Backsearch

A point-in-time web search API from General Reasoning — let agents and models search and fetch the web as it existed on any past date, with no data leakage.

Paid·API·Search·AI Agents·Research·Developer Tools

Paid·API·Search·AI Agents·Research+1

What is Backsearch?

Backsearch is an API from General Reasoning (gr.inc) that lets agents and models search and fetch the web as it existed on any given date, rather than the live internet. Its search endpoint returns documents crawled on or before an as_of date, and its fetch endpoint retrieves article text as it appeared then, gating on crawl date rather than the article's stated publish date to prevent information leakage. Results are stable and replayable (identical queries return identical results indefinitely), with domain filtering and optional summarization.

It's built for backtesting AI forecasting, replaying trading strategies, and building reproducible RL environments. Pricing is $10 per 1,000 searches and $2 per 1,000 fetches against a prepaid balance; failed requests are free.

Features

Point-in-time search + fetch with an as_of date
Gates on crawl date, not stated publish date (no leakage)
Stable, replayable results for identical queries
Domain allow/block filtering + optional summarization
Function-calling ready for models like GPT-5.6 Sol

FAQs

Common questions about Backsearch

What is Backsearch for?

Giving AI agents point-in-time web access, so backtests, forecasting evals, and RL training environments can search the web as it was on a past date without leaking future information.

How much does it cost?

$10 per 1,000 searches and $2 per 1,000 fetches, charged against a prepaid balance. Failed requests are free.