Skip to content
Back to explore

Discovery

Find a repository from a screenshot

Saw a tool in a talk, a thread or a terminal and only have a picture? Upload it and the catalog works out what it probably is, showing you exactly which clues matched.

How this works, and what it will not do

  1. 1. The file is checked by its actual bytes, not its claimed type, and stored privately under your account.
  2. 2. A vision model extracts only what is visible: names, owners, URLs, domains and technology names.
  3. 3. Those strings are matched against the catalog and, if needed, GitHub's search API. A URL in your image is parsed for an owner/name pair, it is never fetched.
  4. 4. Candidates are scored in code from which clues matched. The model never asserts a confidence number.
  5. 5. Nothing is added to the catalog automatically. An unmatched repository goes through the normal submission queue.