Code review that knows what depends on your change.
Veridu is a hosted code review service for GitHub pull requests. It reads your diff, maps the parts of the repository that actually depend on it, and hands that change-impact evidence to two independent AI reviewers before they read the change. The findings are reconciled cross-vendor and returned as a verdict you can act on.
One credential from Veridu. No OpenAI or Anthropic account required.
For every signature the diff modifies, Veridu walks the code to find callers, interface implementers, and references to removed symbols - the kind of context that turns "this function looks reasonable" into "this function has three callers that pass null and one is a test." It is deterministic. It is not a repo-wide vector search.
Every review runs Codex and Claude in parallel on identical evidence. Findings both models raise at the same file and line become agreements. Findings that contradict become disputes surfaced for human review. Findings only one model raises are still shown, ranked by severity. Independent perspectives beat a single confident wrong answer.
Web dashboard, CLI, hosted MCP tool inside your coding agent, or auto-review on every PR via the GitHub App - same reviewer panel and reconciliation engine underneath. Pick the one that fits your workflow; switch whenever you like.
AI reviewers can be wrong. Veridu returns verdicts, not certifications. The terms spell out that reliance on any finding is on you; we do not use your code to train models and never will.
Private beta. Everyone gets 5 free review attempts on any public GitHub PR to try it end-to-end. Paid plans are coming; when they land, the free tier stays free. Private repos, per-call model selection, and team accounts land alongside the paid tiers.
Veridu is operated by T3 Interactive. Small team, direct line to the person who built the thing you're using. Say hi if you have feedback, feature requests, or want us to review a repo we haven't seen yet.