Agents
Understand an Agent as a declarative graph of Templates and data flows.
An Agent declares a computation graph made of Templates. It describes what operations participate and how their data flows, rather than being an imperative sequence of function calls.
An Agent can contain a simple pipeline, independent branches or a later merge:
flowchart LR
I[Input DataContainer] --> A[Template A]
A --> B[Template B]
B --> O[Output DataContainer]
Green-bordered nodes reference Sinapsis templates. Hover or focus them to fetch documentation.
View Mermaid source
flowchart LR
I[Input DataContainer] --> A[Template A]
A --> B[Template B]
B --> O[Output DataContainer]
The graph expresses which Templates are connected, where execution dependencies exist, and where branches split or join. A runtime decides how to schedule that declared graph; it might run locally or distribute operations without changing the graph's intended meaning.
Each normal Template processes a DataContainer. When data must cross a process or service boundary, Universal Data Transport handles its representation and exchange.
When to think in terms of Agents
Use the Agent model to reason about a whole workflow: which steps exist, which depend on others, and how their outputs reach the next step. Think in terms of Templates when reasoning about one individual transformation.
For the formal graph rules, explicit branching with DataSplit, fan-in with DataMerge, and runtime responsibilities, see Agent Model.
Published from Git revision 7b7ba25d627b