Hacker News Search — momentum, not just all-time points

Search Hacker News through its public search API and get one row per story, with the fields you need to tell a post that is climbing right now from a post that was big in 2020.

What the point count hides

Search sorted by relevance puts old winners on top. The first result for rust is a 2020 story with 1,582 points. That number is real, and it is also useless if what you want is "what is getting traction today".

Every row therefore carries:

• pointsPerHour — points divided by hours since posting, but only for stories still inside the 48-hour scoring window. Past that, HN scores stop moving, and dividing a frozen score by an ever-growing age produces a number that shrinks forever. Older rows get null, not a small number. The run summary tells you how many rows were inside the window, so a single high-momentum row is never mistaken for the only story worth reading.
• commentsPerPoint — separates a link everyone upvoted and nobody discussed from a thread that argued for a day.
• domain — the publisher, already normalised (www. stripped), so you can group by source.

Fields

Field	Notes
storyId, hnUrl	The HN item and its discussion page
kind	story, show_hn, ask_hn, poll or job
title, url, domain	url and domain are null for Ask HN and text posts — that is "no external link", not a scrape failure
author	The HN handle only. No profile, no name, no email
points, comments, commentsPerPoint	As of the moment of the run
pointsPerHour, ageHours	pointsPerHour is null beyond 48 hours (see above)
createdAt	ISO 8601, UTC
sameUrlSubmissions, bestPointsForUrl, isBestForUrl	See the limit below

The one field with a hard limit

sameUrlSubmissions counts how many times the same link appears in this run's results, after stripping tracking parameters (utm_*, ref, fbclid). It is not how many times the link was ever submitted to Hacker News. Measured over 500 results per query on 2026-08-23, 8 rows matched for openai, 6 for chatgpt and 37 for hacker news, with a maximum of 4 submissions for any one link. Useful as a flag, not as a headline. If you need the true resubmission history of one URL, search for that URL as the query.

Input

{
  "query": "rust",
  "tags": "story",
  "sort": "date",
  "sinceDays": 7,
  "minPoints": 20,
  "maxItems": 200
}

• sort: "date" is what you want for monitoring. relevance favours well known posts.
• sinceDays matters more than it looks. Without it, search reaches back to 2006.
• minPoints and minComments are applied by the API before results are fetched, so filtered-out stories are never charged for.
• tags: "front_page" returns the front page at the moment the run starts. It is a snapshot; run it on a schedule if you want a history.
• Leave query empty only when you set a tags slice — an empty query with no tag is rejected rather than returning the whole site.

Output

One dataset row per story, plus a SEARCH_SUMMARY record with the total number of matches on Hacker News (which is almost always far larger than what you asked for), how many rows fell inside the scoring window, the median score, and the ten most frequent domains.

Pacing

Requests are serialised with a minimum interval. Please do not lower minIntervalMs. The search API is offered free and without an API key; keeping the load light is the reason it stays that way.

Unofficial. Not affiliated with Y Combinator or Hacker News.