AI coding agents often outpace human review, with many holding direct access to infrastructure—migrations, restarts, configs—without guaranteed reversibility. Synapse solves this with two halves: UNDERSTAND, a live hybrid knowledge graph (AST + LLM edges) for onboarding and impact tracing; and PROTECT, which intercepts risky ops, scores blast radius, checks conflicts, generates reversibility plans, verifies invariants, and auto‑rollbacks on failure. Unknown ops fail closed. Developers and platform teams gain structural safety, live visibility, and verified rollback—unique in combining conflict detection, reversibility contracts, and lightweight deployment (SQLite, networkx, tree‑sitter).
Category tags: