how we protect your data
Annex 2 to the wysdym Data Processing Addendum: the technical and organizational measures Wysdym Inc. maintains. This page is the current version of the Annex, and it may change as Additional Term 9 on the Cover Page of the Cloud Service Agreement provides.
DPA Annex 2 · Version 1.2 · November 17, 2026
your data, walled off
1. Tenant isolation
Each customer’s data is held in its own tenant. The application scopes every query to the tenant. All tables that hold customer data are also protected by PostgreSQL row-level security policies that the application’s database role cannot bypass. Vector embeddings are stored in a separate namespace for each customer.
who gets in
2. Access control
Sign-in through WorkOS AuthKit, with support for single sign-on and multi-factor authentication. Accounts are created by invitation from Wysdym Inc. or a customer administrator, or by registration with a verified business email address. Role-based access applies per user and per agent.
nothing written without approval
3. Governed actions
The customer sets the permissions for each connected system, and changes to those systems follow the approval controls in Section 3 of the Service Description (Exhibit A). Agent actions and Gateway calls are logged; the content of Gateway calls is kept for 30 days.
in transit and at rest
4. Encryption
TLS for data in transit over public networks; the application connects to its database through Google’s encrypted Cloud SQL connector. Data stored in Google Cloud (database, object storage, secret store) is encrypted at rest with Google-managed keys. The vector-store provider encrypts data at rest and in transit.
no long-lived keys
5. Secrets and infrastructure
Application secrets are held in Google Cloud Secret Manager. Cloud authentication uses workload identity, with no long-lived service-account keys. Core infrastructure is defined as code, and changes are applied by a named engineer from a reviewed plan.
fewer identifiers in the pipeline
6. Data minimization
Automated pattern-based redaction removes email addresses, phone numbers, US Social Security numbers and, where supplied, call participants’ names from ingested source text, written Cortex pages and items surfaced for human review. From stored agent transcripts and Gateway logs it removes those identifiers and also card numbers and IP addresses. Call and CRM content is handled as Section 2.5 of the Service Description describes. Web searches containing an email address or phone number are refused. Redaction reduces, but does not eliminate, personal identifiers in content sent to model providers.
no training on your content
7. Model providers
The model providers listed in Annex 3 are used under commercial terms that exclude training on customer content.
reviewed and tested
8. Change management
Every change is made through a pull request and must pass automated tests before it can be merged and released, with separate staging and production environments.
a short list of named people
9. Personnel and access
Production access is restricted to a small number of named engineering personnel and to purpose-specific service identities. Deployment runs through keyless workload identity. Administrative database access goes through IAM-authorized Cloud SQL proxy sessions. Access grants are reviewed against the principle of least privilege.
alerts go to engineering
10. Monitoring
Alerts on deployment failures, background-processing failures and authentication failures on agent Gateway calls are routed to engineering. Cloud audit logs, including data-access logs for the database, object storage and secret store, are retained for investigation.
backups and recovery
11. Resilience
Daily automated database backups, of which the most recent 15 are retained, with point-in-time recovery over the last 7 days.
The short version -
Each customer’s data sits in its own tenant, behind row-level security. Nothing is written to a connected system without an administrator’s approval. Questions go to security@wysdym.ai.
