A new way to search code on GitHub

GitHub is rolling out a technology preview of a substantially improved code search experience. The preview introduces a dedicated interface at https://cs.github.com, separate from the main github.com site for now. The company plans to fold the new search into the standard GitHub experience once it's satisfied with feedback and ready for broader adoption.

The initial index covers more than five million of the most popular public repositories, with support for searching private repositories you already have access to. The tool is designed around several key capabilities:

  • Smart ranking that prioritizes relevant results, backed by an index optimized for code rather than general web content.
  • Exact string matches with support for substrings and special characters, plus regular expression search using / delimiters.
  • Scoping via org: and repo: qualifiers, complete with auto-completion in the search box.
  • Result refinement with language:, path:, and extension: filters, Boolean operators (OR, NOT), and symbol definition lookup with symbol:.
  • Code reading aids including a directory tree, symbol info for the current scope, jump-to-definition, and select-to-search.

Full syntax documentation is available on the preview site, and hitting ? on any page shows the keyboard shortcuts. The project also has an FAQ and a dedicated feedback forum.

Roadmap and open questions

The preview is intentionally limited in capacity, and GitHub is using it to gather input on priorities. Planned work includes expanding the index to cover every repository a user can access, iterating on ranking heuristics, exploring API and integration options, and broadening language-specific features. The company is also soliciting novel use cases and feature ideas from early testers.

Search in the context of developer productivity

GitHub frames the code search overhaul as part of a broader push on developer productivity, alongside Copilot and Codespaces. The stated goal is to minimize the interruptions that come from looking up library usage, diagnosing environment drift, or tracing error messages. The new search aims to reduce that friction by surfacing the most relevant results quickly, offering auto-completion throughout the query process, and providing a browsing experience tuned for reading unfamiliar code even outside an IDE.

GitHub also credits external contributors whose work underpins the project, including @BurntSushi and @lemire. More engineering deep-dives are promised as the preview matures.