Feed Organization
Feeds are organized into categories based on the active variant:Full Variant (Geopolitical)
~25 categories including:- Breaking News (Reuters, AP, BBC)
- Defense & Military (Defense News, Janes, Breaking Defense)
- Geopolitics (Foreign Policy, The Diplomat, Carnegie)
- Energy & Resources (Oil Price, Rigzone, S&P Global)
- Intelligence & Security (The Cipher Brief, War on the Rocks)
- Regional Focus (Middle East, Africa, Latin America, Asia)
Tech Variant
~20 categories including:- AI & ML (VentureBeat AI, MIT Tech Review)
- Startups & Funding (TechCrunch, VentureBeat)
- Cloud & Infrastructure (The Register, Data Center Dynamics)
- Cybersecurity (Krebs on Security, Dark Reading)
- Developer News (Hacker News, The Verge)
Finance Variant
~18 categories including:- Markets & Trading (Bloomberg, MarketWatch, CNBC)
- Crypto & Blockchain (CoinDesk, Decrypt, The Block)
- Central Banking (Central Banking, Macro Hive)
- Commodities (Kitco, Futures Magazine)
- Economic Policy (IMF, World Bank, WTO)
Reading News Items
Each news item displays:Item Components
Threat Badge
Color-coded severity indicator:
- 🔴 Critical - Major conflicts, terror attacks, disasters
- 🟠 High - Significant geopolitical events
- 🟡 Medium - Notable developments
- 🟢 Low - Routine news
- ⚪ Info - Background information
Headline Title
Article headline with entity linking:
- Country names → clickable to country brief
- Leaders → entity profile
- Organizations → context info
Source & Metadata
- Source name with tier indicator (T1/T2/T3)
- Time ago (e.g., “2h ago”)
- Propaganda risk score (if applicable)
Threat Classification
Every news item passes through a 3-stage classification pipeline:1. Keyword Classifier (Instant)
- Pattern-matches against ~120 threat keywords
- Organized by severity tier (critical → high → medium → low → info)
- 14 event categories (conflict, protest, disaster, diplomatic, etc.)
- Word-boundary matching prevents false positives
2. Browser-Side ML (Async)
Transformers.js runs locally:- Named Entity Recognition (NER)
- Sentiment analysis
- Topic classification
- No server dependency
3. LLM Classifier (Batched Async)
- Headlines batched and sent to Groq/Ollama
- Results cached in Redis (24h TTL)
- Overrides keyword result if confidence is higher
- Exponential backoff on API errors
Users see keyword results instantly, with ML and LLM refinements arriving within seconds.
Clustering Mode
News items can be displayed in two modes:Flat List Mode
- All items shown chronologically
- No grouping
- Fastest render
- Best for scanning recent headlines
Clustered Mode (Default)
- Items grouped by topic/entity
- ML-based clustering runs in background
- Progressive enhancement - flat list appears first, clustering applies when ready
- Reduces visual clutter with related stories
Keyword Monitors
Create custom alerts for terms of interest:Creating Monitors
Word-Boundary Matching
Keywords use word boundaries to prevent false matches:
- ✅ “ai” matches “AI breakthrough”
- ❌ “ai” does NOT match “train” or “chairman”
Monitor Scope
Keyword monitors search across:- Headline titles - Primary text
- Descriptions - Article summaries (when available)
- Both fields - Union of matches
Managing Monitors
Editing Monitors
Editing Monitors
Click the monitor chip to edit keywords. Changes apply immediately.
Deleting Monitors
Deleting Monitors
Click the × button on a monitor chip to remove it.
Monitor Persistence
Monitor Persistence
Monitors are saved to localStorage and restored on reload.
Multi-Keyword Support
Multi-Keyword Support
Each monitor supports multiple keywords (comma-separated). An item matches if ANY keyword is found.
Source Quality Indicators
Source Tiers
Feeds are classified into 3 tiers:| Tier | Description | Examples |
|---|---|---|
| T1 | Tier 1 - Highest credibility | Reuters, AP, BBC, Bloomberg |
| T2 | Tier 2 - Established sources | Defense News, TechCrunch, The Diplomat |
| T3 | Tier 3 - Secondary sources | Regional outlets, specialized blogs |
Propaganda Risk Scores
Some sources receive propaganda risk labels:- State Media - Government-controlled outlets (RT, CGTN, Press TV)
- High Risk - Known for bias or misinformation
- Moderate Risk - Editorial bias but factual reporting
Related Assets
News items automatically detect nearby infrastructure:Asset Detection
When a news item is geo-located (74-hub strategic location database):- 600km radius scan - Identifies critical infrastructure
- Asset types - Pipelines, cables, datacenters, bases, nuclear facilities
- Distance ranking - Sorted by proximity
- Clickable links - Jump to asset on map
Proximity Context
Related assets provide instant geopolitical context:- Cable cut near a strategic chokepoint
- Protest near a nuclear facility
- Troop movements near a datacenter cluster
- Pipeline disruption affecting energy routes
Filtering & Search
Category Filtering
Each feed category has a dedicated sub-panel:- Click category tabs to switch between feeds
- Each category maintains separate scroll position
- New item badges per category
Time-Based Filtering
Filter news by recency:- Items outside the selected time range are hidden
- Coordinated with map time filter
- Useful for focusing on breaking developments
Source Type Filtering
Filter by source type:- All - Show all sources
- Tier 1 Only - Highest credibility only
- Exclude State Media - Hide propaganda risks
Feed Refresh
News feeds update automatically:Refresh Intervals
- Default: Every 5 minutes
- High-priority feeds (Breaking News): Every 2 minutes
- Low-priority feeds (Opinion, Analysis): Every 10 minutes
Circuit Breakers
If a feed fails repeatedly:- 5-minute cooldown - Prevents cascading failures
- Exponential backoff - Increases retry delay
- Error indicators - Shows feed status in UI
Manual Refresh
Force an immediate refresh:- Locate the refresh button in panel header
- Click to trigger immediate fetch
- Loading indicator appears during fetch
Advanced Features
Deviation Indicators
News panels show deviation levels when activity is abnormal:- 📊 Elevated - 1.5-2x normal volume
- 📈 High - 2-3x normal volume
- 🚨 Critical - 3x+ normal volume
Trending Keywords
A separate panel shows spiking terms:- 2-hour window vs 7-day baseline - Detects surges
- 3x spike threshold - Filters noise
- Source diversity - Requires 2+ feeds
- CVE/APT extraction - Identifies threat indicators
- Auto-summarization - AI context for spikes (rate-limited)
Entity Extraction
Headlines are analyzed for entities:- Countries - Auto-linked to country briefs
- Leaders - World leader profiles
- Organizations - UN, NATO, IMF, etc.
- Military entities - Specific units, equipment
Export & Sharing
Individual Items
- Copy link - Direct URL to article
- Share to social - Twitter, LinkedIn, WhatsApp, Telegram
- Email - Pre-formatted email with headline and link
Panel Summaries
Generate AI summaries of panel content:- Click ✨ sparkle button in panel header
- AI analyzes top 8 headlines
- Summary appears below header
- Cached for 10 minutes
Localized Feeds
Seven locales have native-language feed sets:- French - Le Monde, Jeune Afrique, France24
- Arabic - Al Jazeera Arabic, Al Arabiya
- German - DW German, Spiegel, FAZ
- Spanish - El País, BBC Mundo
- Turkish - BBC Türkçe, Hurriyet, DW Turkish
- Polish - TVN24, Polsat News, Rzeczpospolita
- Russian - BBC Russian, Meduza, Novaya Gazeta Europe
- Thai - Bangkok Post, Thai PBS
- Vietnamese - VnExpress, Tuoi Tre News
Next Steps
Video Streams
Configure live video channels
Command Palette
Quick search for news and entities
Map Controls
Visualize news on the map
Panels
Customize news panel layout