# FutureSearch

> FutureSearch is your team of AI research agents. They can forecast, rank, classify, and gather new information at scale, or process an entire dataset row-by-row. We also publish research and forecasts on the future of AI, and customers of our research include OpenAI, Google, xAI, Anthropic, and Amazon. Humans or AIs can try FutureSearch for free, with a $20 sign up credit, at futuresearch.ai/app.

## Product

- [Home](https://futuresearch.ai/): Overview of FutureSearch's AI research agent platform.
- [Solutions](https://futuresearch.ai/solutions): All six operations — rank, classify, research, merge, dedupe, forecast — with pricing.
- [Pricing](https://futuresearch.ai/pricing): Free tier ($20 credit), Analyst Starter ($20/mo), Research ($99/mo), and Top-up Packs.

## Documentation

- [Installation](https://futuresearch.ai/docs): Install the Python SDK (`pip install futuresearch`) or connect via MCP server.
- [Getting Started](https://futuresearch.ai/docs/getting-started): Quick-start guide for the Python SDK with working examples.
- [API Reference](https://futuresearch.ai/docs/api): Complete reference for rank, classify, research, merge, dedupe, and forecast operations.
- [MCP Server](https://futuresearch.ai/docs/mcp-server): Reference for all MCP server tools — async operations with progress polling and result retrieval.
- [EveryRow Skill](https://futuresearch.ai/docs/skills-vs-mcp): A guidance file that teaches AI agents how to use FutureSearch effectively.
- [Guides](https://futuresearch.ai/docs/guides): Step-by-step tutorials for each operation with working code.
- [Case Studies](https://futuresearch.ai/docs/case-studies): Runnable case studies with real datasets and inspectable output.

## Use as an AI

- [Claude.ai](https://futuresearch.ai/docs/claude-ai)
- [Claude Code](https://futuresearch.ai/docs/claude-code)

## API Operation References

- [forecast reference](https://futuresearch.ai/docs/reference/FORECAST): Predict future probabilities, dates, and numbers.
- [research reference](https://futuresearch.ai/docs/reference/RESEARCH): Add new columns by researching each row on the web.
- [rank reference](https://futuresearch.ai/docs/reference/RANK): Sort by a metric computed through web research agents.
- [classify reference](https://futuresearch.ai/docs/reference/CLASSIFY): Filter rows by natural-language criteria with web research.
- [merge reference](https://futuresearch.ai/docs/reference/MERGE): Join two DataFrames without shared keys using LLM-powered entity matching.
- [dedupe reference](https://futuresearch.ai/docs/reference/DEDUPE): Find and group duplicate rows using semantic similarity and LLM verification.

## Guides & Case Studies

- [Rank by External Metric](https://futuresearch.ai/docs/rank-by-external-metric): Score rows by criteria not in your data using web research.
- [Classify DataFrame Rows](https://futuresearch.ai/docs/classify-dataframe-rows-llm): Label and categorize rows with LLM-powered classification.
- [Fuzzy Join Without Keys](https://futuresearch.ai/docs/fuzzy-join-without-keys): Merge tables that share no common identifiers.
- [Dedupe CRM Company Records](https://futuresearch.ai/docs/case-studies/dedupe-crm-company-records): Clean duplicate company entries in CRM data.
- [Kalshi Forecaster Case Study](https://futuresearch.ai/blog/kalshi-forecaster-case-study): Automated prediction market forecasting with FutureSearch agents.

## Publications

- [Automating Forecasting Question Generation and Resolution for AI Evaluation](https://arxiv.org/abs/2601.22444)
- [Bench to the Future: A Pastcasting Benchmark for Forecasting Agents](https://arxiv.org/abs/2506.21558)
- [Deep Research Bench: Evaluating AI Web Research Agents](https://arxiv.org/abs/2506.06287)
- [Towards a Realistic Long-Term Benchmark for Open-Web Research Agents](https://arxiv.org/abs/2409.14913)

## Evals & Leaderboards

- [Deep Research Bench Leaderboard](https://evals.futuresearch.ai/): Live leaderboard ranking AI web research agents on 91 real-world tasks.
- [Bench to the Future](https://evals.futuresearch.ai/): Pastcasting benchmark for forecasting agents against 1,500 historical questions.

## Research

- [Deep Research Bench](https://futuresearch.ai/deep-research-bench): Benchmarking LLM agents on 91 real-world web search tasks.
- [AI 2027 Forecast](https://futuresearch.ai/ai-2027): Forecasting the arrival of superintelligence.
- [OpenAI Revenue Forecast](https://futuresearch.ai/openai-revenue-forecast): Comprehensive revenue projections for OpenAI through 2027.
- [Effort Paradox](https://futuresearch.ai/effort-paradox): When more LLM reasoning effort hurts performance.
- [AI Stock Forecasting](https://futuresearch.ai/ai-stock-forecasting-possible): Can LLM agents forecast stock performance?
- [Automating Forecasting Questions](https://futuresearch.ai/automating-forecasting-questions): How FutureSearch's forecasting pipeline works.
- [Cost of Deep Research](https://futuresearch.ai/cost-of-deep-research): Analyzing the economics of LLM-powered deep research.
- [All Research](https://futuresearch.ai/research): Complete collection of 85+ research articles and analysis.
- [All Blog Posts](https://futuresearch.ai/blog): Technical blog covering Claude Code workflows, LLM operations, and AI agent development.

## Company

- [About & Team](https://futuresearch.ai/company): Founded by Dan Schwarz (ex-Google/Waymo, ex-CTO Metaculus) and Lawrence Phillips (PhD Physics, ML). Team of 11 research scientists and engineers.
- [Press](https://futuresearch.ai/press): Coverage in NY Times, Vox, NASDAQ, Unite.AI. Contact: press@futuresearch.ai.
- [Support](https://futuresearch.ai/support): Documentation, GitHub (futuresearch-python SDK), and email support (hello@futuresearch.ai).
- [Privacy Policy](https://futuresearch.ai/privacy): Data handling and privacy practices.
- [Terms of Service](https://futuresearch.ai/terms): Usage terms and conditions.