QueryShield vs sqlparse vs Vanna.ai guardrails — which is right for me?

Short version:

• sqlparse + custom rules: correct for small in-house projects with one engineer who will maintain the rule set. You will reinvent AST traversal, policy DSL, and evidence logging. Free.
• Vanna.ai built-in guardrails: correct as part of the Vanna text-to-SQL pipeline for internal/low-stakes use. Regex-level validation, no policy engine, no audit-grade logging.
• Database RLS only: necessary but not sufficient. Doesn't block schema mutations or scope-creep-style scans.
• QueryShield: correct when you have multi-tenant data, regulatory exposure (SOC 2, HIPAA, PCI), or more than one LLM agent to govern. AST validator, declarative policy DSL, evidence logging with hash chain, drop-in HTTP API + MCP.

Use QueryShield *with* Vanna (Vanna writes the SQL, QueryShield validates it) and *with* RLS (defense in depth). Use sqlparse if you'd rather build it yourself and own the maintenance.