Security controls in every AI request.

RestrictAIChat applies identity, policy, redaction, and tamper-evident records before and after every approved model call.

Try the security demoRequest the security package →

Product controls

Controls enforced in the request path.

RestrictAIChat provides access control, policy enforcement, retention, audit evidence, and export capabilities in one request path.

  1. Decision

    Allow, mask, warn, or block

  2. Evidence

    Hash-chained record with redacted excerpts

  3. Review

    Searchable in one investigation flow

Diagram of linked decision, evidence, and review records, with the review record highlighted. Decision written to the chain Hash links verified Export ready for review

Catch known sensitive data

Detect known secrets and sensitive patterns before an approved provider call.

Approved models by team

Choose which providers and models each tenant or group can use.

Review every policy outcome

Review policy decisions and supporting audit evidence in one investigation flow.

Usage and savings reporting

Show leaders usage, provider cost, routing outcomes, and recorded savings.

The request and evidence path

Plaintext exists in the browser and gateway. Where redaction applies, only placeholders cross the provider boundary. The application has no code-execution subsystem.

  1. Employee browser

  2. Deployment DNS/TLS + reverse proxy

  3. Authentication

  4. Budget/model policy

  5. Outbound gate

  6. Approved provider

  7. Inbound gate

  8. Audit, transcript, and usage in SQLite

What each control domain covers

DomainProduct control
Data pathOutbound and inbound chat gate, encrypted placeholder redaction with authorized local restoration, bounded file upload, and local text extraction
IdentityLocal password, TOTP, Google Workspace and Microsoft Entra ID sign-in, and SCIM Bearer user/group operations
AuthorizationTenant isolation, roles, group-scoped model and provider access
ApplicationInput validation, prepared SQL, CSRF design, security headers, and rate limits
AuditAppend-only application writes, hash chain, exports, and admin events
AvailabilityHealth and readiness endpoints, SQLite WAL, and backup and recovery verification
SecretsHosted service uses platform provider credentials by default with optional encrypted BYOK; Private Edition uses customer-controlled API or supported CLI credentials
Encryption keysVersioned keys for stored credentials and records, with documented rotation that keeps previous keys readable; startup refuses traffic unless every version still in use is configured, and Managed Hosting upgrades are checked before the service is replaced
RetentionTenant-configured content expiry and cleanup jobs

Report vulnerabilities to security@restrictaichat.com.