Tags : Browse Projects

Lightweight SDK that makes agentic AI safer and more auditable.Before executing any tool call or side-effect action, the agent must propose a structured JSON contract declaring: - explicit intent - impact classification (read / write / money / irreversible / privacy / etc.) - provenance sources ... [More] with trust levels (untrusted → trusted via verification) - claims backed by verifiable evidence (SHA-256 file hashes or Ed25519 signatures) Verifier checks everything locally and fails closed if provenance doesn't match or high-impact actions lack trusted evidence. Integrates easily as middleware (LangGraph tool-node wrapper) or via MCP for enterprise tool guarding. [Less]