Skip to content

Features

Diagramm-Generator covers the complete workflow from entering a requirement through generation and refinement to export. Operation takes place via a chat interface, a data panel, and a code editor. Quality assurance functions — pre-validation, LLM-supported correction, and template fallback — run implicitly and are not part of manual operation.

Application scenarios

  • Documentation of support processes — From a description such as "support process with 1st level, 2nd level, and specialist departments", the application produces a swimlane diagram with the typical activities per lane (ticket intake, initial analysis, technical analysis, specialist processing) and the transitions between lanes.
  • Visualization of recruitment and approval processes — For HR processes, default lanes (applicants, HR team, specialist department, management) and matching activities are available. The result can subsequently be adjusted via chat.
  • Project planning with Gantt diagrams — Tabular input with tasks, durations, and dependencies produces Gantt diagrams including phases, milestones, and progress indicators.
  • Creating org charts from hierarchies — Indented lists or JSON trees are automatically converted into hierarchical diagrams suitable for team or department structures.
  • Architecture and sequence diagrams for IT documentation — Descriptions of system components and interfaces are translated into architecture or sequence diagrams, including typical request-response patterns.
  • Data models and ER diagrams — Tabular and relationship descriptions are processed into entity-relationship diagrams, including cardinalities and attributes.

At a glance

  • Input connectors for file upload (CSV, JSON, TXT) and free text input.
  • Language-driven generation via chat with downstream refinement in the conversation.
  • Template library with templates for business, technical, and project scenarios.
  • Multi-stage validation with automatic code correction and template fallback.
  • Export to SVG, PNG, and source code, plus batch export as ZIP.
  • Session persistence with history, versioning, and export/import.
  • Code editor with syntax highlighting, render button, and undo stack.

Input and data formats

The application accepts both free text via chat and structured data in the data panel. The Smart Convert mechanism detects the format automatically and transforms it into a form suitable for diagram generation.

  • CSV — Tabular data with column headers, suitable for relationship and task lists. Processing is handled via pandas.
  • JSON — Structured and nested data, suitable for hierarchical diagrams and configuration data.
  • Indented lists — Plain-text hierarchies via indentation, suitable for org charts and mindmaps.
  • Markdown lists — Bullet lists with sublists, suitable for the rapid capture of structures.
  • Tables and free text — Tabular structures are detected and converted; unstructured text is interpreted by the chat agent.

Uploaded files can be ingested directly from CSV, JSON, or TXT. Maximum input size is configurable.

Diagram types

The application covers the following diagram types. Selection is either explicit via the chat or implicit from the recognized intent.

  • Flowcharts and decision trees
  • Mindmaps
  • Org charts
  • Sequence diagrams
  • Gantt diagrams and roadmaps
  • Swimlane diagrams with lanes and inter-lane transitions
  • Kanban boards with columns
  • State diagrams and user journeys
  • Entity-relationship diagrams
  • Network graphs and architecture diagrams

Connectors and external components

  • Mermaid CLI (mmdc) — External renderer for Mermaid diagrams. The CLI is invoked as a subprocess and produces SVG output via a headless Chromium. A configuration file controls theme and background.
  • N2G (Need To Graph) — External library for generating draw.io XML from node-edge descriptions, used for network and infrastructure diagrams.
  • python-gantt — External library for generating Gantt diagrams with tasks, resources, dependencies, and progress tracking.
  • OpenAI-compatible LLM interface — External connection to a language model service via the OpenAI API convention. The application is agnostic to specific models and supports both external cloud APIs and locally operated services.
  • Internal LLM service — By default, an internally provided LLM endpoint is addressed. The connection uses the OpenAI-compatible API; the endpoint details are set via configuration.
  • cairosvg and Pillow — Libraries for SVG-to-PNG conversion in the export pipeline and for thumbnail generation.

Quality assurance functions

The application contains several mechanisms that ensure correctness and reproducibility of the results.

  • Pre-validation — Before the actual rendering, the Mermaid code is checked for known error patterns (embedded keywords, duplicate diagram declarations, invalid node IDs) and corrected automatically.
  • LLM-supported code correction — If rendering fails, a validation agent corrects the code with reference to a complete Mermaid syntax reference. The number of correction attempts is configurable.
  • Template fallback — If all correction attempts fail, a prefabricated template for the respective diagram type takes effect, so that the application always delivers a renderable result.
  • JSON intermediate format — For structured diagram types (flowchart, mindmap, org chart), the LLM optionally first produces a JSON intermediate that is then converted to Mermaid code in a rule-based manner. This reduces syntax errors.
  • Deterministic plan-to-code conversion — In the IPE architecture, the Mermaid code is generated from the structured plan without an LLM call. Identical plans produce identical code.
  • Confidence-based routing — Intent recognition combines rule-based and LLM-based analysis. On high confidence from the quick analysis, the LLM stage is skipped; on low confidence, the LLM is consulted.
  • Diagram versioning and traceability — Generated diagrams receive unique IDs, version numbers, and parent relationships, so that modifications remain traceable.

Operation and additional functions

  • Chat-driven generation and modification — A chat interface accepts natural-language descriptions, takes the conversation history into account, and distinguishes between creation, modification, extension, and analysis of existing diagrams.
  • Code editor with direct render — Generated code can be edited manually and re-rendered. An undo history and syntax highlighting are included.
  • Gallery and history — Multiple generated suggestions are displayed in parallel in a gallery; a history with thumbnails allows reverting to earlier versions.
  • Export — Single and batch export to SVG, PNG, and source code. Batch exports are delivered as a ZIP archive.
  • Session persistence — Session data can be exported and re-imported elsewhere.