SentinelAI — Local‑First Repository Intelligence

SentinelAI is a local-first repository intelligence runtime for developers who want continuous, explainable insight without sending source code to the cloud. It is not a one-off scanner and not a chatbot. SentinelAI acts like a background intelligence layer: it learns the repo over time, stores memory in local SQLite, and uses that memory to make future scans faster and more precise. The goal is to make code quality and security checks feel automatic and safe, while keeping data fully on the developer’s machine. SentinelAI runs in three modes. Mode 1: Silent Watcher updates a live dependency graph as files change. Mode 2: Commit Guardian runs at commit time and inspects only staged changes using pre-built intelligence, keeping checks fast and reducing false positives. Mode 3: Deep Scan runs on demand and executes all agents across the repo and history to build a full intelligence graph that powers trace, impact, and explain. The system uses eight coordinated agents: SCOUT maps the repo, GHOST HUNTER finds dead code, PRISM traces sensitive data flows, DOMINO computes blast radius, ARCHITECT enforces rules, FIXER applies safe fixes or records dry-run diffs, NARRATOR generates human reports, and VERIFIER validates AI suggestions with confidence scoring. Each finding is stored with severity, context, and history so the system improves over time instead of forgetting. SentinelAI integrates with IBM watsonx (Bob) in a privacy-preserving way. Findings are compressed through a Caveman pipeline before any external call, so Bob receives only compressed intelligence, never raw source files. A live dashboard (Next.js + SSE) shows agent status, health scores, findings, and trends. Everything runs locally: no hosted server, no external database, and no centralized storage. Developers install once, activate per repo with sentinel init, then run sentinel watch for continuous intelligence or sentinel scan --deep for full audits. SentinelAI’s core value is persistence: