Always-on layers

The runtime spine describes what happens to a read. The always-on layers describe what is true at every stage of it. These are not steps in the path, they are properties that wrap the whole path: privacy, x402, and guardrails.

Privacy

Chat and history stay local, and nothing sensitive is stored on intenthybrid AI servers. Your prompts run on infrastructure you control, and your conversation lives in your browser. This holds regardless of which stage of the spine is running. Privacy is not a setting you enable, it is a property of the design. See Privacy model.

x402

Every paid read clears over x402. A read is an HTTP endpoint that an agent or tool can pay for with a small amount of USDC, settled onchain in well under a second, with no keys, accounts, or monthly invoice. Pricing is per request, down to a fraction of a cent. This layer is what makes the product agent-native: a machine can pay for a read without a human in the loop. See x402 overview.

Guardrails

Every agent action is bounded by a cap, an allowlist, a slippage ceiling, and an expiry, with a kill switch on top. No matter what a verdict says or which output fires, an agent cannot act outside the limits you set. Guardrails wrap the act stage on every single call. See Guardrails and lifecycle.

Why these three are layers, not features

A feature is something that runs at one stage. A layer is something that holds at all of them. Privacy, pay-per-call, and bounded execution are properties intenthybrid AI is committed to keeping true everywhere, which is why they are drawn as wrapping the spine rather than sitting inside it. They are the difference between a tool that happens to be private and bounded today, and one that is built to stay that way.