Core Concepts
Core Concepts
Section titled “Core Concepts”Separate the building blocks
Section titled “Separate the building blocks”| Piece | Controls |
|---|---|
| Agent | Role, prompt, model, loop, and output. |
| Tool | Callable capability and side effects. |
| Memory | State across turns. |
| Workflow | How agents run and combine. |
Configuration should make model, loop, tools, and output expectations visible. Hidden defaults make cost and behavior hard to predict.
Keep the pieces separate
Section titled “Keep the pieces separate”An Agent, a tool, memory, structured output, and a workflow fail in different ways. Debugging is easier when each piece can be tested alone.
| Piece | Controls |
|---|---|
| Agent | Role, prompt, model, loop, and output. |
| Tool | Callable capability and side effects. |
| Memory | State across turns or runs. |
| Structured output | Shape expected by downstream code. |
| Workflow | How agents run and combine. |
Make configuration readable
Section titled “Make configuration readable”A future reader should be able to see the model, loop limit, tool access, output expectation, and owner without chasing hidden defaults.