Dataryo is the system of record for documents that matter. We treat that responsibility with the boring, paperwork-heavy seriousness it deserves.
We organize our program around four pillars: data boundaries, identity, observability, and vendor discipline. Every control we ship maps back to one of these.
TLS 1.3 on the wire, AES-256-GCM at rest. Per-tenant encryption keys managed in AWS KMS, with customer-managed keys (CMK) on Enterprise.
Choose US, EU (Frankfurt), or UK (London) for stored workspace data. AI subprocessors follow the configured provider policy; embeddings use paid Google Gemini API.
Uploaded files are write-once. Transformations produce new records; the source is always reachable by hash.
Contractually. Your files, prompts, and outputs are used to serve your workspace only. This is a term, not a toggle.
SAML 2.0 with Okta, Azure AD, Google, and OneLogin. SCIM provisioning and de-provisioning on Team and Enterprise.
Owner, admin, editor, viewer. Folder-level ACLs. Granular controls for Transformations and export.
Every read, write, prompt, citation, and export — timestamped and exportable. Stream to Splunk, Datadog, or S3.
Every AI output carries a cryptographic manifest of the source chunks it was grounded in. Reproducible, forever.
Infrastructure, auth, billing, Azure Direct model inference, and paid Google Gemini embeddings are tracked as approved subprocessors. Public list maintained at a dedicated URL.
Hybrid routing keeps non-embedding calls on Azure Direct and embeddings on paid Google Gemini API. Usage rows store metadata and costs, not prompts or outputs.
We run a continuous compliance program with drift monitoring across AWS, GitHub, Okta, and our code. Every control has an owner and a review cadence.
We run a private bounty with coordinated disclosure. No lawyers, no NDAs for good-faith research. Report to security@dataryo.com.