Self-hosted AI analyst shows all SQL queries, self-checks answers

The core problem this tool addresses is a common frustration in business analytics: AI systems that answer questions confidently but provide no way for users to verify accuracy. Traditional "chat with your data" tools function as black boxes, leaving users to either trust the output or manually reconstruct the reasoning — a time-consuming and error-prone process. By placing the entire working on screen — including the SQL queries, row counts, and explicit statements about which data sources were chosen and why — this tool inverts the typical trust model. Users can now inspect the logic and data at every step.

The developer's example illustrates the transparency in action. When asked whether Engineering's hiring growth reflected actual headcount expansion or was backfilling departures, the tool first explained its reasoning: it found that the governed semantic model lacked a hires metric, so it deliberately chose to fall back to a raw monthly table instead. This kind of explicit acknowledgment of source limitations and substitution decisions is absent from most AI analytics tools. By combining this transparency with automatic self-checking — where the AI flags its own potential errors rather than relying on users to spot them — the tool attempts to combine convenience with accountability.