Architecture

Everything is a plugin.

A lightweight runtime discovers plugins at startup and wires them through typed extension points. An edition is nothing more than the framework plus the plugins you install — so you add, remove, or swap capabilities without forking the app.

The framework

roadmap-framework

A thin runtime: plugin discovery and topo-sorted startup, an extension + service registry, config with encrypted secrets, and an MCP server. It ships no features of its own — every capability arrives as a plugin.

  • Runtime discovery — no rebuild to add a plugin
  • Typed extension points, additive or swappable
  • Backend Python + optional Angular frontend per plugin
  • A shared @roadmap/ui design system for consistent UI
Installed plugins register into the framework
Contexts
roadmap-plugin-context
Context Assistant
roadmap-plugin-context-assistant
Company Brain
roadmap-plugin-brain
Jira / Confluence
roadmap-plugin-jira
AI Providers
roadmap-plugin-ai
Whisper (voice)
roadmap-plugin-whisper
Web Testing
roadmap-plugin-e2e
Selenium Driver
roadmap-plugin-e2e-selenium
AI Test Design
roadmap-plugin-e2e-assistant
File Viewer
roadmap-plugin-fileviewer
PlantUML
roadmap-plugin-viewer-plantuml
Markdown
roadmap-plugin-viewer-markdown
Image
roadmap-plugin-viewer-image
Scratch Dir
roadmap-plugin-scratchdir
Folder Picker
roadmap-plugin-folderpicker
Settings / Admin
roadmap-plugin-settings

Each plugin declares its dependencies and the extension points it fills — so a pattern is just a resolved set of these tiles, pre-configured for a job.

How it works

From framework to AI-ready in three steps

1

Start from the framework

Download the bundle and launch. You get the framework plus a core set of plugins — no database, nothing to provision.

2

Apply a pattern

Install the plugins a pattern needs from the in-app registry, then apply its configuration — sources, AI providers, task prompts.

3

Work, then serve over MCP

Compose contexts and generate artifacts, and expose every context to Claude, Cursor or your IDE with one MCP tool call.

Pattern language

Each way of working, named

A pattern names a recurring way of working with AI — its context, problem and solution — and lists the plugins and configuration that realize it.

A quick tour: compose a context from plugins, preview it inline, and serve it to your AI over MCP.