When to use Net#

Net earns its place when work crosses a machine or authority boundary and the caller cannot reduce the operation to one fixed endpoint.

Use Net when#

• Providers appear, disappear, move, or change availability while the application is running.
• Work spans personal devices, edge nodes, services, agents, or organizations.
• The provider must keep its credentials and enforce policy locally.
• Discovery authority and invocation authority need different rules.
• The call produces streams, durable state, long-running task progress, or content-addressed artifacts.
• Several providers offer the same capability and selection depends on health, locality, resources, organization, or request policy.
• The application must distinguish transport acceptance, execution, and verified outcome.
• The same identity must remain attached to discovery, invocation, state, and artifacts as topology changes.

Typical examples include a personal agent using tools across trusted devices, a fleet application resolving equipment capabilities, a company invoking a partner's service without receiving its credentials, or a scheduler placing work near data and compute.

Use a smaller system when#

• A single server and database already solve the problem.
• The caller knows one stable endpoint and request/response is the complete lifecycle.
• An MCP host has a fixed set of tools and no runtime discovery is needed.
• A normal queue or job runner already matches a fixed producer/consumer workflow.
• NATS or another messaging system already supplies the required subjects, durability, and operational model inside one domain.
• Zenoh or another data fabric already matches a system organized around keys, telemetry, and distributed queries.
• You only need one narrow feature that Net also happens to contain.

The dividing question is not whether the system is distributed. It is whether the application needs to address work under identity and authority, rather than a known endpoint, subject, or data key.

See How Net relates to other systems for the models side by side.