Your data never leaves your cloud.

In your own cloud account. The data plane and compute plane deploy there, and all customer data, including run logs and intermediate artifacts, stays there. Residency follows your regional preferences.

No. Your data never trains or fine-tunes a Varick model, and Varick never shares your data with other customers. Inference runs on a model service in your cloud, such as Amazon Bedrock or Azure AI Foundry, and third-party providers are used under enterprise terms that contractually exclude customer data from their training.

Only through accounts you provision. Our engineers work read-only during the 4-week pilot, then with access scoped to each statement of work once agents go live, and you can revoke it at any time. Varick's control plane holds no standing access: break-glass access is initiated by you, approved, and time‑bounded. The control plane sends releases in and receives aggregate metrics that your environment pushes out. It never pulls data.

Agents work inside your existing systems, under your existing permissions and audit requirements. During the pilot they are read-only. Later write actions are scoped per statement of work, with segregation of duties, authorization limits, controlled change management on models, prompts, and logic, and an immutable audit trail.

Request it on this page. We share the security and architecture overview, the control set, and completed security questionnaires under NDA. We do not publish them for download.