MemoryLint

MemoryLint is the IDE-native trust and governance layer for AI-written code inside JetBrains. AI coding agents don't govern themselves. They can remember deleted code, violate architecture rules, leak context across projects, and produce changes that are hard to audit. MemoryLint sits between the agent and the keystroke: it decides what agents are allowed to remember, recall, and act on — and proves every decision with a signed audit trail. In the demo, MemoryLint shows up as a real JetBrains plugin: a live red squiggly appears on an unsafe System.getenv(...) call, Alt+Enter triggers a governed auto-fix via OpenAI Codex, the line is rewritten to ConfigService.get(...), the missing abstraction is scaffolded automatically, and the decision is logged to audit with a deep link into the dashboard. Underneath that IDE experience, MemoryLint uses: AuthZed / SpiceDB for Zanzibar-style authorization on governed recall Nebius Token Factory for multi-model consensus on high-severity decisions Supabase Storage for signed compliance evidence packs Neo4j for provenance and blast-radius traversal FastAPI + Next.js + JetBrains plugin as the product surface This project maps cleanly to all three hackathon tracks: Writing & Generating Code — governed code rewrites inside JetBrains Testing & Debugging Code — ShadowQA mutation testing and TraceOS incident analysis Reviewing & Deploying Code — Ship Gate, evidence packs, and audit-backed governance Assistants generate. MemoryLint governs. Built solo in 48 hours. 1,900+ tests, 65 dashboard pages, 42 backend services, 22 Kotlin files across 8 IntelliJ SDK surfaces.