Kagi News follows a fully automated server-side pipeline that transforms raw RSS entries from hundreds of curated sources into polished, AI-summarized stories — all published in a single daily snapshot. The front-end you run locally (or visit at kite.kagi.com) simply fetches that already-processed data; no summarization or clustering happens in the browser.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/kagisearch/kite-public/llms.txt
Use this file to discover all available pages before exploring further.
This page describes the server-side processing pipeline. The open-source front-end only reads the output of this pipeline via the Kagi News API. You do not need to understand the pipeline to run or customize the front-end.
The Pipeline
RSS Feed Ingestion
Kagi News reads publicly available RSS feeds from two sources: core feeds (maintained by the Kagi team in
core_feeds.py) and community feeds (contributed by the community in kite_feeds.json). Feeds can be in any language — the pipeline handles translation automatically. Only publicly available RSS feeds are used; Kagi News does not scrape websites.Article Clustering
Articles published within the same time window that cover the same event are grouped into a cluster (also called a story). Clustering is based on semantic similarity so that multiple independent reports about one event are unified into a single story object rather than appearing as separate items. The
unique_domains field on a story records how many distinct publisher domains contributed articles to the cluster, and number_of_titles records how many individual article titles were considered — both are signals of how broadly an event was covered across the media landscape.AI Summarization
Each cluster is passed to a large language model. The model generates all AI-enriched fields on the story: a concise
short_summary, bullet-point talking_points, a representative quote, multiple perspectives from different sources, a timeline of events, historical_background, international_reactions, humanitarian_impact, economic_implications, future_outlook, and a suggested_qna section, among others. The language models powering this step include Anthropic Claude and Google Gemini (via @anthropic-ai/sdk, @google/genai, and @google-cloud/vertexai).Translation
The fully enriched stories are automatically translated into 30+ languages. Every text field in a story — title, summary, talking points, perspectives, and more — is translated so that any supported language displays a fully localized experience. The
sourceLanguage field records the original language of the source articles, while selectedLanguage records the language in which the story content was actually rendered for a given request.Batch Publication at 12PM UTC
Once processing is complete, all stories are packaged into a batch — a time-stamped snapshot of the entire news landscape across every category. A new batch is published once per day at 12 PM UTC. This single daily publish is an intentional design choice: it eliminates the compulsion to refresh constantly, keeps the news diet manageable, and ensures that every story is fully processed before it reaches readers.
What Is a Batch?
A batch is the atomic unit of news in Kagi News. It is a complete, immutable snapshot of all categories and their stories for one day. Batches are identified by a UUID (id) and carry metadata such as createdAt (ISO 8601 timestamp), the total number of categories, clusters, and articles it contains, and the Chaos Index score for that day.
Because batches are immutable, Kagi News supports time travel — the ability to browse the news as it appeared on any past day by requesting a historical batch ID.
What Is a Cluster (Story)?
A cluster is a group of articles from different publishers that all cover the same real-world event. From the reader’s perspective a cluster appears as a single story. Key evidence fields:| Field | Meaning |
|---|---|
unique_domains | Number of distinct publisher domains that contributed articles |
number_of_titles | Total article titles considered during clustering |
articles | The individual source articles included in the cluster |
domains | Metadata about each contributing publisher, including paywall status |
unique_domains count indicates a widely-reported event with diverse sourcing — an important signal of reliability.
The Chaos Index
Every daily batch includes a Chaos Index: a numeric score from 0 to 100 that measures the overall intensity of global news that day. A low score means a relatively calm news day; a high score indicates an unusually turbulent or event-heavy day. The batch also carries achaosDescription — a short human-readable summary explaining the score in context.
The Chaos Index is surfaced prominently in the UI so readers can calibrate their expectations before diving into the day’s stories.
On This Day
Alongside current news, each batch can include On This Day — a curated set of historical events, notable birthdays, and anniversaries that occurred on the same calendar date in past years. EachOnThisDayEvent carries a year, a content description, a sort_year for ordering, and a type of event, person, or people. An optional image URL may also be present.
AI-Enriched Story Fields
The server pipeline populates a rich set of AI-generated fields on every story. Not every field applies to every story type — fields that are not applicable are omitted (returned asnull or absent) in the API response.
| Field | Description |
|---|---|
short_summary | One-paragraph plain-language summary of the event |
talking_points | Bullet list of the most important facts |
perspectives | Array of viewpoints from different sources, each with attribution |
timeline | Chronological sequence of events with dates |
quote / quote_author / quote_attribution | Representative quotation and its source |
historical_background | Context from relevant past events |
international_reactions | How countries or international bodies have responded |
humanitarian_impact | Human cost or displacement, if applicable |
economic_implications | Market or macroeconomic effects |
future_outlook | Expected developments going forward |
suggested_qna | Question-and-answer pairs for deeper understanding |
did_you_know | A surprising or lesser-known fact about the topic |
geopolitical_context | Broader geopolitical framing |
key_players | Notable individuals or organizations involved |