An interactive, AI-powered release-notes explorer for ICM InfoWorks that transforms static software release notes into a searchable, visual, and conversational product history.
Repository: SWMMEnablement/ICMNewView
Replit app: replit.com/@robertdickinson/ICMNewView
ICM InfoWorks — What's New is a web application for exploring the evolution of ICM InfoWorks across nearly two decades of releases. Instead of reading static release-note documents, users can browse a visual timeline, search by keyword, filter by domain, compare upgrade paths between versions, inspect feature-evolution chains, and ask AI assistants questions about any release or feature.
The application spans 48 versions from v1.5 (February 2011) through v2027.0, with a total of 809 documented features stored in the application dataset. It is designed for engineers, product users, consultants, and technical teams who want to understand what changed in ICM InfoWorks and how product capabilities evolved over time. [page:41]
- Interactive release timeline with version cards and feature summaries. [page:41]
- Full-text search across feature titles, descriptions, and categories with inline highlighting. [page:41]
- 7 timeline filter categories and version-range filtering. [page:41]
- My Stack tracking to show what is new relative to a selected baseline version. [page:41]
- Analytics dashboard with bar charts and a full Category × Version heatmap. [page:41]
- Compare Versions workflow for upgrade-path analysis between any two releases. [page:41]
- Feature evolution chains detected automatically using Jaccard title similarity. [page:41]
- Deep-linkable features so each feature can be referenced by URL anchor. [page:41]
- Quad AI chat with four independent models and separate conversation histories. [page:41]
The application includes four built-in AI assistants, each exposed through a separate API endpoint and conversation tab. These are Claude Sonnet 4.5, DeepSeek Chat, Gemini 2.5 Flash, and GPT-4o Mini. [page:41]
Each AI response is instructed to include feature citations in the form [cite:featureId], which the frontend renders as clickable badges that jump directly to the referenced feature in the timeline. This makes the chat useful not just for summaries, but for traceable product-history exploration. [page:41]
The timeline is the main navigation surface of the application. It presents versions chronologically with feature cards, and each card can open a detail modal, pre-fill a chat question, or be targeted by a deep link such as /#feature-2027.0-network-design. [page:41]
Users can search across titles, descriptions, and category text. The filter system includes 7 timeline categories — Cloud/SaaS, Interface/UI, 2D Modelling, Data Import/Export, Automation/Scripting, Administration, and Analysis/Results — along with a version range selector and baseline “My Stack” version tracker stored in local storage. [page:41]
The Compare Versions dialog analyzes all features added between a selected “from” version and “to” version, not just the endpoint releases. It supports By Version, By Category, and Chains tabs, plus markdown export for structured upgrade summaries. [page:41]
The app detects related features across versions using Jaccard title similarity with a threshold of 0.5. Chains are classified into maturity states such as introduced, growing, mature-active, and stable, which helps users see how capabilities developed rather than viewing features as isolated one-off notes. [page:41]
The analytics section is centered on VersionCharts.tsx and includes bar-chart summaries plus an 11-row × 48-column heatmap. The heatmap is sorted by release date, supports hover tooltips, row normalization, multiple color schemes, numeric overlay toggles, and click-through navigation back into the timeline. [page:41]
The heatmap uses 11 analysis categories that are separate from the 7 timeline filter categories. These include Cloud/SaaS, 2D Modelling, UI/UX, Data Management, 1D Hydraulics, GIS Integration, Rainfall, Water Quality, Scripting/API, SWMM, and Analysis/Results. [page:41]
The frontend is built with React 18, TypeScript, and Vite, using TanStack Query for data fetching, Wouter for routing, Tailwind CSS for styling, and shadcn/ui plus Radix UI primitives for interface components. Core UI pieces include TimelineView, FeatureCard, FeatureDetailModal, ChatSidebar, FilterPanel, VersionCharts, CompareVersionsDialog, and EvolutionChainsDialog. [page:41]
The backend uses Express.js with TypeScript and serves version data plus four AI chat endpoints. The primary routes are GET /api/versions, GET /api/versions/:id, and four POST chat endpoints under /api/chat/ for Claude, DeepSeek, Gemini, and OpenAI-backed assistants. [page:41]
All version data is stored in memory, sourced from server/data/versions.ts, a TypeScript file of about 5,202 lines. There is no active database in current operation, although Drizzle ORM infrastructure and an IStorage abstraction exist for future migration. [page:41]
ICMNewView/
├── .agents/ # Agent-related configuration
├── attached_assets/ # Supporting content and source materials
├── client/ # React frontend
│ └── src/
│ ├── components/
│ ├── hooks/
│ ├── lib/
│ └── pages/
├── scripts/ # Version-data utility scripts
├── server/ # Express backend and version data
│ └── data/
│ └── versions.ts # 48 versions / 809 features
├── shared/ # Zod schemas and shared types
├── HANDOVER.md # Detailed project handover
├── IDEAS.md # Feature roadmap and ideas
├── replit.md # Replit notes
├── design_guidelines.md # UI/design guidance
├── package.json
├── vite.config.ts
├── tailwind.config.ts
└── .replit
The repository is primarily TypeScript (98.8%), with a small amount of CSS and other supporting files. [page:40]
GET /api/versions— Returns all versions sorted by version number descending. [page:41]GET /api/versions/:id— Returns a single version by ID. [page:41]
POST /api/chat/claude— Claude Sonnet 4.5 via Anthropic SDK. [page:41]POST /api/chat/deepseek— DeepSeek Chat via OpenRouter. [page:41]POST /api/chat/gemini— Gemini 2.5 Flash via Google GenAI SDK. [page:41]POST /api/chat/openai— GPT-4o Mini via OpenAI SDK. [page:41]
All chat endpoints inject the full version dataset into the system prompt along with citation instructions, so answers can refer directly back to individual features. [page:41]
Install dependencies and start the development environment:
npm install
npm run devThe application runs on port 5000, with the Express backend serving the frontend and Vite providing hot module reloading in development. The handover explicitly notes that server/vite.ts, vite.config.ts, drizzle.config.ts, and package.json should not be modified casually. [page:41]
The app expects AI integration credentials to be provided through Replit AI Integrations. These include API keys and base URLs for Anthropic, OpenRouter, Gemini, and OpenAI, plus SESSION_SECRET; KIMI_API_KEY is also listed but currently unused. [page:41]
A few implementation details are important for maintainers:
- Version IDs are the same as version strings, for example
id: "2027.0"andversion: "2027.0". [page:41] - Release dates do not always increase with version numbers, so heatmap ordering can differ from semantic version ordering. [page:41]
- The app uses two distinct category systems: 7 timeline filters and 11 heatmap categories. [page:41]
- Features matching “ruby” are exclusively mapped to the Scripting/API heatmap row through
EXCLUSIVE_OVERRIDES. [page:41] - All data is in memory, and each AI request includes the full 809-feature dataset in its system prompt. [page:41]
This project is especially useful for:
- InfoWorks ICM users planning upgrades.
- Engineers comparing feature growth across versions.
- Teams documenting platform evolution.
- Product analysts studying development focus areas.
- Anyone who prefers conversational exploration over static release-note PDFs.
The most complete technical reference is HANDOVER.md, which documents the component architecture, API routes, data model, category systems, AI integrations, cross-component state flow, quirks, deployment approach, and future considerations. IDEAS.md captures roadmap concepts for additional diagrams and exercises. [page:41]
The project is designed for Replit deployment. In production, Vite builds the frontend and Express serves the static assets, with publishing handled through Replit’s deployment workflow. [page:41]
Add the project license here if one is intended for public reuse.