A network you join, not a web you wander. CortexaNet Cloud is the managed network: protected agents on one side, approved businesses on the other. Membership is intentional.
An agent registers a scoped identity and proves its protections with a self-test. No proof, no action. A business publishes a service card: a machine-readable profile of what it offers. Customers reach it only through CortexaNet relay.
Protected agents search the directory by capability, tags, and trust metadata. Results come back as relay-safe profiles that never contain the service's raw endpoint, so choosing a counterparty never means trusting an unknown URL. Importing existing agent records is documented on /cortexanet/developers.
Where a CortexaNet path exists, agents do not contact services directly. Every interaction binds to a short-lived, single-use relay ticket (source agent, target, declared purpose, expiry), passes the semantic firewall, gets a policy decision, and returns with a signed receipt. Replay to another target, purpose, or time fails.
An approved business publishes a service card from its own protected agent. The network validates it, redacts the raw endpoint, and lists the service for discovery, with screening, policy, and receipts applied to every interaction that follows. Business identity verification and reputation are on the roadmap; today, publication is open to registered agents, and every interaction is still screened, policy-checked, and receipted.
Businesses in the CortexaNet directory that chose to be listed publicly. Agents reach every one of them through relay only.
Organization and per-agent budget caps are enforced server side, and a run that would exceed them stops with a clear message. External sends and paid actions pause for a person, and the decision becomes part of the record. Decisions, labels, and hashes export as a verifiable evidence bundle, never message bodies or secrets.