Defense AI Appliance

Frontier-grade AI, without CUI leaving the facility.

The productivity of generative AI — for proposals, past performance, and program data that can't touch a public cloud.

DPLYD puts a fully managed AI appliance inside your perimeter. Controlled Unclassified Information stays on your network — and so does every index, embedding, and log derived from it.

DPLYD for defense is a managed on-premises AI appliance for contractors handling CUI: BD search, proposal drafting, and program knowledge retrieval run entirely inside your boundary, aligned with your CMMC 2.0 Level 2 posture.

Book a Demo Talk to Sales
▲ ABOVE · THE CLOUD GovCloud / third-party API
◇ frontier LLM ◇ public cloud ◇ shared tenancy
≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿≈∿
WATERLINE
↑ your facility's boundary ↑
DPLYD Appliance ON-PREM
▾ YOUR PROGRAM DATA — STAYS HOME
CUI proposals past performance teaming data

"What's beneath the waterline is your business."

BUILT TO THE STANDARD OF CMMC 2.0 L2 NIST 800-171
The Boundary

Your assessors drew the boundary. Cloud AI erases it.

NIST 800-171's data-flow controls — 3.1.x access enforcement, 3.13.x boundary protection — exist to keep CUI inside a defined, assessed enclave. Every cloud AI tool asks you to ship that same CUI to a server outside it.

That's not a convenience question. It's a scope question. If the assessment boundary doesn't cover the AI vendor's cloud, the CUI shouldn't be reaching it — full stop.

You built your compliance posture around a defined enclave. That instinct was right. AI shouldn't be the exception that punches a hole in it.

The part nobody tells you

Keeping the file local isn't enough. The index is the leak.

Most "on-prem-friendly" defense AI keeps the document local — then ships an index of it to the cloud. That distinction is marketing, not a control boundary.

An index derived from your proposal file is a second copy of the sensitive part:

[¶]
CHUNKS

Verbatim passages lifted straight out of a proposal or program document. Not a summary. The text itself.

[≈]
EMBEDDINGS

Encode the meaning, and are reversible enough to reconstruct the source.

[#]
KEYWORDS

Program names, teaming partners, contract numbers, proposal themes — in plaintext.

[⊟]
METADATA

Authors, dates, solicitation numbers, file paths: a map of every program you're pursuing.

[⇄]
RELATIONSHIPS

Which teaming partners, primes, and programs connect. Disclosure without a single sentence leaving.

The original file never moved. The CUI did.

A cloud index of local CUI is outside the boundary the moment it can disclose, infer, search, or reconstruct protected material off your enclave. DPLYD keeps the index on the same side of the boundary as the file. Always.

The Approach

Bring the AI to the data — not the reverse.

DPLYD is a managed AI appliance that lives inside your assessed boundary. The model runs where your program data already is.

[⊡]

Inside your boundary

CUI never crosses your enclave. Full generative capability, zero egress of controlled data.

[↻]

Fully managed

We own the hardware, updates, and lifecycle. Your team operates; we maintain. No new servers for your IT to babysit.

[≡]

Predictable economics

Fixed monthly. No per-seat licensing. No per-token metering. No capex. Every BD, capture, and proposal staffer works from the same appliance — price doesn't climb because you gave the whole team access.

The Workload

General-purpose intelligence, with BD and capture where it counts.

DPLYD isn't a narrow point-tool. The appliance runs capable general models — the same drafting, summarizing, and reasoning your team would reach for anywhere — plus program-specific workflows on top.

Proposal & BD drafting on 20 years of past performance Program knowledge search across contract documents and prior submissions Compliance-aware summarization of DFARS clauses and solicitation packages Teaming and capture research grounded in your program history, not the open internet

One appliance serves the whole capture and delivery organization. Program work gets program handling; everything else your people do with AI runs on the same secure box.

The Record

Audit-ready by construction.

Compliance shouldn't be a slide deck you assemble the week before a C3PAO assessment. DPLYD produces an immutable, exportable record of every AI interaction: what ran, on which documents, under what policy, with what result.

When an assessor asks for evidence — or when a program audit surfaces two years later — the record already exists.

"trust us" ──▶ evidence on demand.
SENTINEL · AUDIT LOG immutable
cui_flag  bound
access_scope  verified
enclave_boundary  enforced
control_family  3.1 / 3.13
▸ export evidence 
The Comparison

DPLYD vs. GovCloud SaaS AI.

GovCloud SaaS AI DPLYD
Where the model runs Vendor's FedRAMP/GovCloud tenancy Inside your facility
Where the index lives Vendor cloud, outside your enclave Same side of the boundary as the data
ATO / assessment surface A new external system to scope and trust An asset inside your existing boundary
Evidence at assessment Vendor attestations and contract language Immutable, exportable record on demand
Pricing Per-seat / per-token, scales with usage Fixed monthly — no per-seat, no per-token
Disconnected / air-gapped work Not possible — requires connectivity Supported by design

GovCloud SaaS AI is still someone else's cloud — inside the FedRAMP boundary, but outside yours. DPLYD refuses that trade: the intelligence comes to the data.

Who It's For

Built for the contractors cloud AI can't serve.

Some work simply cannot go to a third-party API: CUI under NIST 800-171, programs preparing for CMMC 2.0 Level 2 assessment, export-controlled technical data, and any contractor that treats its assessment boundary as non-negotiable.

If your answer to "where does the data go?" has to be nowhere outside the enclaveDPLYD is the only version of this that works.

Field Results

Proof from the field.

DEFENSE & AEROSPACE

AI Infrastructure for a Defense Contractor Preparing for CMMC 2.0

Deploy generative AI on 20 years of BD content without violating CMMC Level 2 posture or exposing controlled data to frontier models.

Read the case study →
FAQ

Straight answers.

Q ▸ Can we use generative AI on CUI at all?

Yes, if the entire processing chain stays inside your assessed boundary. That is the design constraint DPLYD is built around: model, index, embeddings, and logs all remain on your network.

Q ▸ Does DPLYD expand our CMMC assessment scope?

The appliance is deployed inside your existing boundary and inherits your enclave's controls; Sentinel produces the audit evidence assessors ask for. It is scoped as an asset within the boundary, not a new external service.

Q ▸ What about the search index — isn't that where tools leak?

Most "compliant" AI tools keep the file local but ship an index to their cloud. An index is derived CUI. DPLYD keeps the index on the same side of the boundary as the data.

Q ▸ Is it air-gapped?

The appliance can run fully disconnected. Managed updates are delivered on a schedule and mechanism that fits your environment, including offline transfer for air-gapped enclaves.

See it on your own program data.

A 30-minute walkthrough of the appliance, running on your data, inside your boundary.

▸ BOOK A DEMO
Book a Demo