NewYour Exa or Tako SDK, up to 5× cheaper.

News API

News API for AI agents.

Real-time news search over our own index: every result with its source, publication time and a calibrated probability that it answers your query. Short verbatim excerpts instead of full pages, so your agent reads less and cites more.

$5 free creditQuickstart

From $1.40 per 1,000 searches · $5 free credit, no card

  • 270+

    news sources

  • 130+

    countries

  • 30+

    languages

  • 66,000+

    articles indexed in the last 7 days

Our index on September 25, 2026. GET /v1/sources returns these numbers live, by country and language, and tells you whether a given domain is covered. We don’t publish the list of sources; every result shows its own.

One request

Ranked news in JSON, in about a second.

Plain JSON over HTTPS, from any language. Already use the Exa SDK? Point it at our compatibility layer and keep your code.

query
What you’re looking for, in any language. Up to five queries per request.
days · published_after
How fresh: the last N days, or since a date.
include_domains · sections
Only the outlets and sections you trust.
mode
ultra, fast, normal or deep: how much we read before we rank.
curl https://api.typesearch.ai/v1/search \
  -H "Authorization: Bearer $TYPESEARCH_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "query": "lithium royalties in Chile", "mode": "fast", "days": 7 }'

Built for agents, not for news readers.

  • Fresh by design

    We re-read news sitemaps and feeds on our own schedule, the busiest sources every five minutes. Every result carries its publication time, so your agent can filter by the hour.

  • A probability, not a rank

    Each result says how likely it is to be about your query: 0.94 means relevant, 0.51 means undecided. Threshold on it instead of guessing from position.

  • Read before ranked

    In normal and deep modes we open the top candidates and verify them. Headlines that oversell are dropped before they reach your context window.

  • Small enough for a context window

    Headline, the publisher’s description and short verbatim excerpts, with the same story from ten outlets collapsed into one result. Never full pages to trim.

  • Local and multilingual

    Regional and national press in 30+ languages, searched in the language it was written. Starting with Latin America.

  • Plugs into your stack

    Plain HTTP and OpenAPI, the Exa, Tako and Perplexity SDKs through our compatibility layers, and LangChain, the Vercel AI SDK, n8n or LlamaIndex.

By country

News API by country.

A page for each country where our coverage clears the bar: at least 5 sources based there that publish in its languages, and 1,000 articles from them in the last 7 days. For any other country, GET /v1/sources returns what we cover.

Measured in our index on September 25, 2026.

Pay per search, by how much it reads.

No monthly minimum and no seats. Cached repeats within 10 minutes are free, and so are failed requests.

Every new account gets $5 in free credit, valid for 30 days: about 3,571 fast searches.

$5 free credit
  • ultraHeadlines only. For classifying at volume.$1.40/ 1k searches
  • fastHeadlines and standfirsts. The default for agents in a loop.$1.40/ 1k searches
  • normalReads the top 4 results before ranking them.$2.20/ 1k searches
  • deepWidens the topic, reads 8, adds highlights and essentials.$5.60/ 1k searches

News API questions

Is there a free news API?

There’s no free tier, but every new account gets $5 in credit, valid for 30 days — about 3,571 fast searches — with no card needed. After that you pay per request, from $1.40 per 1,000 searches, with no monthly minimum.

How real-time is it?

We re-read each source’s news sitemap and feeds on a schedule set by how often it publishes: the busiest every five minutes. Every result has its publication time; filter with days or published_after to keep only the latest.

Which news sources do you cover?

270+ sources in 130+ countries and 30+ languages on September 25, 2026, with an emphasis on local and regional press. We don’t publish the list: GET /v1/sources returns coverage by country and language and tells you whether a given domain is covered. Every result shows its source and links to the original.

Does it return the full text of each article?

No. You get the headline, the publisher’s own description and short verbatim excerpts, with a link to the original. That keeps us fair to publishers and keeps your agent’s context small.

Is it a good news API for LLMs and RAG?

That’s what it’s for: compact results, a calibrated score to threshold on, and excerpts quoted verbatim, so your model grounds its answer in text it can cite. It never writes summaries of its own.

Can I use it from LangChain, the Vercel AI SDK or n8n?

Yes. See the integrations page: LangChain works today through the Exa integration pointed at typesearch, and the Vercel AI SDK, n8n and LlamaIndex call the HTTP API as a tool.

Keep exploring

Your agent reads the news. Make it cheap.

Get a key in a minute, with $5 in free credit. One request, and the results are in your context.

$5 free creditTalk to sales