cAIj: The Cage Between AI Agents

Why it surfaced

Its central proposition is unusually crisp: models and agents are replaceable workers, while deterministic rules and durable state decide what counts as done. The page lays out six concrete guardrails—including local-first storage, expiring authority, token metering, audit trails, and fail-closed behavior—instead of treating AI safety as a matter of better prompting.

cAIj presents a locally owned governance layer for AI-assisted work, keeping goals, evidence, approvals, permissions, and run history on hardware the operator controls.