Semantic information about source code is a central requirement for searching, navigating, and analysing code, all of which are important tools in a developer’s workflow. As the size of the codebase grows, collecting and maintaining accurate semantic information about the code becomes harder; meanwhile our tools demand richer and richer information. In this talk I’ll describe a system called Glean that we’re building at Facebook to address these problems. Glean is a datalog-inspired storage and query engine that collects facts about code from language-specific analysers and allows tools to query the facts. Glean also supports automatically deriving facts so that information can be presented at different levels of abstraction, from very detailed language-specific information needed by analysers, to the high-level language-independent relationships typically needed by IDEs.