Open-Source Semantic news API
Keyword search returns everything that says the word and misses everything that *means *it. Describe what you’re watching for in a sentence and Clair returns the news that means it. Scored, explainable, and yours to reproduce, audit, and tune.
example.py and its live output
Get started
Send a phrase. We embed it server-side and return the news that means it. No vectors, no SDK, no local model.
How it works
Pure vector math, deterministic by design. #
We split each article into smaller parts called chunks, and every chunk gets embedded, using bge-m3, an open model. We employ MaxSim algorithm to find the best-matching chunk. Even a long article that spends one paragraph squarely on your topic still reaches you. No LLM sits anywhere in this path, which is why the same article and the same query always produce the same score.
Every result carries its score and the exact chunk of text that produced it, not an opaque relevance ranking.
Deterministic by construction: the same article and the same query always produce the same score.
Matching is vector math, not generation. It can surface a weak match, but it can't invent an article or a quote.
Access
Call the API, read the algorithm behind it, ask for what’s missing. #
API
Query the news
Describe what you're watching for as a phrase, or send your own vector, and get back the articles that mean it, each with its score. Metered per request.
Explore the API → Open source
Check the math
Don't want to trust a black box? The scoring algorithm is public under Apache-2.0. Read exactly how a score is produced and reproduce it yourself.
View the source → Requests
Shape what's next
Need a source added, higher throughput, or an endpoint that isn't here yet? Tell us what you're working on. We'd love to help you build it.
Build on the world’s news. #
One endpoint, a curated set of trusted sources, and a score you can check. Explore the API, or read exactly how it works.