Norgitov Trends is a personal project by Muhammad Norgitov. Hacker News and Show HN contribute news and projects, GitHub repositories, Hugging Face models and Daily Papers, arXiv research, LessWrong essays and Lobsters development discussions. Hugging Face Blog, Google Research and Cloudflare Blog add model news, scientific results and engineering practice. Both language editions link to the same originals.

Ranking

Popularity is normalized within each source, then balanced with freshness and variety. Exact URL duplicates are combined. Trending covers 48 hours; other periods cover 1, 7, 30 and 365 days. History starts when a source is connected; full coverage is still accumulating.

Reading the metrics

Stars, points and comments are original source counters. Projects in daily, weekly, monthly and annual rankings use measured counter growth, and are excluded without comparable observations. Hugging Face downloads are a rolling counter, so its change is not a count of new downloads. Charts contain real saved measurements.

For you and bookmarks

Likes, opened stories and selected interests shape For you. Not interested removes a story from your personal feed. Bookmarks, history and preferences stay in this browser; export a file or copy your profile to transfer it. Clearing browser data removes your local profile. Reading does not require an account.

Shared reactions and archive

Like and Not interested totals are shared across the site. A vote is linked to a random browser identifier; these are device votes, not verified people. The server stores the reaction and a pseudonymous rate-limit key. The annual section can replay an immutable daily publication: up to the first 100 stories, preserving their order and metrics. This archive starts accumulating when the feature launches.

Summaries and the assistant

AI summaries use available source descriptions and abstracts. Pending translations show the original with a label. The assistant searches, explains and compares feed materials with citations. Item conversations attempt to read the public article; when unavailable, the answer uses only its description and says so. AI can make mistakes: check the original for details.

Models and comparisons

The model section combines the OpenRouter catalog and Hugging Face trends. Compare base input and output token prices and context windows, then estimate your workload cost. OpenRouter ordering reflects usage on that platform; Hugging Face downloads use a rolling 30-day counter. This is not an independent quality ranking. History starts with our first observations and gaps are never backfilled with invented values.

Chat privacy

Your question, previous messages and selected interests are sent to our server and Workers AI to prepare an answer. Conversation history stays in the current tab; the server stores usage counters and cached public articles. Daily request limits apply.

Updates

A background schedule checks sources even without visitors. The feed and bookmarks request updates every five minutes; stories, charts and models every two minutes and when returning to the tab. The model catalog is collected hourly. Daily archive publications preserve their original metrics. One source failure preserves existing records. The feed shows its last successful collection time. A saved snapshot is clearly labeled if the live server is unavailable.

Read your way

RSS · Weekly digest · Chrome extension

The unified feed is inspired by Denis Trends; data is collected from original sources.