8ce68fa779
- Engine now converts content_md to HTML at render time (was dumping raw markdown, causing articles to show literal #/**/- symbols and collapse into wall of text) - /api/publish accepts 'content' key and converts markdown→HTML for API consumers - Added md Jinja filter + md_to_html helper (markdown lib, extra+sane_lists) - orchestrator: log warning when falling back to 'thinking' field (CoT, not prose) - content_pipeline now generates formatted articles via LLM instead of raw scraped HTML
Auto Publisher — README
Autonomous AI Publishing System
A fully autonomous, self-hosted publishing platform that continuously discovers valuable topics, generates original useful content, and publishes it to a collection of evergreen authority websites.
Sites
| Site | URL | Status |
|---|---|---|
| AI | https://ai.thetempleofdoom.com | 🚀 |
| Tech | https://tech.thetempleofdoom.com | 🚀 |
| Science | https://science.thetempleofdoom.com | 🚀 |
| Crypto | https://crypto.thetempleofdoom.com | 🚀 |
| Linux | https://linux.thetempleofdoom.com | 🚀 |
| Gaming | https://gaming.thetempleofdoom.com | 🚀 |
| DIY | https://diy.thetempleofdoom.com | 🚀 |
| Guides | https://guides.thetempleofdoom.com | 🚀 |
Architecture
cron (6AM daily) → Trend Discovery → Topic Scoring → Research (ornith:latest)
→ Multi-Agent Writing → SEO → Site Builder → Deploy to Proxmox CTs
Quick Start
# Initialize database
python3 core/orchestrator.py init
# Discover trending topics
python3 core/orchestrator.py discover
# Run full pipeline (3 articles)
python3 core/orchestrator.py run --max 3
# Check status
python3 core/orchestrator.py status
# Admin dashboard
python3 dashboard/app.py
# → http://localhost:5106
Infrastructure
- Orchestrator: MacBook (cron via launchd)
- LLM Inference:
- MacBook: qwen3.5:4b (fast, cheap tasks)
- GamingPC RTX 3070: ornith:latest (quality writing/research)
- Sites: 8 Proxmox LXC containers (nginx on each)
- Tunnels: Cloudflare home tunnel (d2871458)
- Code: Gitea (10.30.20.149:3000)
- Analytics: Self-hosted, privacy-first
Directory Structure
auto-publisher/
├── core/ # Orchestrator, deploy scripts
├── services/ # Microservice APIs
├── sites/ # Per-vertical static sites
├── shared/ # CSS, JS, assets
├── dashboard/ # Admin control panel
├── analytics/ # Tracking & learning loop
├── cron/ # Scheduled jobs
└── docs/ # Architecture docs
Pipeline Steps
- Discover: Scans HN, Reddit, GitHub, arXiv for trending topics
- Score: LLM evaluates popularity, competition, freshness, evergreen value
- Assign: Routes topics to appropriate vertical site
- Research: Deep research via ornith:latest (GamingPC), builds knowledge package
- Write: Multi-agent pipeline (outline → draft → edit → SEO → fact-check)
- Build: Generates static HTML, RSS, sitemap, JSON-LD
- Deploy: Pushes to Proxmox CT, restarts nginx
- Learn: Nightly analytics feedback loop improves future topic selection
LLM Strategy
| Stage | Model | Host | Rationale |
|---|---|---|---|
| Topic scoring | qwen3.5:4b | MacBook | Fast pattern matching |
| Research | ornith:latest | GamingPC | Deep reasoning, accuracy |
| Writing | ornith:latest | GamingPC | Quality output |
| Editing | qwen3.5:4b | MacBook | Fast iteration |
| SEO | qwen3.5:4b | MacBook | Template-driven |
| Fact-check | ornith:latest | GamingPC | Critical accuracy |
Environment Variables
# Optional overrides
OLLAMA_MACBOOK=http://localhost:11434
OLLAMA_GAMINGPC=http://10.30.20.186:11434
DASHBOARD_SECRET=auto-publisher-2026
Monitoring
- Admin dashboard: http://localhost:5106
- Pipeline logs:
core/orchestrator.log - Analytics: per-CT at
:5199/a/stats