Skip to content
← Explore the product
Point-in-time search API

Search the record as of a past day.

Give a historical question its own information boundary. Set a date, search the indexed record, and use the returned article text as evidence for your agent.

Start free ↗Read the API docs ↗
  1. 01Choose a UTC day
  2. 02Search the eligible record
  3. 03Read the indexed evidence

A date is an information boundary.

Point-in-time search asks what the indexed record contained by a chosen day. For standard search, as_of_date selects documents indexed by the end of that UTC day. Adding recency narrows the time window before the cutoff.

The boundary uses indexing time. A source's published_at describes its publication time, which can differ. An article published on August 18 but first indexed on August 21 is excluded from an August 20 search. Filtering today's results by publication date alone would miss that distinction.

Indexed by Aug 20Eligible evidenceWithin the selected search window
Aug 20UTC day ends
First indexed Aug 21ExcludedEven if published before the cutoff

One request. A historical cutoff.

This example searches a seven-day window ending on August 20, 2026. Remove as_of_date to use the live record with the same API.

cURL · POST /v1/searchHistorical search
curl --fail-with-body https://api.ariselabs-search-api.com/v1/search \
  -H "Authorization: Bearer $ARISELABS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "query": "nvidia blackwell supply",
  "mode": "standard",
  "k": 5,
  "as_of_date": "2026-08-20",
  "recency": "7d"
}'

Set ARISELABS_API_KEY to an API key from your console. Run this on your server or in a local terminal.

Know what the boundary covers.

Historical coverage begins in 2025 and is limited to the corpus we have indexed. The response provides indexed article text and source URLs; it is not a complete archive of every page on the web.

Use mode: "standard" for this workflow. Advanced search can include live web sources, and /v1/fetch reads a URL now. Following a historical result's URL today may also expose newer content. Give the agent the indexed text returned by search.

A cutoff constrains evidence eligibility. Save the request and response when exact replay matters: the index and ranking can evolve. The cutoff also does not remove information already learned by your model.

Use the same boundary across a workflow.

Apply the date to every search an agent makes during a historical task. Keep the question, source text, and resulting prediction together, then evaluate it against the later outcome. The backtesting guide walks through that sequence.

For parameter ranges, response fields, and credit costs, use the Search API reference and pricing page.