In March, we shipped risk intelligence reports and file heatmaps. In April, we gave the risk engine eyes.
The biggest change this month: Re-entry's risk assessments now use AI to read and understand your actual code changes. Not just file names and line counts — the actual intent behind the diff.
AI-Powered Code Comprehension
Until now, risk scoring was deterministic: file types, change size, branch targets, linked tickets, historical patterns. Effective, but limited. A 500-line change to a test file and a 500-line change to authentication middleware scored similarly.
Now the engine runs AI-powered semantic analysis on every assessment. It reads the diff, understands what the code does, considers the context of the repository, and factors that understanding into the risk score. This is was intentional. Our goal was to build a strong and battle tested deterministic model that will continue to be the foundation of our Risk scoring.
What this means in practice:
- Test-only changes are recognized as low-risk even if they're large
- Authentication and payment code is flagged even in small changes
- Refactors that rename variables without changing behavior score lower
- New API endpoints that handle user data are surfaced as higher risk
The AI analysis doesn't replace the deterministic signals - it adds a layer of comprehension on top. Deterministic scoring catches structural risks (wrong branch, no ticket, sensitive file types). AI comprehension catches semantic risks (this code disables rate limiting, this migration drops a column).
The result: dramatically fewer false positives. The changes that matter get flagged. The ones that don't, don't.
Self-Hosted Model Support
For teams where code can't leave your infrastructure, Re-entry now supports self-hosted AI models.
Connect any OpenAI-compatible inference server - including Ollama running locally - and Re-entry will use your model instead of our hosted one. The semantic analysis, policy evaluation, and risk explanations all run through your endpoint.
This matters for:
- Regulated industries where code must stay on-prem
- Air-gapped environments with no external API access
- Cost optimization - run a smaller model for high-volume repos, use the hosted model for critical ones
Configuration is a single endpoint URL and optional API key in your team settings.
Assessment Feedback
You can now tell Re-entry when it gets a risk score wrong.
Every assessment has a feedback button: mark it as accurate, or flag it as a false positive. Your feedback is recorded and used to calibrate the engine to your team's specific risk tolerance.
Over time, this creates a feedback loop: the more assessments you confirm or correct, the more the engine learns what "risky" means for your team specifically.
Deep Codebase Scanning
Risk intelligence reports now go beyond historical data with three new capabilities:
Team Health Baselines - see how your PR sizes, file change counts, and directory spread compare to your own team's rolling averages. Spot anomalies before they become incidents.
Active Scan - run an on-demand scan that checks your GitHub repositories for unassessed PRs, unprotected default branches, and stale pull requests that have been open too long.
Dependency Audit - scan package.json, requirements.txt, go.mod, and other manifests across all monitored repos. Re-entry checks each dependency against the GitHub Advisory Database and flags known vulnerabilities with severity ratings.
All three run as background jobs and report results directly in the risk intelligence panel. No new tools to install, no CI pipeline changes.
Settings Redesign
The settings pages got a full visual overhaul. They now match the operational design language used across the rest of the dashboard - clean cards, consistent spacing, and inline confirmations instead of browser dialogs.
Bitbucket Security Hardening
The Bitbucket integration now uses explicit OAuth scopes, proactive token refresh before expiry, and mandatory webhook signature verification. If you connected Bitbucket before this update, the improvements apply automatically.
What's Next
We're focused on incident correlation (connecting Sentry errors to file risk profiles), GitLab support, and the first compliance guard packs (SOC 2 and ISO 27001 policy templates). Follow the changelog for real-time updates.
Questions? Reach us at info@re-entry.ai or through the command bar in the dashboard.