Security & Trust
Governance is the product. Security is its foundation.
An accountable AI operation has to be a safe one. The controls below are how Tapestry protects customer data — described plainly, and limited to what we actually do.
Built into the platform.
These controls are part of how every Tapestry operation runs, not options bolted on afterward.
01Tenant isolation
Every customer's data is scoped to their company at the database layer. Reads and writes go through company-scoped access paths, an automated isolation check runs on every code change, and requests for another tenant's records return the same response as records that do not exist.
02Role-based permissions
Access is governed by defined customer roles — owner, admin, approver, operator, and viewer — each granted a specific set of actions. Permissions are enforced on the server for every write; interface-level checks are a convenience, never the control.
03Human approval controls
Consequential actions route through an approval system with configurable risk thresholds. Work above a company's risk boundary cannot be approved past the gate, and outbound communication follows a propose-then-approve path — nothing sends without sign-off.
04Audited operating history
Actions taken in an operation are recorded to an activity log: who acted, on what, and when. AI calls are separately audited without storing prompt or response content — the audit record keeps a keyed one-way hash, never the text.
05Connected-system credentials
Credentials for connected systems are stored encrypted at the application layer with dedicated keys, separate from application data. Connections are scoped to the role and workflow that require them.
06Machine-to-machine intake
Inbound machine endpoints authenticate with signed requests — HMAC signatures over exact request bytes, with replay-window and constant-time verification.
07Encryption in transit and at rest
Tapestry runs on managed cloud infrastructure. Connections are encrypted in transit with TLS, and underlying storage is encrypted at rest by the cloud provider.
08Data retention
Retention policies are configurable per company and enforced by scheduled sweeps rather than manual cleanup — for example, lead data supports a defined retention period after last activity.
09AI model access
All model access is routed through a single controlled gateway operated by Tapestry, with per-environment credentials. Specialists reach models only through that gateway — never with customer-held keys.
Discussed openly, documented on request.
Enterprise review usually goes deeper than a marketing page should. Where we are not ready to publish detail, we say so — and cover it directly in your security review instead.
Model-provider data handling
Details of provider-level data handling — including whether customer data is used for model training — are documented per engagement. Ask us during evaluation and we will put the commitments in writing for your agreement.
Incident response & subprocessors
Incident-response procedures and the current subprocessor list are available on request as part of security review.