama runs locally, and no cloud API is required. Privacy behavior changes if the project is modified to call external services, remote models, hosted databases, telemetry systems, or third party analytics.