DataContainers and Packets
Understand DataContainers, typed DataPackets, payloads, and enrichment.
A DataContainer is the common data boundary between normal Templates and Agents. Instead of requiring a different interface for images, audio, text, files and structured results, Sinapsis sends typed packets within one container.
flowchart TD
DC[DataContainer]
DC --> I[ImagePacket]
DC --> A[AudioPacket]
DC --> T[TextPacket]
DC --> F[FilePacket]
DC --> D[DataFramePacket]
DC --> C[Custom packets]
I --> P[Image content and annotations]
Green-bordered nodes reference Sinapsis templates. Hover or focus them to fetch documentation.
View Mermaid source
flowchart TD
DC[DataContainer]
DC --> I[ImagePacket]
DC --> A[AudioPacket]
DC --> T[TextPacket]
DC --> F[FilePacket]
DC --> D[DataFramePacket]
DC --> C[Custom packets]
I --> P[Image content and annotations]
A DataPacket is a typed unit within the container. Its payload is the actual content, which can be an image, array, tensor, text, file or another supported type. Packets may also carry metadata and annotations.
Enrichment, not just replacement
A Template can return a container with its existing data plus new annotations or derived packets. For example, an image can gain object-detection bounding boxes and later be used for crops or embeddings without changing the universal DataContainer interface.
An empty DataContainer is also valid: a workflow may intentionally produce no packets or create an initially empty container for later processing.
The UDT concept explains how such data crosses boundaries; UDT architecture provides exchange details.
Published from Git revision 7b7ba25d627b