Email Assistant: NL to SQL Queries with Phi-4 Mini & Postgres

This n8n workflow transforms natural language requests into precise SQL queries executed against a PostgreSQL database. Powered by the efficient Phi-4 Mini model running locally through Ollama, it ensures complete privacy with no external API calls or data exposure. On first run, it automatically extracts the full database schema by listing all tables, storing it for reference. Users can interact via n8n's chat interface or trigger it as a sub-workflow, receiving both the generated 'query' and 'sqloutput' results.

Benefits include high accuracy in query generation balanced with low resource use, making it ideal for sensitive data like email archives. It eliminates manual SQL writing, reduces errors, and speeds up data retrieval. Customizable prompts in the AI Agent node allow adaptation for specific tables or entirely new datasets, with easy modifications like filtering table lists via SQL conditions.

Primary use cases: Analyze structured email data (e.g., sender, subject, dates) stored in Postgres from other ingestion workflows. Extend to CRM reports, sales analytics, inventory queries, or any tabular data. Perfect for non-technical users querying databases conversationally, integrating seamlessly into larger automation pipelines for real-time insights.

Setup is straightforward: Connect your Postgres credentials and Ollama instance. The workflow handles schema discovery and query execution autonomously, outputting structured JSON for downstream processing.