Watch it work. Then try it yourself.
The intelligence of top-tier hedge funds, for lean investment teams. Start with the three-minute film.
Request access and we email your login within one working day. Free, demo data only, no commitment.
A voice at the table
Medusa sits in your investment committee.
It listens, answers on your book, and logs what you decide.
Start here · 3 min
Saltmarsh Capital is a fictional fund. Every number shown is from our demo book, as of 7 August 2026.
Deep dives
Two walkthroughs for when you want the detail.
A fund's morning, end to end
Chapter by chapter.
Deep dive · 7 min
Read what the walkthrough covers
One morning at Saltmarsh Capital, a demo multi-strategy book: 291 positions, five sleeves, six asset classes. A multistrat because it is the hardest book to run. Seven files land before the desk opens, from brokers, fund admin, and the risk engine, in different formats and often in disagreement. The book is reconciled and mapped to the desk's own sectors and themes before the first question is asked.
Then the morning starts: what moved, what broke, and what needs a decision? The agent writes the brief with every number cited to its source, and the dashboard shows the same two breaches. Then the decisions. What did the committee change this year, and did it add value, measured with hindsight. What would the agent change today. The final call stays with the human, by design. And every conversation is recorded, so any answer can be replayed months later, citation by citation.
Saltmarsh is a fictional fund. The market data behind it is real.
Go inside the risk engine
Medusa on a representative long/short equity book: a genuine factor model, factor attribution, and stress scenarios designed on demand. Investment risk, in the language you already use.
This walkthrough runs on our risk-model demo, in the same login: open it there to try the stress scenarios yourself.
Deep dive · 3 min · For investment risk professionals
What the walkthrough shows
The video walks through four things Medusa does, each as a live prompt on a representative long/short equity book:
Monitor
"What are my beta risk limits, and am I complying with them?" The agent fetches the exposures, pulls the limits, flags any breach, and reasons about the implications.
Analyze
"What is the impact of closing my Apple position?" The agent runs the counterfactual, decomposes the impact across factors, and judges whether the move is risk-efficient.
Stress test
"Design a stress scenario for a global oil supply shock and apply it to my portfolio." The agent builds the scenario from scratch, shocks interest rates, FX, sector spreads, and volatility, and runs them against the book. Scenario design that takes a risk team a full afternoon, in under a minute.
Advise
"How do I reduce my Leverage factor tilt?" The agent traces the contributors, simulates trims, searches for replacements, and cross-checks every idea against the other limits.
Every number is traceable
Each answer carries citations back to the source data, and every conversation is recorded end to end: the question asked, the answer given, and every data query the agent ran in between. When an investor or a regulator asks how a number was produced, you can show them, step by step.
From the demo to your data
These walkthroughs run on demo books: Saltmarsh is a fictional fund on real market data, and the risk engine walkthrough runs on our own equity risk model. Yours runs on your data. We connect to the risk engine you already use (MSCI, Axioma, Bloomberg), on the files and systems you already run, and you ask the questions.
Every vendor shows you their version of your portfolio. Medusa shows you yours.
Show us your workflows, and we will build agents fitted to them. In about six weeks, on your own data, you will see what your team can achieve.
Start with a free review of your risk process.
One call, one page back: what a large firm's risk desk would see, and what can be automated.
Download the one-pager (PDF) to share with your team.
Runs in your own cloud · Externally reviewed · Your data never trains a model · Security and data ownership →