Futuresearch
FUTURESEARCH is an AI-powered tool designed specifically for geopolitical risk professionals. It enables users to produce complete risk ana...
Last verified:
What is Futuresearch?
FutureSearch is an AI‑powered platform that runs automated research and forecasting agents on entire datasets, treating each row like a mini‑research task. It is designed to help teams perform large‑scale, reproducible web research, ranking, classification, matching, deduplication, and forecasting directly from spreadsheets or via an API. The platform lets users describe what they want in plain English and then dispatches LLM agents to search the web, read pages, and synthesize structured results for every row in their data.
Key features include deep dataset‑scale research where an agent examines each row, automatic classification of entities by any written‑down criteria, scoring and ranking of rows to surface the best candidates, and automated matching and merging of records across different datasets. It also offers forecasting agents that can predict probabilities, future dates, or numeric values for each question in a dataset, backed by internal evaluation benchmarks to track accuracy. The API exposes all these operations—rank, classify, dedupe, merge, forecast, and agent‑map—as callable functions that take a DataFrame and a natural‑language instruction.
The tool is aimed at researchers, analysts, product teams, and startups who need to quickly screen, prioritize, and understand large sets of companies, projects, or questions without manual web‑scraping or one‑off analyses. It is especially useful for discovery workflows, due‑diligence pipelines, competitive intelligence, and quantitative screening, where traditional tools would require manual queries or brittle rule‑based systems. FutureSearch integrates into existing data workflows and can be added to existing Python stacks or spreadsheet‑adjacent tools that speak to APIs.
FutureSearch works by treating each row in a DataFrame as a research task, sending it to an LLM agent that performs live web searches, reads documents, and reason about the result. The platform supports multiple operations—research, classify, rank, dedupe, merge, forecast—each taking a natural‑language instruction and returning enriched or reordered data. Users upload or send a dataset, define the operation and the criterion in plain English, and get back a new DataFrame with extra columns, scores, labels, or forecasts computed for each row.
The platform also emphasizes transparency and reproducibility, letting teams re‑run research on the same dataset to see how results evolve or to compare different prompts. Each operation is billed per researcher, with cost scaling roughly with the complexity of the research needed for each row. The API and web app aim to keep the UX spreadsheet‑like, so analysts can iterate quickly over columns, filters, and scores without writing complex code.
Futuresearch pricing
Pricing model: Free
FutureSearch offers a free tier with $20 in credits to get started, with no credit card required. Users pay only for what they use, with costs scaling based on the complexity of the research needed for each row. The platform exposes several operations—research, classify, score/rank, matcher (dedupe and merge), and forecast—each priced at different per‑researcher rates, typically ranging from fractions of a cent to under a dollar per researcher depending on the operation. Paid usage is billed per researcher, and the web interface clearly shows how many credits remain and how much each operation will consume.
Futuresearch pros
- Runs automated web research at dataset scale
- Handles each row of a DataFrame as a separate research task
- Supports natural‑language instructions for research tasks
- Adds new columns of researched data to existing datasets
- Performs deep web reading and synthesis, not just keyword lookups
- Enables ranking and sorting of entire datasets by custom criteria
- Automatically classifies entities by user‑defined categories
- Identifies and deduplicates similar rows across datasets
- Matches and merges records across different tables or sources
- Generates calibrated forecasts for probabilities, dates, or numbers
- Scales forecasting to hundreds or thousands of questions at once
- Offers a well‑documented API for Python and DataFrame workflows
- Integrates with existing data pipelines and spreadsheets
- Provides reproducible research runs for audits and comparisons
- Bills per researcher, with costs tied to research complexity
Futuresearch cons
- Pricing is per‑researcher, which can add up for large datasets
- Web research may occasionally miss niche or non‑English sources
- API usage requires some familiarity with DataFrames and Python
- No built‑in GUI for advanced prompt versioning or A/B testing
- Does not replace human‑judged deep domain expertise for legal or highly regulated topics
- Latency per row depends on how hard the research problem is
- Limited control over individual web sources or domains to include/exclude
- Free tier is credit‑based and may run out quickly on complex tasks
Frequently asked questions about Futuresearch
What does FutureSearch actually do?
FutureSearch runs LLM‑powered web‑research agents on entire datasets, where each row is processed as a separate research task. You upload or send a DataFrame, define what you want—such as classification, ranking, or forecasting—in plain English, and get back a new DataFrame with enriched or reordered data. The platform handles live web searches, reading pages, and reasoning to produce structured outputs like categories, scores, merged records, or forecasts for each row.
How is FutureSearch different from a normal search engine or scraper?
Unlike a basic search engine or scraper, FutureSearch applies LLM agents that read and synthesize multiple pages per row, then write structured outputs rather than raw links. It operates at dataset scale, so each row is researched and scored or classified in parallel, and it integrates directly into DataFrame‑based workflows. The platform also supports forecasting, ranking, and matching operations that are not standard in generic scrapers or UI‑driven search tools.
Who is FutureSearch for?
FutureSearch is designed for researchers, analysts, product teams, and startups who need to screen, prioritize, or understand large sets of companies, ideas, competitors, or questions. Typical users include people doing market research, competitive intelligence, startup discovery, quantitative screening, and internal due‑diligence workflows that would otherwise require manual web searches or brittle scripts. The API also appeals to teams that already use Python and DataFrames and want to plug in automated research agents.
How does the ranking feature work?
The rank operation takes a DataFrame and a natural‑language scoring criterion, then dispatches agents to compute a score for each row based on live web research. The operation returns the same DataFrame sorted by that score, so you can quickly surface the best candidates for any written‑down criterion. The sort key does not need to exist in your original data; agents derive it at runtime by researching and synthesizing evidence from the web.
What kinds of forecasts can FutureSearch produce?
FutureSearch can forecast probabilities for binary outcomes, numeric values as percentile distributions, and future dates as percentile dates. The forecasting agents research each question in a dataset, synthesize evidence, and return calibrated forecasts with rationales, so you can see why each probability or date range was chosen. The platform is particularly useful for teams that want to systematically forecast many questions at once, such as market‑entry timing, product launches, or event outcomes.
How does classification work on FutureSearch?
The classify operation takes a DataFrame and a set of categories or labels, then runs agents that research each row and assign it to one of the categories. You can describe nuanced criteria in plain English—for example, classifying companies by whether they use a specific technology stack or business model—and the agents will attempt to infer the label from web evidence. The result is a new column in the DataFrame with each row’s category and, optionally, reasoning for the classification.
What are dedupe and merge operations for?
The dedupe operation finds duplicate or near‑duplicate rows in a DataFrame by a natural‑language equivalence relation, groups them into clusters, and selects a canonical row per cluster. The merge operation left‑joins two DataFrames using agents to resolve the key mapping instead of relying on exact or fuzzy key matches, which is useful when identifiers are inconsistent across tables. Both operations help clean and unify datasets across sources without manual mapping.
How is FutureSearch priced and billed?
FutureSearch offers $20 in free credits to start, with no credit card required. After that, users pay only for what they use, with costs measured per researcher and scaling with the complexity of each research task. Each operation—research, classify, rank, matcher (dedupe/merge), and forecast—has a different per‑researcher price point, and the platform updates your remaining credits and estimated costs in the UI as you run jobs.
Can I use FutureSearch without touching code?
Yes, you can use FutureSearch through its web app to upload or connect datasets, choose operations, and run research or forecasting workflows without writing code. The interface is designed to feel spreadsheet‑like, so you can explore operations, filters, and new columns visually. For more advanced use cases, you can then switch to the API or Python SDKs to integrate FutureSearch into your existing pipelines.
Is there an API, and how does it work?
FutureSearch exposes a full API that supports operations like rank, classify, dedupe, merge, forecast, and agent‑map over DataFrames. Each operation takes a DataFrame and a natural‑language instruction, dispatches agents to process every row in parallel, and returns an enriched DataFrame. The API is aimed at teams that already use Python and DataFrames and want to plug in automated web‑research agents into their existing ETL or analysis code.