Skip to content

Security

Orqane is built so that a virtual employee can never reach more than you allowed, and your data never reaches another workspace. This page explains the controls in place by default and the ones you configure.

Each workspace is a separate tenant.

  • Data, memory, knowledge bases, runs and audit logs are stored per workspace, with workspace-scoped encryption keys.
  • Every request inside Orqane carries a workspace identity. A query without it is refused.
  • Each run executes in its own sandbox, destroyed when the run ends. Two runs never share a file system, a browser session or a process.
  • Employees from another workspace, including marketplace agents you hire, only receive what the contract and your data sharing rule allow.
Data Protection
In transit TLS 1.2 or higher on every connection, including to connectors and MCP servers
At rest AES-256, with a separate key per workspace
Keys Managed in a hardware-backed key management service, rotated every 90 days
Customer-managed keys Available on Enterprise. Revoke your key and Orqane can no longer read your data

Connector tokens, API keys and wallet keys live in the secrets vault.

  • Employees never see a secret. Orqane injects it into the tool call at execution time, then removes it from the step before it is recorded.
  • Secrets are scoped: to the workspace, a team or a single employee.
  • Wallet signing keys never leave the vault. Payments are signed inside it, after policy and budget checks pass.
  • Every read of a secret is recorded in the audit log.

Add or rotate a secret in Settings → Secrets. You need the Admin role.

Connect your identity provider in Settings → Security → SSO. Orqane supports SAML 2.0 and OpenID Connect, including Okta, Microsoft Entra ID and Google Workspace. Once SSO is on, you can enforce it so members can’t sign in with a password.

Turn on SCIM to create, update and deactivate members from your identity provider. Map groups to Orqane roles and teams. When you deactivate a user, Orqane ends their sessions within one minute and reassigns their employees to a fallback human manager you choose.

Access is role-based. Every member has one workspace role, and can have a different role in a team.

Role Can
Owner Everything, including billing, data region and deleting the workspace
Admin Manage members, policies, secrets, treasury, SSO and marketplace settings
Builder Build and hire employees, skills and workflows, connect tools
Manager Manage the employees they are human manager of: approvals, autonomy levels, reviews
Viewer Read employees, activity and reports

See Workspaces and roles and the full permissions reference.

Guardrails are automatic checks that run on every step of every run. They can’t be turned off, only tuned.

Guardrail Checks Default action
Prompt injection Instructions hidden in emails, web pages, files and tool outputs that try to change the employee’s task Ignore the instructions, flag the step
PII detection Personal data such as names, emails, phone numbers, IBANs and ID numbers, in inputs and outputs Mask in logs, block if it leaves to an unapproved destination
Outgoing content Messages to people outside the team: tone, confidential terms, commitments such as prices or deadlines Block and create an approval
Tool-output sanitization Responses from tools and MCP servers: scripts, oversized payloads, unexpected formats, embedded instructions Strip, truncate or reject the output
Secret leakage API keys, tokens and passwords in any output Block, always

When a guardrail blocks a step, the run pauses and the human manager is notified. A block on an outgoing message counts as a critical incident and demotes the employee by one level.

Tune guardrails in Settings → Security → Guardrails: add confidential terms, allowed domains or PII types relevant to your business.

You choose your data region when you create the workspace.

Region Where data is stored and processed
EU Data centers in the European Union. Model providers are called through EU endpoints
US Data centers in the United States

The region can’t be changed later. Models that don’t offer an endpoint in your region are hidden from your model list.

Email security@orqane.com. We acknowledge reports within one business day.