Security is the product.
Private AI only works if you control the boundary. Here's how we build systems you can bet your business on.
Deploy where you need it
Choose the deployment model that fits your security requirements.
VPC Deployment
Runs in your AWS/Azure/GCP environment. Data stays within your cloud account.
- Your cloud account
- Your network rules
- Your encryption keys
On-Premise
Deploys to your physical infrastructure. Full hardware control.
- Your data center
- Your hardware
- Your physical security
Air-Gapped
Zero network connectivity. For classified and high-security environments.
- No internet required
- Complete isolation
- Classified-ready
What stays inside. What you control.
Complete control over your data boundaries with full transparency.
Inside your boundary
- All documents, emails, and communications
- All AI inference (LLM runs locally)
- All embeddings and vector storage
- All audit logs and artifacts
Default posture
(regulated deployments)
- Prompts and responses stay inside your boundary
- No external API calls for inference
- Credentials managed in your vault
Optional external
(if configured)
- Explicitly governed and audit-logged
- Disabled by default in regulated deployments
- Configurable per-workflow with approval controls
- Customer controls which workflows can use external models
Audit trail
Every AI action is logged with timestamp, user, input hash, output hash, and purpose. Exportable for compliance review.
System Architecture
Every component runs inside your security boundary. External data flows in — nothing flows out.
Identity, secrets, access — controlled by design
Enterprise-grade governance built into every layer.
Identity
- SSO integration (Azure AD, Okta, SAML)
- Role-based access control
- Division of Authority enforcement
Secrets
- No secrets stored in application code
- Integration with your secrets manager (Vault, AWS Secrets Manager)
- Credential rotation support
Access Controls
- Document-level permissions
- Approval workflows with audit trails
- Out-of-office delegation routing
Three phases to operational AI
A proven approach that prioritizes control before capability.
Stabilize Control
Establish boundary, identity, and audit infrastructure. Define what data flows where. Set up logging before AI touches anything.
Automate the Money Path
Deploy 1-2 workflows that directly impact revenue or speed. Proposal generation. Lead routing. Executive briefing automation.
Compound
Expand playbooks. Add memory. Refine governance. Each cycle makes the system smarter and faster.
The compliance and execution clock is ticking.
- CMMC third-party assessment is paused, not repealed — build controlled boundaries now to avoid painful retrofits when it resumes.
- Shadow AI is already inside most organizations. Unmanaged tools create governance blind spots.
- Proposal cycles are compressing. Manual processes can't keep pace.
- Institutional knowledge walks out with every departure.
The organizations that control their AI boundary now will compound advantage. The ones that wait will be retrofitting under pressure.
Aligned with frameworks that matter
Built to support your compliance requirements, not replace your responsibility.
NIST 800-171
Architecture designed to support "control the flow of CUI in accordance with approved authorizations" (3.1.3) through on-premise deployment, audit trails, and access controls.
CMMC 2.0
Deployment model supports organizations pursuing Level 2 certification by keeping CUI within controlled boundaries.
FedRAMP
FedRAMP-aligned architecture. Not currently FedRAMP Authorized. Contact us for deployment options in FedRAMP environments.
Compliance is a shared responsibility. Cabrillo provides architecture and controls; customers are responsible for their own authorization and certification processes.
Compliance Framework Alignment
Platform architecture designed to address the frameworks that matter most to defense contractors.
Architecture aligned with listed frameworks. See compliance roadmap for certification status.
Compliance Roadmap
We hold ourselves to the standard we implement for you. Here is exactly where our own certifications stand — stated plainly, because with third-party CMMC verification suspended, the accuracy of a self-attestation is the whole game.
Cabrillo software: not FedRAMP authorized
FedRAMP
We deploy inside your accreditation boundary, on AWS GovCloud (US) — itself FedRAMP High authorized — so deployments inherit its physical and environmental controls. That is inheritance, not authorization of our software: Cabrillo holds no FedRAMP authorization of its own. For an in-boundary deployment that is the point. Your data stays in your environment, so our authorization is not what your CUI rests on.
Not yet certified
SOC 2 Type 1
We completed an internal readiness assessment against all five trust service criteria in May 2026 and are working the gaps it identified. Type 1 is the current target. We are not SOC 2 certified and will say so until a report exists.
Not CMMC-certified
CMMC Level 2
We run the same program we build for clients: a documented System Security Plan, all 110 NIST SP 800-171 Rev 2 controls enumerated in our own compliance system with the 87 applicable to our enclave tracked continuously, POA&Ms where remediation is outstanding, and a self-assessment in progress. We are not CMMC-certified. We are also not a C3PAO and will not become one — that is deliberate: it is what lets us prepare a contractor without ever competing for their assessment.
Ready to see how it works in your environment?
25 minutes. You'll leave with an automation map and boundary recommendation.
Built for defense contractors. Cleared workforce. Your CUI never leaves your environment. CMMC 2.0 ready.