Isonapse

About Isonapse

Isonapse is an independent runtime governance layer for AI agents that sits between agents and the resources they can access: files, the shell, the web, and custom tools. It allows you to define policies that specify what agents may do, what requires human approval, and what is never allowed.

The system operates through three verbs: DECIDE (you set the rules as the source of truth), ENFORCE (every action passes through a checkpoint that allows, asks, or blocks it), and PROVE (every decision is written to a tamper-evident, signed ledger for offline verification).

Isonapse is vendor-neutral, working across any agent, model, or harness. It uses deterministic code for critical decisions rather than relying on language models for governance, ensuring that a compromised agent cannot escalate its own privileges. The system works locally without calling home, maintaining full enforcement capability even when offline.

Built in Switzerland with European compliance in mind, Isonapse is offered as a license for your own hardware rather than a SaaS account, with versions for individual machines, organizational servers, and enterprise fleets.

• self-hosted
• compliance
• local first
• audit trail
• ai governance
• policy enforcement
• deterministic

FAQ

How does Isonapse prevent prompt injection attacks?

The governance gate sits outside the agent, so a hijacked or poisoned agent cannot argue with or bypass it. A malicious prompt cannot talk its way past the deterministic policy enforcement.

What happens if the network is disconnected?

Everything degrades gracefully and enforcement keeps working offline. A pulled network cable does not stop enforcement because the critical control path is deterministic policy and bounded local state with no network round-trip required.

How is Isonapse different from built-in model controls?

Every agent harness and model vendor ships its own controls, but those controls stop at vendor boundaries when you use multiple frameworks or open-weights models. Isonapse is a neutral layer across all of them, owned by you, with one policy and one audit chain.