Fireworks Tech Graph — Generate Technical Diagrams from Natural Language

Introduction

Fireworks Tech Graph converts plain-text descriptions of systems and workflows into clean technical diagrams. You describe the architecture in natural language and receive production-ready SVG or PNG output, removing the need to manually position boxes and arrows in a GUI diagramming tool.

What Fireworks Tech Graph Does

• Parses natural language descriptions and produces technical diagrams in SVG or PNG format
• Supports 7 visual styles: minimal, blueprint, whiteboard, dark, corporate, sketch, and neon
• Generates UML diagram types including sequence, class, component, and activity diagrams
• Renders AI/agent workflow patterns with specialized shapes for models, tools, and decision points
• Accepts iterative refinement commands to adjust layout and labels without regenerating

Architecture Overview

The tool uses a two-stage pipeline. The first stage parses natural language to extract entities and relationships into a structured graph. The second stage applies the selected visual style and computes node positions using force-directed or hierarchical layout. SVG is generated directly; PNG goes through a headless renderer.

Self-Hosting & Configuration

• Install via pip install fireworks-tech-graph or use the Docker image
• Configure default style and output format in ~/.fireworks-graph/config.yml
• Set FTG_OUTPUT_DIR to control where diagrams are saved
• Custom color palettes defined in YAML can be referenced by name with --style
• Integrates with CI pipelines to auto-generate diagrams from description files

Key Features

• Produces clean, deterministic SVG suitable for version control alongside code
• Seven visual styles match output to different team and presentation contexts
• Specialized AI/agent workflow support including tool-call loops and multi-agent topologies
• Iterative refinement adjusts individual elements without full regeneration
• CLI-first design makes it scriptable for documentation automation

Comparison with Similar Tools

• Mermaid — requires learning a text-based DSL; Fireworks Tech Graph accepts plain natural language
• PlantUML — established UML tool with its own markup; Fireworks Tech Graph removes the need for a formal syntax
• D2 — declarative diagram scripting; Fireworks Tech Graph targets users who prefer English prose descriptions
• Excalidraw — freehand whiteboard tool; Fireworks Tech Graph automates layout from text input

FAQ

Q: Does the tool require an internet connection? A: The layout engine and style rendering work offline. The natural language parsing stage may use a local or remote LLM depending on your configuration.

Q: How does it handle large diagrams with many nodes? A: The layout engine switches to a hierarchical algorithm for graphs above 30 nodes and supports pagination for very large diagrams.

Q: Can I export the intermediate graph for use in other tools? A: Yes. Pass --export-json to output the parsed graph structure as JSON for programmatic use.

Sources

• https://github.com/yizhiyanhua-ai/fireworks-tech-graph
• https://fireworks-tech-graph.readthedocs.io/