Standard Profiles

Profile is one of the core abstractions in NNRP. Its purpose is to keep the public layer stable while moving domain-specific interpretation into profile and schema combinations.

The public layer is responsible for:

1. Message skeletons.
2. Length models.
3. Shared semantics such as sessions, operations, and flow control.

Profiles are responsible for:

1. Identifying the semantic family of a payload.
2. Working with schemas to define the actual interpretation.
3. Preventing one business scenario from bloating the public layer.

Current public standard directions

tensor, token, and the preview4 runtime control profiles are the current standard profile directions.

Tensor Profile

For numeric blocks, region-oriented payloads, real-time enhanced rendering, and parts of media processing.

Token Profile

For discrete tokens, streaming text, AI NPC dialogue, agent collaboration, and tool-result streams.

Runtime Control Profiles

For cancellation, deadlines, partial results, backpressure, runtime objects, cache references, route hints, and trace context.

Reading path

1. Start with Tensor Profile Overview, then read its Descriptor Common Header / Schema and Body / Payload Frame child pages.
2. Continue with Token Profile Overview in the same order.
3. Read Runtime Control Profiles before implementing preview4 control, runtime object, cache reference, or wire conformance behavior. Its child pages split the frozen value registries, control-frame metadata, and object/cache metadata.
4. For cross-profile descriptor fields, read Typed Payload Descriptor first.
5. When you need to decide whether a field belongs to the public layer, cross-check Common Header or the version-frozen preview pages.