Security Overview
Last updated: 27 September 2026
A short summary of how Vaultless Guard, provided by Techmek LLC, protects your data. The technical detail is in our Data Processing Addendum (Annex 2), the Trust page and the product's security review, which your account team can share.
1. Architecture: your data stays with you
- The gateway runs on your machines. It inspects each AI request and replaces sensitive values with tokens before anything leaves the machine, restoring them only there. Prompts, code, files and protected values are not sent to us.
- Your AI provider (for example Anthropic or OpenAI) receives tokens in place of protected values. You contract with it directly.
- The console is optional. Our hosted console receives only content-free operational data: counts per data category, versions and health, device identifiers and labels, audit-chain hashes, organisation settings, and console user accounts. Sensitive-term rules are stored only as salted digests. You can print the exact telemetry payload with
vguard org telemetry. - Your choice of deployment. Use our hosted console, host it yourself, or run fully air-gapped.
2. Hosted console controls
| Area | What we do |
|---|---|
| Encryption | TLS in transit. Secrets the service must reuse (MFA seeds, webhook and SSO secrets) are encrypted at rest (AES-256-GCM, key held outside the database). Passwords and tokens stored only as one-way digests. |
| Tenant isolation | PostgreSQL row-level security forced on every tenant table; isolation tested against the live service on every release. |
| Identity | Single sign-on with your own OpenID Connect identity provider, SCIM provisioning, MFA enforceable by role, session limits, IP allow lists. |
| Access control | Role-based access; two-person approval for rule changes and privileged changes. |
| Vendor access | Our staff use a separate, address-restricted application. Every staff view or action on your organisation is recorded in your audit trail and sent to your webhooks, and you control what staff may see. |
| Integrity | Tamper-evident, hash-chained audit logs on each gateway, anchored at the console. Rule sets can be signed with your organisation's own key, so no one without it can change what your gateways enforce. |
| Backups | Every 6 hours; each backup is restore-tested automatically, encrypted to an offline key, and copied to Amazon S3 (us-east-1) with a 30-day object lock and 90-day expiry. |
| Network | Default-deny network policies; the database cannot reach the internet; rate limiting; no application access logs. |
| Releases | Signed releases and offline bundles, pinned and hash-checked dependencies. |
3. Resilience
Protection never depends on the console: if it is unavailable, or your licence lapses, gateways keep protecting data locally. The hosted console currently runs two application replicas on a single node at Contabo in the United States. Availability commitments for Business and Enterprise are in our SLA.
4. What we have not done yet
We state this plainly so you do not discover it later:
- No SOC 2 report and no ISO 27001 certification yet.
- No third-party penetration test yet.
- Control mappings to frameworks such as PCI DSS, GDPR and the EU AI Act are mappings, not certifications.
5. Limits of detection
No detection tool catches everything. The gateway can miss values in formats it has not been taught, values split across text, identity described in prose, and traffic that does not pass through it. Add your own rules for identifiers specific to you, and use vguard prove to test what your AI provider would actually receive under your policy.
6. Reporting a vulnerability
Email devopsplus11@gmail.com or use the contact in /.well-known/security.txt. Please allow us 90 days before public disclosure. We credit reporters who want it.