Canary is an independent tester for code written by AI coding agents. It runs changes in sandboxes, looks for runtime failures, and reports issues before they reach production. The product is aimed at teams using tools like Claude Code, Codex, Cursor, or Copilot.