July 12th, 2026
Generated reports are now stored and retrievable at any time. No need to regenerate a report you already produced: open the history, find it, download it.
Retention by plan: 90 days on Scale, 365 days on Enterprise.
VibeDefend's Action Guard no longer relies on pattern matching alone. Every tool call is now evaluated by a hybrid engine combining deterministic regex rules with LLM judgment.
Fewer false positives: legitimate actions are no longer blocked because they superficially resemble a dangerous pattern.
Broader coverage: complex or obfuscated dangerous actions performed by an AI agent, which no regex could reasonably catch, are now detected and blocked.
VibeDefend's fourth capability, fully integrated with Claude Code, Codex, Cursor, Copilot, Windsurf. On every file modification, the generated source code is analysed on the spot to surface security vulnerabilities introduced by the agent, in the session, before the code ever reaches a commit.
No context switch: the finding is raised where the code is written, not hours later in a scan report.
Covers what the agent just changed, not the whole repository, so feedback stays immediate.
Fixed "Autopilot proposals as you code" on Claude Code: business and logic rule mining is working again.
Various frontend fixes.
Please update VibeDefend to v1.3.0 to access these features.