For the people doing the homework.

Metadata Minder is a containerized pipeline composed of well-known, auditable open-source components — orchestrated by Rietta's hardened glue code. Every stage is replaceable, inspectable, and runs inside your perimeter.

We can provide an architecture diagram and component inventory under MNDA on request.

• No external data transfer by default. The deployment ships with outbound network access to public AI providers disabled.
• Local LLM inference via Ollama. Document content is never sent to OpenAI, Anthropic, Google, or any third-party LLM.
• Encryption. Transport is TLS 1.2+. At-rest encryption is provided by the host filesystem or AWS GovCloud KMS — keys stay with you.
• Data residency. All processing occurs on agency-controlled infrastructure. No cross-border replication.
• Retention & disposal. Configurable per jurisdiction. Findings reports and source artifacts can be purged on a defined schedule with cryptographic deletion confirmation.

Pilot Phase 2

During the pilot phase, access is managed directly by Rietta.

• SBOM on request in CycloneDX format for each release.
• Signed container images. Releases are published with verifiable signatures so your IT team can validate provenance before promotion.
• Continuous dependency scanning. Third-party packages are tracked against published CVE feeds; patches are scheduled to outrun disclosed vulnerabilities, not chase them.
• Reproducible builds. The same containerized build process runs on developer machines, in CI, and in production.

Metadata Minder inherits Rietta's appsec practice in full: mandatory security-oriented code review on every change, branch-protected merges, automated test gates, and a Continuous Blue Team relationship between engineering and security review.

Rietta has spent over two decades patching production systems within hours of disclosed vulnerabilities. We extend the same SLAs to Metadata Minder deployments, with optional managed-update channels for agencies that want patches applied on our cadence rather than theirs.

• Critical CVE response: patch released within 72 hours of public disclosure when exploitable.
• High CVE response: patch released within 14 days.
• Coordinated disclosure: [email protected] is monitored. PGP-encrypted channel available on request.

• Append-only audit log of authentication, document access, finding generation, and report export — exportable to your SIEM (Splunk, Elastic, Sentinel) via syslog or JSON.
• Reproducible findings. Each report is keyed to a discovery-pass ID; rerunning the pass against the same archive snapshot reproduces the same findings.
• Health and performance metrics exposed via Prometheus-compatible endpoints, scoped to your monitoring stack.

Severity classification, communication SLAs, and customer-side containment steps are documented in the Operations Manual provided with each deployment. Rietta engineers are reachable directly — not through a tier-1 ticket queue — for any production-impacting incident.

Metadata Minder itself is not currently FedRAMP-authorized. Deployment inside your existing FedRAMP boundary (such as AWS GovCloud) inherits that boundary's controls.

Our crawler honors robots.txt, contains scope to authorized URIs, uses HEAD requests for change detection, and limits re-examination to roughly once per calendar month with non-predictable timing. Full disclosure on the dedicated crawler page.

References from existing public-sector and HIPAA-covered customers available under MNDA during procurement.