Your data stays in your cloud.

• Medusa runs in your own cloud account (Azure, AWS or Google Cloud) or on your own servers, in the region you choose, behind a VPN with no public address.
• Your data is stored and processed there. We come to your data; it does not come to us.
• Nothing is migrated. Medusa reads the systems you already run, so there is nothing to move back.

• Your users, with named accounts. Sign-in with your corporate identity is on the roadmap. Sharing happens only within groups your administrator defines, and every share and withdrawal is recorded.
• Named Medusa engineers, through access you delegate to one isolated part of your account.
• You can withdraw our cloud access in minutes. A documented checklist removes the rest (VPN, server and application credentials), and the deployment keeps running throughout.
• We cannot give anyone permissions in your cloud account, change the access you delegated, or reach your other resources.

• Runs in your cloud, on the model you choose: a leading model through your own Azure or AWS private endpoint, or an open-weight model so nothing leaves your network. With a private endpoint, each question to the agent, with the figures it retrieved to answer it, goes to your endpoint, under terms that exclude your data from model training. With an open-weight model, nothing leaves your network. Open-weight models are tested on our risk agent, and we test them on your own questions when you ask.
• During onboarding we work with your files the way you prefer: inside your environment, on synthetic samples, or on working copies that we delete at the end of the engagement and confirm in writing.

• Every change we make to your cloud resources, in your own activity log, which we cannot alter.
• Every answer is recorded step by step, attributed to the user: the question, each lookup the agent made and the data it returned, and the answer itself. Any answer can be replayed and checked months later.
• Every number in an answer is checked against the data the agent retrieved.

In August 2026, before he took the role of fractional CTO, Balki Kodarapu carried out a paid review of our deployment design against SOC 2, NYDFS 500 and CIS Azure criteria. Balki is a senior engineering leader with a background in security and compliance in regulated environments. He found the design architecturally sound, and every finding was at the configuration level, with no redesign needed. The findings were remediated in October 2026, and a verification review follows.

SOC 2. We do not hold a SOC 2 report today. Type I is next; Type II follows once the controls have run for the observation period. Until then we answer your security questionnaire in full.

• Petr Merkuryev, founder. 22 years at the intersection of finance, technology and data science, including seven years running quantitative research and risk management at Pictet’s multi-strategy hedge fund.
• Balki Kodarapu, fractional CTO. Owns architecture and security; your IT team’s technical counterpart.
• References. A client reference call is available on request, by arrangement.

• If an incident affects your deployment, we tell you within 24 hours of confirming it, with a written report afterwards.
• Your firm-specific work (mappings, configuration, agent instructions) belongs to you.
• You can export your data in standard formats at any time, at no charge.

The free review is a screen share: we receive no files and no access to your systems. The live demo holds a fictional fund on public market data, and no client data.