Security center
Encryption, the tool firewall, the AI kill switch, access rules, single sign-on and the audit log in one place.
Open Security in the ORYVOX Core menu group. Sensitive changes ask for your password again.
Pause all AI
One switch, for emergencies: Pause all AI now. No chatbot answers through a model, no tool runs and API requests are refused, until you Resume AI.
Scorecard
Blocked actions and actions held for confirmation or approval in the last 24 hours, injection attempts and data-loss hits in the last 7 days, and what's in place or still open.
Encryption and keys
Customer content is encrypted with your organisation's own key; that key is wrapped by a key-management service and never stored in plain form. Rotate it by hand, or set days between automatic rotations.
Security policy
The model only proposes actions; these rules decide whether they run, wait for a person, or are blocked.
| Setting | |
|---|---|
| Security mode | Standard; Strict: only providers that don't train on data, no semantic cache; Confidential: also zero retention at the provider, personal data masked before the model, zero retention by default. |
| Financial actions | Refunds, credits and payments: up to a first amount the tool's own policy applies; up to a second a person approves; above it the action is blocked. |
| Allowed outgoing domains | Custom tools, webhooks and MCP servers may call only these domains. Empty allows any public address. |
| Data loss prevention | What happens when an answer or tool result contains each kind of data: allow, mask or block. |
| Shadow mode | The firewall only records what it would decide. Use it to try stricter rules; the kill switch and quarantine still hold. |
Firewall decisions lists the latest decisions with the rules behind them.
Agents in quarantine
An agent in quarantine answers nothing and runs no tool. Agents are quarantined automatically after repeated blocked actions; you can quarantine or release one by hand.
Secrets vault
The API keys, tokens and passwords of your integrations, with when each was last used. Values are never shown, only a hint.
Access rules
For the whole company, the dashboard and every API key:
- IP allowlist: one IP or CIDR range per line;
- Require two-factor authentication: members without an authenticator app are sent to set one up;
- Session length: sign everyone out after N hours;
- Alert email: security alerts also go here.
Single sign-on
Sign in with Okta, Microsoft Entra ID, Google Workspace or any OpenID Connect provider (SAML too). Verify your domain first; people on verified domains join with the default role the first time. Enforce SSO makes your identity provider the only way in for those domains.
Security events
Blocked secrets, masked personal data, unknown IPs, failed sign-ins and policy changes, by severity. Warnings and critical events are emailed to owners and admins.
Audit log
Every change in the company: who, what, before and after, when and from where. Filter by person and action, search, and export to CSV or JSON. Each entry carries the hash of the one before it, so changing or removing an entry breaks the chain; a signed checkpoint is made every night. Some features, such as the full audit log and SSO, are part of higher plans.