Composed authority
An agent has no trust by default. The safety layer (substrate) combines all rules. It decides exactly what the agent is allowed to do. It issues a temporary permission for each task.
Give your agent a backbone.
Most tools ask what your AI agent can do. OpenSpine asks what it's allowed to do — and proves it.
Three pillars
Section titled “Three pillars”One gate
The agent must pass through a single gateway for any action. This checks what the agent does in the real world (effectful action). We write every decision to a tamper-evident log (hash-chain audited).
Provable claims
We do not just say OpenSpine is safe. Every security claim maps to a real test. Our check script runs all tests on every change.