GitHub: Internal Analytics Agent Qubot Reduced Query Resolution Time by Around 66 Percent

GitHub described the architecture of Qubot, an internal AI analytics agent that lets employees query the data warehouse using natural language instead of writing queries by hand. A data warehouse is a central database into which data from many systems flows, and querying it normally requires knowledge of query languages.

How Qubot Was Built

Qubot uses a three-tier federated context layer — bronze, silver and gold — which progressively refines and structures data so the agent receives more reliable context. It runs on Copilot Cloud Agent infrastructure, accessible through Slack, VS Code and CLI. The agent automatically routes queries between two engines, Kusto and Trino, depending on the type of data.

Results

According to GitHub, Qubot serves hundreds of users running thousands of queries, and reduced query resolution time by around 66 percent. Structured context, as opposed to sending raw data directly to the model, increased answer accuracy. The example shows how large technology companies are increasingly building agentic systems for their own internal needs, not just as products for the market.