GitHub’s AI agent turns raw data into instant insights
GitHub employees can now skip the SQL and dashboards by typing questions into Slack or VS Code and receiving near-instant answers from Qubot, the company’s new Copilot-powered analytics agent. Instead of waiting for data teams to craft queries or interpret telemetry, any team can explore metrics in plain language and receive markdown reports they can reuse or share.
From idea to instant answers
Qubot started as a way to make GitHub’s sprawling data warehouse self-service. Traditional BI tools and dashboards often leave teams stuck when they need to dig deeper or test new hypotheses, so GitHub built an agent that interprets natural-language questions and turns them into validated answers. The system is designed for exploratory questions—like which user cohort has the best retention on a feature or which product drove a metric shift last week—rather than scheduled reports.
How the three-layer engine works
The architecture splits the workload into three parts. First, the user interface lets employees ask questions in Slack, VS Code, or the Copilot CLI. In Slack, a dedicated channel spawns a Qubot instance that posts the answer back into the thread and creates a markdown report in a pull request for later use. In VS Code and the CLI, users install Qubot as a plugin and query datasets alongside other custom tools.
A federated context layer feeds the engine with curated knowledge. Bronze-level raw telemetry carries schema and metadata provided by product teams, silver-level conformed data includes query examples and mandatory filters maintained by analytics, and gold-level curated datasets contain business rules and metric definitions from their owners. The context is refreshed continuously by ETL pipelines that add derived metadata, ensuring answers stay aligned with the latest definitions.
Built for speed, zero maintenance
Qubot runs as a Copilot Cloud Agent on github.com and incurs no ongoing maintenance cost, letting teams ramp up quickly on unfamiliar datasets. Because every answer is saved as a markdown snippet, users can iterate in Slack threads or copy the query directly into dashboards, reducing duplicate work and shortening the path from question to insight.
Source: GitHub Blog. AI-assisted editorial synthesis — TechnoExpress.