SITUROOM

The zero-backend global situation room — real-time earthquakes, conflicts, hazards and markets in one static page. No server. No API keys. No paywall.

Live demo: https://nguyenminhduc9988.github.io/situroom/

Why

Most "world monitor" intel dashboards are hosted products: a big server fleet renders the picture, the scoring is a black box, and the useful parts (APIs, agent access) sit behind a paid key. SITUROOM is the opposite bet — everything a situation dashboard needs is available from free, keyless, CORS-friendly public APIs, so the entire product can be one static folder.

Six free sources today, PRs for more welcome.

Related MCP server: publicfinance

Quick start

1. Just open it — https://nguyenminhduc9988.github.io/situroom/

2. Run it locally (any static server works; no install, no build):

git clone https://github.com/nguyenminhduc9988/situroom
cd situroom
python3 -m http.server 8080
# open http://localhost:8080

3. Host your own — fork the repo, enable GitHub Pages (deploy from main, root). Done: it's static files.

Features

• Canvas world map — vendored d3-geo projection, graticule, country hover, magnitude-scaled quake markers, category-colored natural events, tension heat blobs. Toggleable layers, pulse animations (disabled under prefers-reduced-motion).

• Tension index — transparent per-country score over the current news window, top-10 bars colored green→amber→red. The formula is documented below and in the UI ("formula?" button).

  

• Headlines — conflict-flavored world news from GDELT (English sources by default, all-languages toggle in settings).

• Quakes / Events / Tech panels — USGS earthquakes, NASA EONET natural events, Hacker News front page.

• Markets ticker — BTC/ETH/SOL with 24h change, USD FX reference rates.

• Status chips per source — LIVE / STALE (cached) / SNAPSHOT / ERROR, so you always know what you're looking at.

• Watchlist — pin countries you care about to the top of the tension board.

• Responsive — panels stack on narrow viewports; the six status chips collapse into one aggregate chip.

Data sources

All free, keyless, fetched directly from your browser (CORS-friendly). Attribution and thanks:

Every source degrades independently: live fetch → last good response from localStorage (STALE) → committed capture in data/snapshot.json (SNAPSHOT) → ERROR chip. The dashboard never renders blank.

The tension index formula

No black box — the entire computation runs in your browser over the raw GDELT article list, in js/score.js:

• volume(C) — articles published in country C or naming C (or a common alias) in the headline

• severity(C) — mean headline keyword weight: 3 war/strike/attack/missile…, 2 sanctions/military/nuclear…, 1 protest/unrest/clashes…, floor 0.5

• score(C) = 100 × norm(0.6·log1p(volume) + 0.4·severity) — the hottest country of the moment reads 100, everything else is relative

Honesty note: GDELT's article-list API doesn't expose its per-article tone field, so severity is a documented keyword proxy — cruder than GDELT tone, but fully auditable, which is the point. The full country/alias gazetteer and keyword table are plain data at the top of the file; the unit tests in test/score.test.js pin the behavior.

AI sitrep (optional, bring your own key)

Off by default and never required. If you point settings (gear icon) at any OpenAI-compatible endpoint — including a local Ollama at http://localhost:11434/v1 — an AI SITREP button appears that summarizes the current headlines + tension board into a 5-bullet brief. Your endpoint, key and model live only in your browser's localStorage and are sent nowhere except the endpoint you configured.

MCP server (agents get the feeds for free)

mcp/situroom_mcp.py is a zero-dependency Python 3.10+ stdio MCP server exposing the same feeds as tools: get_quakes, get_events, get_headlines, get_tension, get_markets — normalized to the web app's shapes, tension formula faithfully ported from js/score.js. Config snippets for Claude Desktop/Code and Cursor in mcp/README.md.

claude mcp add situroom -- python3 /path/to/situroom/mcp/situroom_mcp.py

Architecture

Vanilla ES modules. No framework, no build step, no npm install. The only third-party code is vendored and committed: d3 v7 + topojson-client (plus the world-atlas 110m topology) in vendor/, so the app runs with zero network access to any CDN.

index.html
├── css/app.css           dark ops theme (system fonts only)
├── js/app.js             boot, source registry, 60s scheduler, panels
│   ├── js/map.js         canvas map (offscreen-cached base + marker layer)
│   ├── js/score.js       tension index (pure, documented, tested)
│   ├── js/store.js       localStorage cache + settings
│   ├── js/brief.js       optional BYOK sitrep
│   └── js/sources/*.js   one module per feed: fetchX() + pure normalizeX()
├── sw.js                 PWA: precached shell, network-first API fallback
└── data/snapshot.json    committed real capture — the floor of the chain:

      live fetch ──fail──▶ localStorage last-good ──miss──▶ committed snapshot ──miss──▶ ERROR chip
        (LIVE)                    (STALE)                        (SNAPSHOT)

Run the tests (pure functions only, fully offline):

node --test test/*.test.js    # or: npm test — 44 tests, zero dependencies

Contributing

PRs welcome — especially new free, keyless, CORS-friendly data sources. The bar for a source module: one file in js/sources/, export a pure normalizeX() (unit-tested against a captured payload in data/snapshot.json) and a fetchX() that returns null on any failure. Open an issue first for anything bigger than a source or a fix.

License

MIT © 2026 Minh-Duc Nguyen

Built by BigWinner.work — AI Tools & Consulting