Frequently asked questions

Medusa is the AI-native operating system for lean investment teams. It ingests the data your firm already has (Excel, cloud storage, broker files, market data feeds, legacy scripts), reconciles it into one consistent data layer, and runs AI agents on top:

• Risk Agent: answers exposure, factor, and stress-test questions in plain language, grounded in your own numbers.
• Optimizer: suggests daily rebalancing that keeps exposures on budget; it advises, you decide.
• Reporting & Communications: drafts scheduled client reports, investor Q&A, and RFP responses from the same data.
• Custom agents: cover the workflows specific to your firm, built alongside your team.

You keep your existing systems. Medusa runs on top of them.

Boutique hedge funds, new launches, independent asset managers, and family offices. Large firms surround the desk with risk, reporting, and quant teams. Medusa delivers the same institutional-grade answers without the headcount or the enterprise price tag. If your middle office runs on a web of spreadsheets and a few key people, you are who we built this for.

No. Medusa substitutes for the hires you never made, not the people you have: the risk, reporting, and quant teams that large firms staff and lean teams do without. Your analysts stay in the loop on every answer, and if you have an ops or reporting person, they get a team under them, not a replacement.

No. Medusa is provider-agnostic and built on your stack. It works on top of Bloomberg, MSCI, Axioma, Enfusion, or whatever you already run, and it uses your existing subscriptions rather than replacing them. No rip-and-replace, no migration project.

No. Medusa deploys into your environment (your cloud or on-premises) and builds around the systems and files you already run. Your data is not migrated onto a vendor's data model: the reconciled data layer is built inside your deployment, from your own sources, and it stays yours. The onboarding work is ours, not your team's; the pilot ingests your data where it lives. And there is no new interface to learn: you ask questions in plain language, and briefings arrive in your inbox.

Conversational AI on top of a single vendor's data is quickly becoming standard, and that is good for everyone. The difference is structural. Incumbent tools each answer from their own silo, so risk, reporting, and the portfolio system can still give you three different numbers for the same question. Medusa's agents all read from one reconciled data layer built across your sources, so the answers agree. And because the platform is AI-native, it is priced for lean teams rather than enterprise budgets, and it works across whichever vendors you already use instead of locking you into one.

Because those workflows need the firm's own data, and that is where most AI stalls.

Mercer's 2026 survey of 131 asset managers measured the share of firms with AI already integrated, workflow by workflow:

• Workflows on external data: 49% in idea generation, 40% in processing external data sets.
• Workflows on the firm's own book: 16% in client reporting, 15% in risk management, 13% in portfolio construction.

The same survey names the reason: 69% of firms cite data quality or access as the main barrier.

AI adoption stops where your internal data begins.

Research tools work on external data, so they ship anywhere. Risk, reporting, and portfolio construction cannot answer until positions, exposures, and conventions from every source are assembled into one reliable picture of your book. Medusa builds that foundation first, which is why its agents work in exactly the workflows where generic AI stalls.

In your environment. Medusa deploys in your cloud (Azure, AWS, or Google Cloud) or on-premises, in your choice of EU, US, or Switzerland regions. We come to your data; your data does not come to us. Every number an agent produces traces back to the exact file, report, or feed it came from, so any answer can be audited.

No. Medusa uses commercial AI models under agreements that exclude client data from model training, and your data stays inside your deployment.

Medusa is model-agnostic by design. Today the agents run on a leading commercial model under agreements that exclude your data from training, inside your own deployment. Our edge is fitting agents to your book and the way your team sees it, not the underlying model, so when a better model ships we can swap it in without rebuilding, and you are never locked to one AI lab. For firms that cannot send data to any external model, locally hosted open-weight models are on our roadmap; we are testing them now and will scope that deployment with you rather than promise it before it is ready.

Yes, completely. Every conversation is recorded end to end: the question asked, the answer given, and every data query the agent ran in between. Each number in an answer carries a citation to its source, so any answer can be replayed and verified after the fact. When an investor or a regulator asks how a number was produced, you can show them, step by step. Nothing the agents do is a black box.

Two loops. First, your feedback: every answer can be rated directly in the interface, and that feedback guides how we tune the agents for your workflows. Second, evaluation suites: every agent ships with a fixed set of graded cases, and a model upgrade or a prompt change has to pass them before it reaches you. You get the pace of AI improvement without the regression risk.

The Risk Agent is live: conversational risk, factor exposures, stress testing, and proactive monitoring. You can watch it work at medusaip.ai/demo. The Portfolio Optimizer and Client Reporting & Communications agents are in progress. A standalone Reconciliation agent is in design; the reconciled data layer it will write to is already how the platform organizes your data today. We are deliberate about this distinction: what the demo shows is built, and the roadmap is labeled as roadmap.

Some firms should. A large fund with a ten-person engineering bench can build and maintain its own stack. For everyone else, the honest answer is scope. Reading Excel files is the easy 20 percent. The hard 80 percent is reconciliation across sources, entity resolution, the tooling that keeps AI answers exact, and maintaining all of it as formats and feeds change. Internal builds of this kind tend to decay within months of the person who built them moving on. Medusa exists so a lean team gets the outcome without staffing the discipline.

Three steps. First, a demo and fit call: a live demo and a first look at your workflows, with no preparation needed on your side. Second, a paid pilot: about six weeks, on your own data, with a defined deliverable. Pilots deliver a working system, not a prototype. Third, production: a platform subscription plus custom agents built alongside your team.

From CHF 10K for the pilot and from CHF 3.5K per month for the platform, which typically replaces $100K to $500K per year of enterprise risk analytics. The pilot fee is credited toward the first quarter's platform fee on conversion. Custom forward-deployed work is priced as a separate retainer that declines as it standardizes into the platform, so our incentives sit with delivering value, not with keeping bespoke work perpetual. Larger firms are scoped on a fit call. Priced for lean teams, not enterprise budgets.

The label fits part of it. Medusa is the AI-native operating system for lean investment teams: it automates the workflows that surround the investment desk, investment risk, client reporting, and the reconciled data layer both stand on. We describe it from the investment seat because that is where the value lands. The desk gets its answers in seconds, and the operations burden stops consuming investment time.