Kosmos - An AI scientist system launched by FutureHouse

What is Kosmos?

Kosmos is a next-generation AI scientist, an upgrade from Robin, an automated research system launched by FutureHouse. Employing a structured world model, Kosmos can efficiently integrate massive amounts of information, parsing 1500 papers and executing 42,000 lines of code in a single run, far exceeding the processing scale of similar systems. Kosmos can rapidly complete complex research tasks in fields such as neuroscience, materials science, and statistical genetics, delivering the equivalent of six months' worth of work for a human scientist in one day, with 79.4% of its conclusions being accurate and reliable. Kosmos is fully auditable, ensuring transparent and traceable research.

Kosmos's main functions

• Automated scientific discoveriesIt can independently complete the entire scientific discovery process, from literature search and data analysis to hypothesis generation, without human intervention.
• Efficient information integrationBy efficiently integrating massive amounts of information through a structured world model, processing content on the order of tens of millions of tokens, we ensure the consistency of research objectives.
• Large-scale data analysisA single run can parse 1,500 papers and execute 42,000 lines of analysis code, far exceeding the processing scale of existing intelligent agent systems.
• Cross-domain applicationsIt is applicable to multiple disciplines such as neuroscience, materials science, statistical genetics, and cardiovascular medicine, and can reproduce and propose new scientific discoveries.
• Full auditabilityEach conclusion can be traced back to the code or literature fragment that inspired it, ensuring that the research process is transparent and verifiable.
• Accelerate the scientific research processThe amount of work completed in one day is equivalent to six months of research effort by human scientists, significantly improving research efficiency.
• Propose innovative methods: Able to independently develop new analytical methods, such as proposing new time series analysis methods in Alzheimer's disease research.
• Support scientific researchIt provides scientists with tools to help them quickly verify hypotheses, explore new directions, and generate traceable scientific reports.

Kosmos's technical principles

• Structured World ModelKosmos uses a structured world model to manage information sharing and task coordination among multiple agents, ensuring consistency and coherence in complex tasks. By continuously updating the world model, Kosmos can efficiently integrate information from different agents, supporting large-scale parallel processing and multi-step inference.
• Multi-Agent SystemIt consists of multiple intelligent agents, each responsible for a specific task, such as data analysis, literature search, or hypothesis generation. The agents collaborate through a structured world model, enabling parallel processing and efficient task allocation, significantly improving work efficiency.
• Deep Language ModelsBased on a deep language model, Kosmos can understand and generate natural language text for use in literature search and analysis. Through context management, Kosmos maintains coherence in multi-step reasoning, handling complex scientific problems.
• Efficient information integration and reasoningKosmos can efficiently integrate information from different sources, including experimental data and literature knowledge, and gradually approach scientific discoveries through multi-step reasoning. This ability enables it to discover hidden patterns and relationships in large-scale data.
• Full TraceabilityEvery conclusion in Kosmos can be traced back to specific code or literature snippets, ensuring the transparency and verifiability of the research process. Each statement in the generated scientific report is accompanied by supporting citations, facilitating verification and reproduction by scientists.
• Dynamic updates and iterationsKosmos' world model is dynamically updated based on the results of each run, progressively optimizing the analysis methods and hypotheses. Through multiple iterations, Kosmos improves the accuracy and reliability of its discoveries.
• Scientist-in-the-loopAs an auxiliary tool for scientists, it provides high-quality data and research objectives, and delivers initial findings through automated analysis and reasoning. Scientists then evaluate and validate these findings, creating an efficient feedback loop.

Kosmos project address

• Project official websitehttps://edisonscientific.com/articles/announcing-kosmos
• arXiv technical paper: https://arxiv.org/pdf/2511.02824

Application scenarios of Kosmos

• NeuroscienceTo reproduce the key role of nucleotide metabolism in the hypothermic mouse brain and reveal the molecular mechanism of neuronal vulnerability in Alzheimer's disease.
• Materials ScienceTo determine the decisive influence of absolute humidity during thermal annealing on the efficiency of perovskite solar cells, thus contributing to the optimization of material performance.
• Statistical geneticsThis study proposes a novel mechanism by which single nucleotide polymorphisms (SNPs) reduce the risk of type II diabetes, providing a new perspective for genetic research.
• Cardiovascular medicineThe study found that high levels of SOD2 may alleviate myocardial fibrosis, providing a potential therapeutic target for cardiovascular disease research.
• Alzheimer's disease research: Develop new methods to analyze the molecular event sequences of tau protein accumulation, and advance research on the pathological mechanisms of Alzheimer's disease.