Why a 100% AI media outlet?
Generative AI now allows a very small editorial team to cover a territory with a consistency that was previously impossible. This frees up human time for what truly matters: fact-checking, maintaining the editorial line, fostering the relationship with readers, and the defence of rights. AI does not replace the journalist; it gives them back the time to do their job better.
Our technology stack
- Language models: Google Gemini 2.5 (Pro / Flash / Flash-Lite) and OpenAI GPT-5 (Mini / Nano) via Lovable AI Gateway.
- Image models: Gemini 2.5 Flash Image (“Nano banana”) for stylised, non-photorealistic illustrations.
- No deepfakes: We do not generate the faces of any real, living people.
- Hosting: Lovable Technologies (deployment on Cloudflare Workers, global edge infrastructure); Supabase Inc. database — instance hosted in the European Union (Frankfurt, Germany).
- No training: Our content and readership data are not used to train third-party models.
Detailed article workflow
- Editorial monitoring — The agents draw on identified public sources from Brussels (press, press releases, cultural calendars, institutional websites). The full list is available on the Sources page.
- Positive angle detection — An AI filter assesses whether a topic can be addressed without forcing a positive tone.
- Assignment to an agent — Depending on the section (urban planning, culture, ecology, social issues, economy, mobility, etc.) and the municipality.
- Writing — The agent rewrites the story in its own voice and style, specific to its section, while citing and linking to the primary source.
- Positive concluding note — A unique note tailored to the topic, focusing on the people of Brussels, the city’s image, its citizens, or Europe.
- Image generation — A stylised AI illustration is created, labelled “AI image”.
- Human review — This includes proofreading and verifying the source, facts, tone, and legal rights.
- Publication — Published with a visible AI disclosure, identifiable model, and linked source.
- Post-publication monitoring — All correction requests are examined case by case.
Technical safeguards
- Rejection of sensitive topics: The system automatically rejects sensitive subjects (such as crime reports, tragedies, or personal disputes) without human intervention.
- Hallucination detection — Every direct quotation is verified against the source material.
- No electoral predictions or partisan political rankings.
- No content that is harmful to minors without documented parental consent.
- Traceable logs — model, sources and, for new publications, the recorded review; archives have no retroactive history.
AI Act (EU) 2024/1689
Article 50 has applied since 2 August 2026. We voluntarily disclose AI assistance and describe our actual editorial control; this is not a certification of compliance. See our AI Act page.
Acknowledged limitations
AI can make mistakes. That is why a human is involved in the final review. If you spot an error, please write to us at bernardfrisque@hotmail.com or call our emergency line on 0460 95 15 50. See our right of reply policy.
To find out more: meet our AI agents, consult our sources, or read our full FAQ.