Sonar Vortex cuts coding-agent token costs 5-36% with a code graph
Swapping grep-and-read for typed graph queries also cut agent-made defects 92%. Plus: Sonar's fix-it agent hits GA, and Copilot's review quota is set to rise.
Copy markdown
The context tax, in dollars
On one ~800-line PR, the grep-then-read loop burned 156M context tokens—about $41—and averaged ~$65 across 18 PRs. Vortex answers 'what does this method return?' from a typed dependency graph (SemSitter) instead of re-reading whole files across 400+ turns, cutting agent token use 5-36%.
92% fewer bad diffs, over MCP
Vortex front-loads architectural context before generation, then runs CI-grade analysis on the diff in seconds—cutting the issues a leading coding agent produced by 92%, at a 3.2% false-positive rate. It drops into Claude Code, Cursor, or Copilot via a local MCP server or the SonarQube CLI.
Sonar's fix-it agent goes GA
The SonarQube Remediation Agent left beta: it hunts reliability, security, and maintainability issues on its own and opens PRs for you to approve, on GitHub and Azure DevOps. It's bundled with Vortex under the new Sonar Agent Essentials tier.
Elsewhere: Copilot review is about to cost more
From Sep 28, GitHub Copilot code review flips its default effort from 'Lite' to 'Balanced,' spending more premium requests per review unless you switch it back—and Chat history retention stretches from 28 days to the life of your account.