Autonomous AI Newsroom That Discovers And Publishes Stories

Autonomous AI news portal that discovers, verifies and publishes original multi-source articles - fully hands-off. Includes codebase, domain, infrastructure, admin dashboard and 50+ page ops documentation. ~$3/day to run.

The entire pipeline (discovery → fact-checking → multi-agent writing → publishing) runs on a cron schedule for ~$3/day. Includes full codebase, domain, admin dashboard, and 50+ pages of ops documentation.

What you're buying

A production-ready AI newsroom that runs itself. You point it at any topic, language or market - and it produces verified, original journalism without human intervention.

This isn't a ChatGPT wrapper that rewrites a single article. It's a complete editorial pipeline:

Discovery - The system continuously scans RSS feeds, trending topics and web sources to find what's newsworthy right now. A category rotation engine ensures balanced coverage. Breaking stories bypass the queue automatically.

Research & Verification - Every story must pass a two-source rule: at least 2 independent publisher domains confirm the facts before anything gets written. Sources go through a Link Quality Gate (redirect chains, consent walls, readability scoring, publisher reputation) and a Topic Relevance check. A domain cooldown system automatically sidelines sources that block scrapers.

Fact Extraction & Cross-referencing - Each verified source is independently analysed to extract facts, quotes and key claims. These fact sheets are then merged using intersection logic - only claims confirmed by multiple sources survive. If the merged fact sheet is too thin (< 4 facts or < 600 chars), the story is automatically rejected. No fluff gets through.

Multi-Agent Writing - Multiple AI "journalist agents" (each with a distinct voice, editorial style and system prompt) independently write their version of the story. Each version includes headlines, summaries, bullet-point highlights, and is reviewed by an AI editor. The best variant gets published - or you pick manually from the admin panel.

Visuals - Cover images are generated automatically using AI image models, with compression and Storage upload built in.

Deduplication (VESNA) - A dedup engine prevents the system from re-covering stories it already published, even when sources keep reporting on the same event.

The entire pipeline - from "trending topic spotted" to "article live on the website" - runs on a cron schedule with zero human input. Or use the admin dashboard for full manual control over every step.

What's included

• Full Next.js (App Router) + TypeScript codebase - clean, typed, well-structured
• Production domain: subjektiv.media
• Firebase backend (Firestore + Storage) - all schemas, security rules, indexes
• Docker + nginx setup for Hetzner VPS with blue/green deployment script
• Comprehensive admin dashboard: cycle management, agent configuration, category settings, ops monitoring, cost diagnostics, source probing tools
• 50+ pages of operations documentation (dev setup, architecture, runbooks, incident postmortems, environment variables reference)
• Seed scripts for demo content and agent configurations
• Multi-provider AI support (Gemini, Vertex AI, with fallback chains)
• Multi-provider web search (Brave, Mojeek, SerpAPI, Google CSE)
• Multi-provider scraping with proxy fallback
• Public API with caching, rate limiting and payload guards
• RSS integration for breaking news detection

Why this is valuable

1. It actually works in production. This isn't a weekend prototype - it survived a real billing incident, got hardened with guardrails, and has detailed postmortem documentation explaining exactly what went wrong and how it was fixed.

2. The hard problems are solved. Two-source verification, fact-sheet intersection, domain cooldown, AI cost control (per-cycle budget guards, token caps, concurrency limits), egress optimization, deduplication - these took months to build and debug.

3. It's language and market agnostic. Point it at Croatian politics, US tech news, sports, finance etc... - the discovery and writing pipeline adapts. The agents, categories and sources are all configurable through the admin UI.

4. Operating cost is ~$3-4/day. Hosting (Hetzner VPS) + AI API calls + Firebase. The system includes centralized cost monitoring, per-cycle budget guards, and automatic kill switches if spend spikes.

5. It's documented like enterprise software. Not just code comments - full architecture diagrams, state machine specs, threat models, SLO definitions, deploy procedures, and a troubleshooting runbook. A developer can get productive in a day.

Who is this for

• Media entrepreneurs who want to launch a news site in any niche without hiring a newsroom
• AI/automation builders who want a sophisticated, production-tested pipeline to learn from or extend
• Agencies looking for a white-label news engine for clients
• Content businesses that need high-volume, verified content generation
• Developers who want a serious Next.js + Firebase + AI reference architecture

Tech stack

• Frontend & Backend: Next.js 14 (App Router), TypeScript, React
• Database: Firebase Firestore
• File Storage: Firebase Storage
• AI: Gemini API / Vertex AI (configurable, multi-model)
• Search: Brave / Mojeek / SerpAPI / Google CSE
• Hosting: Docker + nginx on Hetzner VPS (blue/green deploy)
• Monitoring: Custom ops dashboard + GCP Monitoring integration

Extra Info

• Niches: News & Media, AI / Machine Learning, Content Automation
• Technologies: Next.js, TypeScript, Firebase, Docker, Gemini AI
• Skills Needed: JavaScript/TypeScript, basic DevOps (Docker, VPS)
• Monetization: Ads, Sponsored Content, Subscription, White-label licensing