Welcome to Core Sentinel

Your command center for privacy, model health, admin security, and active learning. Monitor risk trends, review recent activity, and keep AI usage compliant across teams.

Current production support: Windows 10/11. Linux and iOS support are planned for a future version.

Clipboard monitored · Risk engine active

New user path (recommended order)

1) Interactive Demo Experience document-to-LLM paste interception, warn/block flow, and remediation. 2) Privacy Dashboard Understand personal risk trends, streaks, and recent events. 3) ML + Admin Views Validate model health in ML dashboard, then inspect org-level risk in Admin Dashboard.

Safe Streak

—

sentinel_streak

Total Pastes Scanned

—

sentinel_total

Admin Events

—

Live

Model Quality

F1 0.826

Recall (L/M/H): 1.00 / 0.41 / 0.99

Last retrain: --

Safety first: catching high/critical risk matters more than a single headline score—see repo README (Training and Evaluation).

🛡️

Privacy Dashboard

Charts, streaks, paste history, risk timeline from local data

Risk trend over time + high-risk windows
PII category frequency and streak behavior
Recent paste timeline with action outcomes

Why click: understand personal exposure patterns and where leaks are happening.

🧠

ML Health Dashboard

Precision-Recall curves, confusion matrix, drift detection, threshold simulator

Class-level precision/recall and confusion matrix
Threshold tuning and false-positive controls
Model quality drift and QA snapshot view

AUPRC: 0.751 Precision: 0.945 FPR: 52.6%

Why click: verify model reliability before demos, reviews, or releases.

🏢

Admin Dashboard

Organization-wide monitoring, per-user risk profiles, policy enforcement

🔒 Admin sign-in required

Locked content includes user-level analytics, alert queues, risk-category breakdowns, and policy controls.

🔒 Admin sign-in required

📜

Recent Activity

Last 20 paste events with risk scores and actions taken

Time	Risk	PII Type	Action	Status

🧪

Feedback & Active Learning

Your corrections improve the model. See training progress and impact.

Correction volume by class and direction
Retrain readiness and data quality signal
Feedback-to-training conversion visibility

47 corrections collected 12 used for training Next retrain: 53 more needed

Pattern: high→med (12) Pattern: med→low (9) Pattern: high→low (6)

Why click: identify where users disagree with the model and prioritize retraining.

Launch Interactive Demo View GitHub

📚 Knowledge Base Manager (Admin)

Upload compliance documents

Drop files here or click to upload

Active KB entries