Core concepts

This section explains how the system works and which problems each mechanism solves. Start with Architecture, then choose a topic. Look up operational commands and parameters in guides and reference pages.

Question	Topic	Scope
What do Desktop and Runtime do?	Architecture	Both
How does the desktop host coordinate execution, IDE, and automation?	Desktop architecture and data flow	Desktop
How does a request become a result?	Agent loop	Shared method, Runtime details
How do multiple entry points reuse execution?	Events and delivery	Runtime
How are conversations, memory, and identity separated?	Workspace, session, and identity	Both, separately described
How are durable facts retained and recalled?	Memory	Runtime with Desktop comparison
How are long histories and tool outputs handled?	Compression and output preservation	Runtime
Which extension mechanism fits?	Tools, skills, and plugins	Both, separate formats
How do permissions differ from isolation?	Security	Both
How does experience become a skill?	Evolution and evaluation	Runtime
How do Agents divide and summarize work?	Multi-Agent	Both, separate mechanisms

Diagram conventions

Architecture uses C4-style views with explicit levels, boundaries, and relations. Interactions use UML sequence diagrams; lifecycles use UML state diagrams. Flowcharts use terminals, processing rectangles, and decision diamonds with labeled branches. Data-flow diagrams use Yourdon / DeMarco notation: external rectangles, process circles, parallel-line data stores, and named directional flows.

Data-flow diagrams do not show execution order. Dependencies do not imply shared state. Optional conditions affecting usage are stated explicitly. See Documentation contribution for the complete conventions.

Click SVG figures to view them at full size.