# Optimal, LLC

> Optimal builds, migrates, and manages CMMC-ready Microsoft 365 GCC High
> enclaves for the United States defense industrial base — the sovereign
> environment a contractor needs to handle Controlled Unclassified Information
> (CUI) and ITAR-controlled data on DoD contracts — and stands up the governed
> AI that runs safely inside them. Optimal is an SBA-certified Service-Disabled
> Veteran-Owned Small Business (SDVOSB). Advisory-led, not a SaaS product.
> Sovereign by default.

## About

- **Legal name:** Optimal, LLC
- **Website:** https://gooptimal.io
- **Contact:** ryan@gooptimal.io
- **Business model:** Advisory- and services-led engagements. Optimal delivers
  scoped engagements and managed operations, not a software subscription.
- **Where Optimal works:** Microsoft 365 GCC High, Azure Government, commercial
  cloud (Azure, AWS, GCP), on-premises, and fully air-gapped networks.

## Leadership

- **Ryan Gutwein** — Founder & Chief Executive Officer. Service-disabled
  veteran. Scoping calls and engagement delivery are led by the founder
  directly; there is no layer between the sales conversation and the
  engineering. Profile: https://gooptimal.io/team

## Certifications & set-aside eligibility

- **Service-Disabled Veteran-Owned Small Business (SDVOSB)** — SBA-certified.
- **Veteran-Owned Small Business (VOSB)** — SBA-certified.
- Certified through the U.S. Small Business Administration (SBA) Veteran Small
  Business Certification (VetCert) program; publicly listed on the SBA
  certification registry at https://search.certifications.sba.gov/.
- Eligible to compete for **SDVOSB and VOSB set-aside contracts**.

## What Optimal does — two pillars

Optimal does two things that are designed to work together: it builds and runs
the compliant environment, and it makes frontier AI genuinely usable inside
that environment.

### Pillar 1 — Managed Microsoft 365 GCC High enclaves (for the defense industrial base)

- **Build the enclave.** A compliant Microsoft 365 GCC High tenant — Entra ID,
  Conditional Access, Defender, and Purview configured to a CUI-ready baseline
  from day one.
- **Migrate cleanly.** Mailboxes, files, Teams, and identities moved from
  Microsoft 365 Commercial or GCC into GCC High with spillage controls, so no
  CUI is left in the wrong boundary.
- **Engineer the controls.** The NIST SP 800-171 control set implemented and
  documented — System Security Plan (SSP) and POA&M, evidence collected — ready
  for a third-party CMMC Level 2 assessment.
- **Run it.** Ongoing administration, patching, monitoring, user lifecycle,
  security operations, and help desk. Optimal also sources and provisions GCC
  High G5 licensing.
- **Aligned to:** DFARS 252.204-7012, NIST SP 800-171, CMMC Level 2. Supports
  CUI and ITAR / export-controlled data.

### Pillar 2 — Governed AI and AI security

- **Governed AI Engineering** (signature engagement) — a Zero Trust access layer
  for AI in the cloud the client already licenses: one authenticated gateway to
  every approved model, per-key virtual credentials and rate limits, fail-closed
  prompt and response guardrails, default-deny egress pinned to an allowlist,
  and a request-ID-joined audit trail reviewers read directly. Self-hostable and
  client-owned; no SDK rewrite, no new accreditation boundary.
- **AI Security Assessment** — structured evaluation of AI systems, pipelines,
  and integrations against real adversary attack paths. Covers prompt trust
  boundaries, retrieval/RAG pipelines, tool- and function-call authorization,
  output handling, and secret/PII/training-data exposure. Findings mapped to the
  OWASP LLM Top 10 and MITRE ATLAS.
- **Adversarial AI Testing (red team)** — direct and indirect prompt injection,
  jailbreaks and system-prompt extraction, model and context manipulation, data
  and model exfiltration, and agent abuse / excessive-agency chains, with
  reproducible attack chains and the control that closes each one.
- **Secure AI Adoption Advisory** — usage patterns, threat models, and
  secure-adoption roadmaps scoped to the organization's risk profile.

## Scope boundary — what Optimal does not do

This matters for accuracy, and Optimal states it publicly:

- Optimal is **not a C3PAO** and does not perform certified CMMC assessments.
- Optimal **does not grant an Authority to Operate (ATO)** and does not issue
  certifications.
- **No environment makes an organization CMMC compliant on its own.** CMMC
  Level 2 is a third-party assessment against the 110 NIST SP 800-171 controls.
  Optimal engineers the environment to be assessment-ready and prepares the
  evidence; the certified assessment is performed independently by an accredited
  C3PAO. Building the environment and grading it are deliberately separate
  roles.

## Bring your own cloud

If a client already licenses Microsoft Azure, AWS, or Google Cloud, Optimal
builds governed, audited access to the frontier models in that existing cloud —
Azure OpenAI Service (Azure and Azure Government), Amazon Bedrock (AWS), and
Google Vertex AI (GCP) — plus self-hosted models. No new vendor or procurement
required. Available to the defense industrial base, federal agencies, SLED
(state, local, and education), and enterprises.

## Reference pattern (Governed AI Engineering)

- Authenticated gateway — one governed endpoint to every approved model, per-key
  virtual credentials (LiteLLM), no SDK rewrite. Caller auth via Cloudflare
  Access and Okta OIDC.
- Prompt guardrails — NeMo Guardrails, fail-closed and pre-call; unit-tested
  secret and PII detectors.
- Default-deny egress — Squid allowlist proxy plus security-group enforcement;
  no configuration opt-out.
- Audit trail — structured JSON joined to Postgres by request ID; defensible
  without vendor-UI screenshots.

## Reference implementations (public, inspectable)

- **AI-Gateway** — open-source Zero Trust AI gateway reference design, the same
  pattern Optimal stands up in client environments:
  https://github.com/optimal-cyber/AI-Gateway
- **NINELINE** — working reference implementation of AI-native application and
  supply-chain security: real SBOM and vulnerability scanning, EPSS and CISA KEV
  enrichment, and frontier-model reachability analysis that compressed 874 raw
  findings to the 18 that required action in the reference scan. Overview:
  https://gooptimal.io/nineline — live demonstrator: https://9line.gooptimal.io/

## Standards (engineering targets)

DFARS 252.204-7012, NIST SP 800-171, CMMC Level 2, NIST AI RMF, OWASP LLM Top
10, MITRE ATLAS, NIST 800-207 Zero Trust, CIS Benchmarks, DISA STIG, ISO 42001,
SOC 2.

## Common questions

**What does Optimal do?**
Two things that work together: it builds, migrates, and manages CMMC-ready
Microsoft 365 GCC High enclaves for the defense industrial base, and it stands
up the governed AI that runs safely inside those environments — including
assessment and red-teaming of the AI systems teams deploy.

**What is Microsoft 365 GCC High, and who needs it?**
GCC High is a US-sovereign version of Microsoft 365, physically and logically
isolated, with US data residency and screened US-person support. A defense
contractor handling CUI or ITAR/export-controlled data generally needs it to
support the DFARS 252.204-7012 and CMMC Level 2 obligations in its contracts.

**Does moving to GCC High make an organization CMMC compliant?**
No. GCC High is a foundation built for the controls, not the certification.
CMMC Level 2 is a third-party assessment against the 110 NIST SP 800-171
controls, performed by an accredited C3PAO. Optimal implements and documents the
control set, collects evidence, and gets the environment assessment-ready.

**Is Optimal a product company or a services firm?**
Services. Optimal delivers engagements and managed operations, not a SaaS
subscription. The control pattern it implements is published as an open-source
reference architecture, so a client can read exactly what will be built and keep
operating it independently afterward.

**What is prompt injection, and why does it matter?**
Prompt injection manipulates an AI system through crafted input, causing it to
ignore instructions, expose data, or take unauthorized actions. It is among the
most active threats to any organization deploying large language models and
agents, and it is one of the first things Optimal tests.

**Can Optimal work with federal, SLED, and enterprise teams?**
Yes. Optimal is an SBA-certified SDVOSB eligible for SDVOSB and VOSB set-aside
contracts, and works with the defense industrial base, federal agencies, SLED
entities, and enterprises across GovCloud, commercial cloud, on-premises, and
air-gapped networks.

## Key pages

- Home: https://gooptimal.io/
- Managed GCC High / CMMC: https://gooptimal.io/gcc-high
- NINELINE reference implementation: https://gooptimal.io/nineline
- Team / leadership: https://gooptimal.io/team
- Field notes (blog): https://gooptimal.io/blog
- Contact / book a scoping call: https://gooptimal.io/contact
- Use case — Regulated cloud: https://gooptimal.io/use-cases/regulated-cloud
- Use case — On-prem & air-gapped: https://gooptimal.io/use-cases/air-gapped
- Use case — Agentic workloads: https://gooptimal.io/use-cases/agentic
- Sitemap: https://gooptimal.io/sitemap-index.xml