← All companies

Science / Company profile

FutureHouse

What FutureHouse does

FutureHouse is a San Francisco-based 501(c)(3) nonprofit research organization building AI “scientist” systems intended to automate and accelerate fundamental discovery in biology and other complex sciences. The organization describes its core approach as pairing domain scientists with advanced AI agents, with a flagship initiative that employs early-career researchers via its AI-for-Science Independent Postdoctoral Fellowship and provides them access to FutureHouse’s AI systems, compute, and wet-lab resources. FutureHouse also operates an agent platform for scientific literature search and synthesis and chemistry-related workflows, and it publishes research on enabling components such as benchmark creation, agent training in realistic environments, and scientific reasoning models. A key current strategic position is that FutureHouse’s nonprofit research work is complemented by a for-profit spinout, Edison Scientific (spun out in 2025), which commercializes the AI tools; FutureHouse states it has access to Edison Scientific’s technology while continuing to focus on nonprofit, public-good fundamental research.

News

Company record

May 19, 2026 · Official · FutureHouseDemonstrating end-to-end scientific discovery with Robin: a multi-agent system (Nature-published update)

FutureHouse announced Robin as its integrated multi-agent workflow orchestrating literature search, evaluation, and data analysis to generate hypotheses, design experiments, and produce an AI-driven first discovery; the page includes an update stating the work is now published in Nature and describes its therapeutic hypothesis, experimental validation, and mechanism investigation.

Apr 08, 2026 · Official · FutureHouseDISCO: Inventing Enzymes for Chemistry that Nature Never Explored

FutureHouse published research describing DISCO (Diffusion for Sequence-structure CO-design), a multimodal generative model intended to co-design amino-acid sequence and 3D protein structure for enzymes targeting reactions not explored by natural evolution (as presented on the FutureHouse research page).

Dec 10, 2025 · Official · FutureHouseOXtal: Generative Molecular Crystal Structure Prediction

FutureHouse published a research highlight page for OXtal, described as a generative diffusion model for predicting experimentally realizable molecular crystal structures from 2D molecular graphs.

Nov 05, 2025 · Official · FutureHouseAnnouncing Edison Scientific

FutureHouse announced Edison Scientific as a commercial spinout and framed it as the vehicle for commercializing FutureHouse’s AI-agent platform and related tooling, summarizing multiple technical and product steps achieved in FutureHouse’s first two years.

May 01, 2025 · Official · FutureHouseFutureHouse Platform: Superintelligent AI Agents for Scientific Discovery

FutureHouse announced the launch of its public agent platform accessible via web and API, describing four initial agents (Crow, Falcon, Owl/HasAnyone, and Phoenix as an experimental ChemCrow deployment) and the intended use cases, including literature search and chemistry experiment planning.

Date not disclosed · Official · FutureHouseether0: a scientific reasoning model for chemistry

FutureHouse published a research note describing ether0 as a chemistry-specific reasoning model trained using reinforcement learning, and characterizing its capabilities and limitations in molecule design and chemistry tasks.

Source map · 2 recurring channels · 14 references

Still resolving: Newsroom · Official X

What it builds

FutureHouse Platform (web interface + API)

A public agent platform made available via a web interface and API, launched with multiple specialized AI agents for scientific discovery tasks such as literature search/synthesis and precedent (“has anyone done X before”) retrieval, with an experimental chemistry-planning agent deployed via Phoenix.

source ↗

Milestones & partnerships

May 19, 2026
First integrated discovery demonstration with Robin (Nature-published)

FutureHouse reported Robin’s end-to-end automated workflow and described its therapeutic hypothesis generation, experimental validation, and subsequent discovery elements in a Nature-published update.

External collaborators referenced on the Robin page
Apr 08, 2026
DISCO milestone: multimodal sequence-structure co-design model for enzymes

FutureHouse presented DISCO as a multimodal generative model that co-designs protein sequence and 3D structure conditioned on biomolecules, targeting enzymatic functions for chemistry not explored by nature (as described in the research page).

Caltech, Mila, and other collaborating entities named on the page
Dec 10, 2025
OXtal milestone: generative crystal structure prediction model

FutureHouse described OXtal as a generative diffusion model predicting experimentally realizable molecular crystal packings directly from 2D molecular graphs (as stated on the research page).

Caltech, University of Oxford, Mila, and other collaborators named on the page
Nov 05, 2025
Spun out Edison Scientific to commercialize AI tools for science

FutureHouse announced Edison Scientific, positioning it as a for-profit commercialization pathway for FutureHouse’s AI-agent technology while FutureHouse continues its nonprofit fundamental research mission.

Edison Scientific
May 01, 2025
Launched the FutureHouse Platform (web + API) with specialized scientific agents

FutureHouse released a public platform featuring four specialized agents for literature search/synthesis, precedent (“has anyone done X”), and an experimental chemistry-planning workflow via Phoenix.

Researchers and research groups using the platform
Jan 01, 2024
Built and validated benchmarked agent capabilities for scientific literature tasks (PaperQA2 referenced)

FutureHouse’s About page characterizes PaperQA2 as an AI agent achieving superhuman performance on scientific literature tasks and used as evidence of its progress in automated scientific reasoning components.

Peer/benchmarking processes mentioned by FutureHouse
Date not disclosed
Aviary research milestone: training language agents on challenging tasks

FutureHouse described Aviary as enabling training language agents on scientific tasks in realistic environments and discussed safety framing and evaluation using LAB-Bench.

LAB-Bench evaluation framework
Date not disclosed
ether0 research milestone: reinforcement-learning driven chemistry reasoning model

FutureHouse described ether0 as a chemistry-focused reasoning model trained with reinforcement learning to improve molecule working/consistency behaviors.

Chemistry benchmark-style tasks mentioned by FutureHouse