Glossary
| Term | Definition |
|---|---|
Agent |
Autonomous component that iterates independently using its own LLM client |
C4 Model |
Context, Containers, Components, Code - hierarchical architecture diagram approach |
Context Efficiency |
Minimizing token consumption in parent LLM conversations |
Feedback Loop |
Iterative process: generate → validate → analyze → refine |
Headless Mode |
Non-interactive operation suitable for CI/CD pipelines |
Kroki |
Unified API for creating diagrams from textual descriptions (PlantUML, Mermaid, etc.) |
Kroki Fat-JAR |
Self-contained Java archive (~80MB) running complete Kroki service locally |
LiteLLM |
Python library providing unified interface to 100+ LLM providers |
MCP (Model Context Protocol) |
Standard protocol for exposing tools to LLMs via HTTP/SSE |
Vision Model |
LLM capable of analyzing images (e.g., Claude Sonnet with vision, GPT-4V) |
PlantUML |
Text-based diagram syntax, particularly popular for UML and architecture diagrams |
Syntax Validation |
Checking if diagram source code is valid for the target renderer |
Design Analysis |
Vision-based evaluation of diagram layout, clarity, and conventions |
URL Reference |
Returning short URL instead of file path or base64 data to save context |
Remote Confirmation |
Interactive consent before sending data to remote services |
SSE (Server-Sent Events) |
HTTP protocol for server-to-client streaming, used in MCP |
Iteration Limit |
Maximum number of refinement cycles before returning current result |
Example Provider |
Component managing curated sample diagrams for each type |
Feedback
Was this page helpful?
Glad to hear it! Please tell us how we can improve.
Sorry to hear that. Please tell us how we can improve.