icon of FutureSearch

FutureSearch

Forecasting platform that turns any question about the future into a probability, number, or date, backed by a public track record on real markets and benchmarks.

Community:

Product Overview

What is FutureSearch?

FutureSearch preview

FutureSearch is a forecasting engine built by Varuna AI (founded by former Metaculus and Google Waymo engineers) that answers open questions about future events with probabilities, numeric ranges, or dates rather than vague guesses. Every forecast is grounded in web research and a persistent 'world model' that compounds accuracy as more questions are asked within a domain. The platform proves its calibration publicly: it trades live on Kalshi, Polymarket, and tracks S&P 500 positions with every trade and the research behind it disclosed, and it currently ranks first among 199 bots in Metaculus's live FutureEval tournament. Beyond single-question forecasting, FutureSearch extends into structured research through its Agent Map tool, which deploys a dedicated research agent per row of a dataset to fill in missing columns, and it plugs into existing workflows via a Python SDK, REST API, and an MCP server for Claude Code.


Key Features

  • Probabilistic Forecasting

    Ask any question about a future event and get back a probability, number, or date with the underlying research and reasoning shown, from $0.15 to a few dollars per question depending on effort level.

  • Public Track Record

    Live trading positions on Kalshi, Polymarket, and simulated S&P 500 calls are published alongside every research report, including losing trades, and standings on independent benchmarks like ForecastBench and Metaculus FutureEval.

  • Agent Map Dataset Research

    Assigns a dedicated web research agent to each row of a spreadsheet or dataset to look up missing values, such as pricing tiers or job classifications, at a cost of roughly 1 to 11 cents per row.

  • Decision Forecasting

    Handles conditional queries in the form of 'if I do X, will I achieve Y', going beyond simple yes/no odds to support real decision-making scenarios.

  • Compounding World Model

    Maintains a persistent latent representation of the future that improves in accuracy as a user or team asks more forecasts within the same domain over time.

  • Developer Integrations

    Offers a Python SDK, API access, and an MCP connector for Claude Code, plus spreadsheet integrations with Google Sheets and Microsoft Excel for embedding forecasts into existing tools.


Use Cases

  • Investment Research : Analysts and quant teams use FutureSearch for quantitative screening, stock outlook estimates, and market-moving event forecasts like tariff impacts.
  • Startup and Market Analysis : Founders and product studios research industries and validate startup ideas using FutureSearch's deep-research capabilities.
  • Spreadsheet Data Enrichment : Operations and growth teams fill in missing dataset columns at scale, such as SaaS pricing lookups or vendor classification, without manual research.
  • Prediction Market Trading : Traders reference FutureSearch's published Kalshi and Polymarket positions and reasoning to inform their own bets.
  • Policy and Geopolitical Forecasting : Research organizations use FutureSearch forecasters to contribute to reports on AI progress and geopolitical timelines, as seen in the AI 2027 report.
  • Professional Decision Support : Law firms and research nonprofits apply calibrated probability estimates to strategic planning and risk assessment.

FAQs

FutureSearch Alternatives

πŸš€

Analytics of FutureSearch Website