Analyze Article
ℹ️ How to use — Analyze
- Paste any news article URL and pick an analysis type.
- Propaganda Analysis — scores the article 0–1 for propaganda techniques (loaded language, appeals to fear, etc.).
- Fact Check — extracts verifiable claims and gives each a verdict (true / false / unverifiable).
- Review — runs a quality gate to decide if the analysis is good enough to publish (approve / reject / hold).
- Wiki Update — generates a wikitext page summarising the analysis (requires a prior analysis run).
- 💡 Tip: Run Propaganda Analysis first, then Fact Check on the same URL. The system links them internally.
Articles
ℹ️ How to use — Articles
- Enter an article UUID (shown in analysis results or search hits) and click Load Article.
- Shows full metadata: title, source, propaganda score, detected techniques, claims, and linked wiki pages.
- 💡 Tip: After analysing an article, copy its ID from the raw JSON to view the full detail here.
Semantic Search
ℹ️ How to use — Search
- Search across all ingested articles using natural language queries.
- Hybrid (recommended) — combines semantic meaning + keyword matching for the best results.
- Semantic — finds conceptually similar content even without exact word matches. Try
economic inequalityto find articles about wages, poverty, wealth gaps. - Keyword — PostgreSQL full-text search for exact term matching. Best for specific names/places like
NetanyahuorSri Lanka. - 💡 Tip: Use natural phrases, not single words.
climate change policy impactworks better thanclimate. - ⚠️ Search only returns articles that have been ingested (via Analyze or source polling).
RSS Sources
ℹ️ How to use — Sources
- Register a news source by name and base URL. Add an RSS URL if the site has one.
- If no RSS URL is provided, the system scrapes the homepage for article links as a fallback.
- Poll fetches new articles from a source (via RSS or scraping) and adds them to the database.
- Polling discovers articles but does not auto-analyse them — go to Analyze to run analysis on discovered URLs.
- 💡 Tip: Try sites like
bbc.com(RSS) oraljazeera.com(scraper). JS-heavy sites like Reuters may return 0 results via scraping.
Register New Source
Registered Sources
Loading...
Poll Sources
Wiki Pages
ℹ️ How to use — Wiki Pages
- View generated wiki pages for analysed articles.
- Enter a page title (format:
Analysis:<short-id>) and click Load Page. - Wiki pages are created by running Wiki Update on the Analyze tab (requires a prior propaganda analysis).
- Pages start as draft → pass Review to be approved → then published.
Job Queue & Pipeline
ℹ️ How to use — Jobs & Pipeline
- The system continuously crawls registered sources every 5 minutes and auto-analyses discovered articles.
- Each article goes through the full pipeline:
Propaganda Analysis → Fact Check → Review → Wiki Update. - Summary shows counts by status. Recent Jobs shows the last 50 jobs with their status.
- Dead jobs failed too many times. Click Requeue to retry them.
- 💡 The pipeline runs automatically in the background — no manual intervention needed!
Pipeline Summary
Loading...
Recent Jobs
Loading...
Dead Jobs
Loading...
🤖 Agent Load & GPU Assignment
Last 24 hours · Auto-refreshes every 30s
Loading...
🔄 Active Pipeline Jobs
Loading...
LLM Token Usage & Cost Estimation
Total LLM Calls
—
Tokens In (Prompt)
—
Tokens Out (Completion)
—
Avg Latency
—
Articles / Hour
—
Tokens / Article
—
Avg Pipeline Time
—
Queued Now
—
Active Sources
—
💵 Anthropic API Cost Estimate
If this workload ran on Anthropic — "current" = same tokens, "projected" = scaled by inference speed (faster API → more articles processed):
Claude Sonnet 4
$3/M in · $15/M out · 8× speed
$—
current usage /month
$—
projected at full speed
Claude Haiku 3.5
$0.80/M in · $4/M out · 10× speed
$—
current usage /month
$—
projected at full speed
Claude Opus 4
$15/M in · $75/M out · 8× speed
$—
current usage /month
$—
projected at full speed
📊 Cloud Provider Cost Projections
"Current usage" = same token volume. "Projected" = scaled by speed multiplier (faster GPU → more articles → more tokens).
| Provider / Model | Speed | In Rate | Out Rate | Current Usage /mo | Projected /mo |
|---|
🤖 Usage by Agent
| Agent | Calls | Tokens In | Tokens Out | Total |
|---|
🧠 Usage by Model
| Model | Calls | Tokens In | Tokens Out | Avg Latency |
|---|
📈 Hourly Trend (Last 24h)
🆓 Current Runtime Cost (Local Ollama)
$0.00
You're running locally on your RTX 2060 — all inference is free!
🔬 Explore Graph
🏷️ Topics
🎯 Techniques Detected
🕸️ Article Network
🔵 Articles
🟢 Topics
🔴 Techniques
— semantic links
📊 Recent Articles — Graph Connections
| Article | Topics | Techniques | Claims | Links |
|---|
Product Crawler Dashboard
🔴 Live Crawling Activity
No active scrapes
🏪 Merchants
| Merchant | Products | w/ Price | Canonical | Last Scraped | Scraping | Visible | Actions |
|---|
➕ Add Merchant
📊 Recent Scrape Runs
| Merchant | Status | Pages | Found | New | Updated | Errors | Started | Duration |
|---|