Introduction and Goals

Requirements Overview

diag-agent is an intelligent diagram generation tool that assists LLMs and developers in creating high-quality architecture diagrams. It addresses the common problems of syntax errors and poor layout in LLM-generated diagrams through an autonomous feedback loop.

Key Features

  • Autonomous diagram generation with syntax validation and design feedback

  • Multi-format support via Kroki integration (PlantUML, C4, BPMN, Mermaid, etc.)

  • Flexible deployment: CLI tool, MCP server (local/remote), or Docker container

  • Privacy-first: Local-first approach with optional remote rendering

  • LLM-agnostic: Works with any LLM via LiteLLM abstraction

  • Context-efficient: Safes main context by working in agent context. Additional minimal context consumption through help system and reference URLs

Quality Goals (Overview)

Die vollständigen Qualitätsziele stehen in Kapitel 10 (Quality Goals). Kurzüberblick:

  • Kontext-Effizienz

  • Einfache Installation

  • Privacy & Security

  • Autonomie

  • Erweiterbarkeit

Stakeholders

Role Expectation Concern

Software Architects

Generate C4 and other architecture diagrams quickly

Diagram quality, AsciiDoc integration

Developers

Simple CLI tool for documentation

Easy installation, good defaults

LLM Applications

Delegate diagram generation without context overhead

Clear interface, reliable results