Pollar does not do original journalism. We gather reporting already published, summarise it with AI, and run automated and human checks on the result. This page covers the models, prompts, and oversight.
Our approach
Every event you read is AI-built from several original sources, then edited for clarity and brevity, and published in English, Polish, German and French. The summary and translation steps run on cloud language models, processing only the published article text from publishers we cite. No reader data, identifiers, preferences or queries ever enter that pipeline. We pick models for quality and efficiency, not novelty.
People & the planet
AI does the heavy lifting: drafts, summaries on every story. Automated checks catch weak or inconsistent output. Selected content gets editorial review. We set editorial direction and decide which sources to trust. We don't pretend a human reads every line. See a mistake? Report it at contact@pollar.news.
Efficiency is a design goal, not a measured number. We summarise existing reporting instead of generating endless new content. We never train our own energy-heavy models. We keep a light, mostly self-hosted infrastructure. We prefer efficient models, not flashy ones.
Sources
Every event lists its original articles at the bottom of the page, each linked to its publisher. The headline image is credited to its source. AI can still get things wrong. When a detail matters, go to the source.
Personalisation & your preferences
If you describe in Settings what you want to read, AI interprets the text into feed rules: boost, mute, skip. This runs on Google Vertex AI Gemini in europe-west4 (Google Cloud, Netherlands). Your text stays in the EU. Vertex AI does not use prompts to train Google's models. You can revoke consent and stop the processing from Settings, in-app, at any time.
Where your data goes, and doesn't
Reading an article sends nothing about you to any AI provider. The app fetches the prebuilt summary from our servers. That's it. Asking a question about an article runs on-device through Apple Intelligence. Your question never leaves your iPhone. Feed personalisation is the one feature that uses your data: if you opt in, your free-text preferences go to Google Vertex AI Gemini in europe-west4 (Google Cloud, Netherlands) to be turned into feed rules. We ask for explicit consent the first time, you can revoke any time in Settings, and the underlying news pipeline stays untouched. Apple's on-device features run locally with no third party involved.
Two Pollar News features send your input outside the European Union: Pollar Chat messages and search queries. To provide the best answer quality, chat messages are processed by a US-based AI provider (OpenAI, via OpenRouter) as the primary model, with an EU provider (Google Vertex AI, europe-west4) as fallback. When you search, your query is converted into a numeric representation and the results are ranked by US-based AI providers (OpenAI and Cohere, via OpenRouter) to find matching articles; no account or device identifiers are attached to it. Chat inputs and search queries are processed only to provide those features and are not used to train any model. All other Pollar News features keep your data within the EU. You can withdraw chat consent at any time in Settings.
Contact
Questions, corrections or ideas? Write to us at contact@pollar.news.