Sloth Brief: an AI news agent that reads overnight so I can be lazy in the morning

Here is my own account of the story based on the provided source material:

An AI news agent called Sloth Brief has been created to allow users to stay updated with AI and technology news without having to actively read through numerous sources each morning. The tool fetches 9 RSS feeds, filters to the overnight window, removes duplicates and presents the latest items first. It then uses an AI model called Mastra running locally via Ollama to edit the fetched items.

The editor picks the top 3 stories, clusters the remaining items by theme and summarizes them. The final brief is generated as a structured Markdown document called briefs/brief-YYYY-MM-DD.md. The entire workflow runs on a laptop computer.

Key features include fetching overnight items from multiple sources, filtering to the last 12 hours, deduplicating items, summarizing stories, and outputting a concise Markdown brief. The system runs on open-source components with no API costs, using the Ollama platform to host the Gemma 3 model. The architecture separates the fetching and filtering step from the editing and brief generation step to avoid the AI model processing URLs and layout details which it handles poorly.

Validation checks ensure links are correct, stories are not duplicated, titles match their items, and summaries don't make unsupported claims.

The project emphasizes open-source principles - no subscription fees, private reading data, and the ability to swap out models without vendor lock-in. It aims to provide an efficient way to stay informed on AI news without requiring significant time investment in reading. Users simply run the tool in the morning and end up with a brief that highlights the most important stories to read further.