NILScript

NILScript is a governed action layer for AI agents that enforces a structural boundary between proposing and committing actions, ensuring that only declared intents can be executed. It solves the problem of AI agents performing unauthorized or unintended actions by making undeclared actions impossible to name, rather than relying on probabilistic filtering. The product targets developers and IT teams in B2B settings, particularly those building or integrating AI agents into enterprise systems. It is positioned as an infrastructure API that provides a deterministic, model-independent guarantee of safety, contrasting with traditional filter-based approaches that accumulate risk over time.

Key features

• Propose-approve-commit-rollback loop
• CLI toolkit for building adapters
• Scaffold shim for any backend
• Local reference demo
• Verb catalog from standard
• Structural guarantee of zero effect
• Model-independent safety
• O(1) boundary instead of O(n) checks