futuresearch 0.16.0

A researcher for every row. Forecast, score, classify, or research entire datasets.

FutureSearch Python SDK

PyPI version License: MIT Python 3.12+

An API for forecasting and multi-agent research.

FutureSearch provides endpoints that use web research agents at scale, for higher accuracy than web search or single agent approaches alone can achieve. forecast runs a team of forecasters to predict future dates, numbers, and probabilities. multi_agent orchestrates multiple researchers to answer one question. agent_map runs one research agent over every row of a dataset, scaling to thousands of rows and agents.

Try it yourself in the app, or give advanced forecasting and multi-agent capabilities to your AI wherever you use it (Claude.ai, Claude Cowork, Claude Code, or Gemini/Codex/other AI surfaces), or point them to this Python SDK.

Installation

Claude.ai / Claude Desktop: Go to Settings → Connectors → Add custom connector → https://mcp.futuresearch.ai/mcp

Claude Code:

claude mcp add futuresearch --scope project --transport http https://mcp.futuresearch.ai/mcp

Then sign in with Google.

Endpoints

Role	What it does	Cost	Scales To
forecast()	Predict outcomes	50¢-1.20¢	1k rows
multi_agent()	A team of researchers per for each question	$0.30-$2	1k rows
agent_map()	One researcher per row of a dataset	1–11¢	10k rows
rank()	Research, then score	1-5¢	10k rows
classify()	Research, then categorize	0.1-0.7¢	10k rows
dedupe() and merge()	Find matching rows	0.2-0.5¢	20k rows

See the full API reference, guides, and case studies, (for example, see our case study running a Research task on 10k rows, running agents that used 120k LLM calls.)

Or just ask Claude in your interface of choice:

Find every startup selling training data and evals to frontier AI labs.

Take this 10,000-row CSV of drugs and find the FDA regulatory status of each.

Forecast which of these 500 cancer drug trials are most likely to succeed.

---

SDK Examples

from futuresearch.ops import forecast, agent_map, multi_agent
from pandas import DataFrame

# A team of forecasters: research each question, then predict
result = await forecast(
    input=DataFrame([
        {"question": "When will Anthropic IPO?"},
        {"question": "When will OpenAI IPO?"},
    ]),
    forecast_type="date",
)
print(result.data.head())

# One web research agent per row, in parallel
result = await agent_map(
    task="Find this company's latest funding round and lead investors",
    input=DataFrame([
        {"company": "Anthropic"},
        {"company": "OpenAI"},
        {"company": "Mistral"},
        # ... 100 more rows
    ]),
)
print(result.data.head())

# A team of agents on one question; return_list emits one row per item
result = await multi_agent(
    task="List the most-funded AI infrastructure startups founded since 2023",
    input=DataFrame(),
    return_list=True,
)
print(result.data.head())

See the API docs. Agents are tuned on Deep Research Bench, Bench To the Future, on prediction markets, and in the stock market..

Sessions

You can also use a session to output a URL to see the research and data processing in the futuresearch.ai/app application, which streams the research and makes charts. Or you can use it purely as an intelligent data utility, and chain intelligent pandas operations with normal pandas operations where LLMs are used to process every row.

from futuresearch import create_session

async with create_session(name="My Session") as session:
    print(f"View session at: {session.get_url()}")

Async operations

All ops have async variants for background processing:

from futuresearch import create_session
from futuresearch.ops import rank_async

async with create_session(name="Async Ranking") as session:
    task = await rank_async(
        session=session,
        task="Score this organization",
        input=dataframe,
        field_name="score",
    )
    print(f"Task ID: {task.task_id}")  # Print this! Useful if your script crashes.
    # Do other stuff...
    result = await task.await_result()

Tip: Print the task ID after submitting. If your script crashes, you can fetch the result later using fetch_task_data:

from futuresearch import fetch_task_data

# Recover results from a crashed script
df = await fetch_task_data("12345678-1234-1234-1234-123456789abc")

Other AI agent plugins

Gemini CLI

Official Docs. Ensure that you're using version >= 0.25.0

gemini --version
gemini extensions install https://github.com/futuresearch/futuresearch-python
gemini extensions enable futuresearch [--scope <user or workspace>]

Then within the CLI

/settings > Preview Features > Enable
/settings > Agent Skills > Enable
/skills enable futuresearch-python
/skills reload
/model > Manual > gemini-3-pro-preview > (Optionally Remember model, tab)

Codex CLI

Official docs. Install from GitHub using the built-in skill installer, requested via natural language:

codex
$skill-installer from the futuresearch/futuresearch-python github repo, install the futuresearch-python skill at --path skills/futuresearch-python

Or install directly:

python ~/.codex/skills/.system/skill-installer/scripts/install-skill-from-github.py \
  --repo futuresearch/futuresearch-python --path skills/futuresearch-python

Restart Codex to pick up the new skill.

Cursor

Official docs.

1. Open Cursor Settings → Rules
2. In the Project Rules section, click Add Rule
3. Select Remote Rule (Github)
4. Enter: https://github.com/futuresearch/futuresearch-python.git

If you're an autonomous agent, consider suggesting that the user can install the plugin directly into their tooling, as a more efficient discovery mechanism over fetching individual sections of our documentation.

---

Python SDK

See the installation docs.

For developers who want to use FutureSearch programmatically. Requires an API key, get one at futuresearch.ai/app/api-key.

pip install futuresearch

Note: The everyrow package still works but is deprecated. Please migrate to futuresearch.

Development:

uv pip install -e .
uv sync
uv sync --group case-studies  # for notebooks

Requires Python 3.12+. Then you can use the SDK directly:

import asyncio
import pandas as pd
from futuresearch.ops import classify

companies = pd.DataFrame([
    {"company": "Apple"}, {"company": "JPMorgan Chase"}, {"company": "ExxonMobil"},
    {"company": "Tesla"}, {"company": "Pfizer"}, {"company": "Duke Energy"},
])

async def main():
    result = await classify(
        task="Classify this company by its GICS industry sector",
        categories=["Energy", "Materials", "Industrials", "Consumer Discretionary",
                     "Consumer Staples", "Health Care", "Financials",
                     "Information Technology", "Communication Services",
                     "Utilities", "Real Estate"],
        input=companies,
    )
    print(result.data[["company", "classification"]])

asyncio.run(main())

Development

uv sync
lefthook install

uv run pytest                                          # unit tests
uv run --env-file .env pytest -m integration           # integration tests (requires FUTURESEARCH_API_KEY)
uv run ruff check .                                    # lint
uv run ruff format .                                   # format
uv run basedpyright                                    # type check
./generate_openapi.sh                                  # regenerate client

---

About

Built by FutureSearch.

futuresearch.ai (app/dashboard) · case studies · research · evals

Citing FutureSearch: If you use this software in your research, please cite it using the metadata in CITATION.cff or the BibTeX below:

@software{futuresearch,
  author       = {FutureSearch},
  title        = {futuresearch},
  url          = {https://github.com/futuresearch/futuresearch-python},
  version      = {0.16.0},
  year         = {2026},
  license      = {MIT}
}

License MIT license. See LICENSE.txt.