OpenAPPA

OpenAPPA is an information-flow policy engine that sits between AI agents and their tools, checking whether data is allowed to reach each destination before tool calls execute. It uses APPA (Agentic Permissions Policy Algebra) to track data sensitivity and trust across agent actions. Unlike probabilistic classifiers, OpenAPPA makes deterministic decisions from event logs alone, returning identical results on every run. Policy is written in declarative TOML and runs in-process or as a sidecar. Built with Claude Code, it's MIT-licensed and vendor-agnostic. Benchmarks show 88–90% task completion with zero successful attacks across 1,320 evaluations, outperforming alternatives like Microsoft FIDES and Claude's auto-mode.