War News Creator - Local MySQL Starter

This project gives you a fully local starter stack:

• n8n automation
• FastAPI website/API
• MySQL database

Included workflow files:

• n8n-warnews-local.json (local ingest test workflow)
• n8n-warnews-15feeds.json (phase 2: resilient 15-feed workflow with validation/enrichment)
• n8n-ai-agent-starter.json (chat agent starter, optional)

1) Setup

1. Copy .env.example to .env.
2. Update secrets in .env.
3. Start services:
   • docker compose up --build

2) Open apps

• Website/API: http://localhost:8000
• n8n: http://localhost:5678

3) Import n8n local ingest workflow

1. Open n8n.
2. Go to Workflows -> Import from File.
3. Select n8n-warnews-local.json.
4. Run with Manual Trigger.

The workflow sends a test article to FastAPI using POST /ingest.

4) Verify article publishing

1. Visit http://localhost:8000/articles.
2. Open the new article detail page.

API notes

• Ingest endpoint: POST /ingest
• Header required: x-api-key: <INGEST_API_KEY>
• Health check: GET /health
• Admin review UI: GET /admin/review?admin_key=<ADMIN_API_KEY>
• Admin actions:
  • POST /admin/review/{article_id}/publish?admin_key=<ADMIN_API_KEY>
  • POST /admin/review/{article_id}/reject?admin_key=<ADMIN_API_KEY>

Editorial workflow (pro MVP)

1. n8n ingests incoming news to POST /ingest.
2. Each item is placed in pending editorial review with an auto confidence score.
3. Public pages show only published articles.
4. Open admin queue and publish/reject per item.

Phase 2 quality upgrades

• n8n pipeline now validates content quality before ingest.
• Each item includes:
  • region
  • event_key (for event clustering)
  • confidence_hint
• API computes:
  • source-agreement adjusted confidence
  • confidence band (high/medium/low)
  • source match count for same event
• Optional auto-publish for strong items:
  • set AUTO_PUBLISH_HIGH_CONFIDENCE=true in .env

Phase 3 claim extraction

• Each ingested article now stores claim blocks with:
  • claim text
  • citation URL
  • per-claim confidence
  • verdict (confirmed/provisional/disputed)
• Article detail page now shows citation blocks and confidence per claim.
• Extraction mode:
  • OpenRouter recommended: set OPENROUTER_API_KEY and model
  • fallback: OPENAI_API_KEY
  • if no key is set: heuristic fallback extraction
• Optional env:
  • OPENAI_EXTRACTION_MODEL=qwen/qwen3-coder:free
  • LLM_API_BASE=https://openrouter.ai/api/v1
  • OPENROUTER_API_KEY=...
  • OPENROUTER_SITE_URL=http://localhost:8000
  • OPENROUTER_APP_NAME=war-news-agent

Phase 4 contradiction detection

• Detects likely contradictions between claims in the same event_key.
• Flags conflicts when:
  • opposite polarity language is found (e.g., "confirmed" vs "denied")
  • key numeric mismatch appears in similar claim topics
• Stores conflict alerts in claim_conflicts.
• Admin conflict dashboard:
  • GET /admin/conflicts?admin_key=<ADMIN_API_KEY>

Phase 5 conflict actions + feedback loop

• Conflict actions available in dashboard:
  • resolved
  • false_positive
  • needs_investigation
• New endpoint:
  • POST /admin/conflicts/{conflict_id}/status?admin_key=<ADMIN_API_KEY>
• Feedback logic:
  • when needs_investigation is selected, linked article confidence is reduced
  • low-confidence published items are moved back to pending automatically

Phase 6 source reliability + analytics

• New reliability model per source (source_reliability) learns from conflict outcomes.
• Source reliability score updates on:
  • ingest volume
  • open conflicts
  • resolved conflicts
  • false positives
  • needs investigation
• New admin analytics page:
  • GET /admin/analytics?admin_key=<ADMIN_API_KEY>
  • shows conflict trend snapshot
  • region distribution
  • lowest reliability sources first
• Reliability repair endpoint:
  • POST /admin/analytics/rebuild-reliability?admin_key=<ADMIN_API_KEY>
  • recalculates all source reliability counters/scores from current database state

Deploy on Vercel

This repo now includes Vercel config:

• api/index.py (serverless entrypoint)
• vercel.json (route all requests to FastAPI app)

1) Prerequisites

• Host MySQL externally (Vercel has no local MySQL service).
• Keep n8n hosted separately (n8n Cloud / VPS / Railway) and point it to your Vercel URL.

2) Required Vercel env vars

Set these in Vercel project settings:

• DATABASE_URL (remote MySQL URL using mysql+pymysql://...)
• INGEST_API_KEY
• ADMIN_API_KEY
• AUTO_PUBLISH_HIGH_CONFIDENCE (true or false)
• OPENROUTER_API_KEY (if using model extraction)
• OPENAI_EXTRACTION_MODEL (default: qwen/qwen3-coder:free)
• LLM_API_BASE (https://openrouter.ai/api/v1)
• OPENROUTER_SITE_URL (your Vercel domain)
• OPENROUTER_APP_NAME (any app name)

3) Deploy commands

From project root:

1. npm i -g vercel (if needed)
2. vercel login
3. vercel
4. For production: vercel --prod

4) Post-deploy

• Test health: /health
• Update n8n POST /ingest URL to your Vercel domain
• Keep admin routes protected with admin_key

Add ADMIN_API_KEY in your .env, then restart:

• docker compose up -d --build

Next step

After this local test works, replace the Build Payload node in n8n with:

• RSS/News source nodes
• AI summarization node
• same POST /ingest node

Home Page

Articel

##Admin Panel

##Analytic

##n8n Workflow