OSM
Physarum polycephalum, a brainless single-celled organism that solves shortest-path problems. read the paper

We'll spend a billion to learn something.
We won't lose a euro to a decision we could have gotten right.

Every company runs on decisions. Almost optimal is expensive. OSM returns the provably optimal decision, from the data you already have.

DECISION MACHINES

Every company runs on decisions. Almost optimal is expensive.

OSM turns your data into provably optimal decisions.

About

Every company runs on decisions. Almost optimal is expensive. OSM returns the provably optimal decision, from the data you already have.

WE BUILD
DECISION MACHINES.

Enterprise margin is decided millions of times a day. Pricing, dispatch, scheduling, allocation, computed continuously by software over live data. Agents made these actions cheap to take. But a good action and the best action look identical in the moment, and the gap between them compounds into billions. Acting is solved. Acting optimally is not.

OSM closes that gap. We turn the lakehouse, where the state of the business already lives, into a world model and run a decision engine on top. Mathematical optimization with AI in the loop computes the best action under uncertainty, with proofs of feasibility and optimality, in real time.

And it runs on the stack you already have. No migration. No solver team. No brittle pipeline. The hard optimization research stays our problem. The best action becomes the easiest one to take.

People

WE'VE DONE THIS BEFORE.

A small team of researchers and engineers, in Germany and the US. We do mathematical optimization research and put it into production. Between us, we've built systems that optimized billions of transactions and saved millions of dollars.

Built by people from

TUM Microsoft Research LMU München Porsche Consulting NYU IBM UC Berkeley MIT Apple CDTM Snowflake National University of Singapore University of Massachusetts Qualcomm
Join us
Research

BUILT ON PUBLISHED RESEARCH.

Our team has done this research for decades and continues to publish. Both papers behind the decisionhouse were accepted at VLDB 2026.

  1. [1]

    Decisionhouse: Prescriptive Analytics in the Data Stack

    VLDB 2026 · Accepted

  2. [2]

    DeQL Studio: Declarative Decision-Making over Relational Data

    VLDB 2026 · Accepted

  3. [3]

    DeQL: A Decision Query Language for Prescriptive Analytics over Relational Data

    arXiv 2026 · Preprint

Follow the research