SECURITY & TRUST

Built to keep your knowledge yours.

Last updated: September 2026

This page describes Atlas Vault's security architecture in plain language, for customers evaluating the product during beta. It is not a certification, an audit report, or a substitute for your own security review. We have not obtained SOC 2, ISO 27001, or similar third-party certifications, and we won't claim otherwise.

Local-first architecture

Atlas Vault is designed to run on infrastructure your organization controls. The language model that answers questions and the vector search that retrieves relevant passages from your Vault both run locally in a typical self-hosted deployment — your document content and conversations are not routed through an external AI provider to generate answers. The one deliberate exception is Web Research: when a question genuinely needs current public information, Atlas queries public search infrastructure and public websites over the internet. That reach is scoped to public information retrieval and never includes your private Vault content.

Customer-controlled Vault

Your documents live in your own deployment's storage and database, not a shared multi-tenant cloud you don't control.

Workspace & role controls

Access is scoped by organization, workspace, and a five-level role model (viewer through owner) enforced on every server-side request.

Evidence & provenance

Answers are built to trace back to a source — a Vault page or a public citation — so claims can be checked, not just trusted.

Human-approved Agent writes

Scheduled Agents can research and propose, but nothing reaches your Vault until a person explicitly approves it.

Account & session security

Password storage: passwords are hashed with Argon2 (a modern, memory-hard hashing algorithm) and are never stored, logged, or transmitted in plain text.
Password policy: a 12-character minimum is enforced, with no arbitrary complexity theater — long passphrases are supported, not penalized.
Sessions: authentication uses a server-side session record referenced by an HttpOnly cookie — the session identifier is never readable by page JavaScript, which closes off an entire class of token-theft attacks (e.g. via a malicious script).
Sign-out: logging out revokes that session server-side immediately, not just on the browser. Changing your password revokes every other active session automatically.
Login protection: repeated failed sign-in attempts against one account or from one source are automatically throttled.

Authorization

Every privileged action — approving an Agent's proposal, deleting a document, managing workspace members, administering an organization — is independently checked on the server against the requesting user's actual role. Hiding a button in the interface is never the only thing standing between a user and an action they're not authorized to take.

What we're still building

In the interest of not overselling a beta product, here's what's explicitly not in place yet:

Self-service password recovery via email — today, a workspace administrator resets a forgotten password directly. We have not built outbound transactional email yet, and we'd rather say so than fake it.
Third-party security certifications (SOC 2, ISO 27001, or similar).
A formal, published data-retention schedule.

Reporting a security concern

If you believe you've found a security issue in Atlas Vault, we want to hear about it. We don't yet have a dedicated security-reporting mailbox, so for now, please reach us through the Request a demo form and mark your message as a security report — it reaches a real inbox, and we'll follow up directly.

For how we handle the information Atlas Vault processes, see our Privacy Policy.