Your most sensitive communications deserve the most advanced protection

No. Email body text, subject lines, and recipient addresses are processed in memory and discarded when the serverless function execution completes. There is no database table, file, cache, queue, or log that holds email content. This is enforced by code architecture (the analysis function has no code path that writes content to any persistent store), verified by automated CI tests (canary string tests, metadata schema tests), and auditable by any customer or third-party assessor (source code review available under NDA).

The only data persisted is analysis metadata: analysisId, timestamp, riskScore, riskLevel, categories[] (names only), riskCount, processingTimeMs, emailCharCount, and truncated. No content, no excerpts, no subject lines, no recipient addresses.

No. We use Anthropic's commercial API and/or AWS Bedrock. Under Anthropic's commercial terms, API inputs are not used for model training and are not retained after the response is generated. Under AWS Bedrock's terms, inputs are not stored or logged, and are not used for model training. Both providers offer contractual commitments to this effect via their enterprise agreements. For customers requiring additional assurance, our Maximum tier runs a local LLM within the customer's own infrastructure with no external API calls.

No. This is not a policy constraint. It is an architectural constraint. There is no system for a Praelyx employee to access email content because email content does not exist in any accessible store. Application logs contain only metadata. The metadata store contains only metadata. The admin dashboard shows aggregate statistics only. In Enhanced and Maximum tiers, content is processed inside hardware-isolated enclaves with no SSH, no debug tools, and no remote shell access. Even a compromised Praelyx employee with root access to our infrastructure cannot extract email content.

We cannot provide what we do not have. If served with a legal request for email content, we can demonstrate that no email content exists in our systems. The most a request could compel is analysis metadata (risk scores, categories, timestamps, hashed user IDs), which contains no email text, subject lines, or recipient addresses. We maintain a quarterly transparency report and monthly warrant canary, both cryptographically signed and timestamped. Praelyx is registered in France (Seraphim SAS, 229 rue Saint-Honore, 75001 Paris), and legal requests are evaluated under applicable French and EU law.

An attacker who compromises our infrastructure would find: (1) Analysis metadata with no content. (2) Application logs with no content. (3) No database of email bodies, subjects, or recipients to exfiltrate. The attack surface is minimal because the attack target (email content) does not exist at rest. In a worst-case real-time attack on a running Lambda function, an attacker could access the content of one email being processed at that instant, within a 1-3 second window, before the function completes and the content is discarded. Enhanced tier mitigates this further with enclave processing where even the host OS cannot access function memory.

In Standard tier: the email body (HTML stripped, max 10,000 characters), subject line, and recipient domains (not full addresses). Sender email address and full recipient addresses are never transmitted. Attachment content is not processed. In Enhanced tier with client-side NER: PII is replaced with typed placeholders ([PERSON_1], [PHONE_1]) before transmission, so personal identifiers never leave the device. The local rule engine handles 50-80% of checks without any network call. In Maximum tier with customer-hosted deployment: nothing leaves your network.

Yes. The admin dashboard supports Azure AD / Okta SSO via SAML 2.0 and OIDC. SCIM provisioning is supported for automated user lifecycle management. Role-based access control separates admin roles: Organization Admin (full config access, aggregate analytics), Department Admin (department-scoped analytics, sensitivity configuration), and Viewer (read-only aggregate dashboards). MFA is enforced for all admin access. Session timeout is enforced. No admin role has access to email content because no content exists in any accessible system.

Our incident response plan follows NIST SP 800-61 guidelines. Detection: automated monitoring via cloud-native alerting, anomaly detection on API usage patterns, and failed authentication monitoring. Triage: security officer evaluates severity within 1 hour. Notification: affected customers notified within 72 hours of confirmed breach (or within 24 hours for high-severity incidents). Remediation: root cause analysis, containment, eradication, recovery. Post-incident: public post-mortem for security incidents affecting customers, with timeline, root cause, and corrective actions. Due to our zero-retention architecture, the impact scope of any breach is inherently limited to metadata.

Yes. We provide multiple audit paths: (1) Source code review under NDA for the analysis function. (2) Infrastructure audit covering serverless configuration, metadata store schema, and log configuration. (3) Canary token testing where your team sends marked test emails and searches for them across our persistent storage. (4) SOC 2 Type II report provided annually. (5) Penetration test results shared under NDA. (6) Customer-hosted deployment for organizations that want to audit their own infrastructure rather than ours. Audit rights are included in all enterprise agreements.

Fundamentally different in architecture and purpose. DLP tools intercept email in transit and maintain persistent content stores for policy enforcement and forensic review. Email monitoring tools continuously analyze and archive employee communications. Praelyx analyzes the email before it is sent, returns a risk assessment to the sender, and discards the content. The sender sees the results. Management sees aggregate statistics (total flags by category, override rates) but never individual emails or per-user flag details. We are a risk spell-check, not a surveillance tool. Our zero-retention architecture means there is no content archive to subpoena, breach, or misuse.