PLAYBOOK · GOVERNMENT & DEFENSE
Air-gapped document classification and triage.
In defense and sovereign government settings, no document leaves the perimeter — yet classification, redaction, and routing still need to happen at speed. This playbook composes on-prem SLMs, intent templates, and human-in-the-loop review into one auditable network.
Air-gapped environments do not get the luxury of "try a hosted assistant first". Every model has to run inside the boundary, every routing decision has to be explainable, and every reviewer has to remain in the loop. VDF AI was architected for this constraint from day one.
The problem
Sovereignty is non-negotiable
Defense and government workloads cannot ship documents to public clouds. Yet classification queues are massive, and human throughput alone is insufficient. The right balance is automation that augments — never replaces — the cleared reviewer.
The VDF AI approach
Local models, strict policies, transparent runs
VDF AI runs entirely on customer hardware, supports local SLMs, and exposes every routing decision. Domain-level policy enforces who can see which document and which model can process it.
WHY THIS MATTERS NOW
Sovereignty is the requirement, not a feature
In defense and sovereign-government settings, the question is not "which assistant is best" — it is "which assistant can run with no external dependency". Most enterprise AI vendors fail that test by design.
VDF AI runs entirely on customer hardware, supports air-gapped operation, and lets you register local small language models alongside any cloud model your environment permits. The five-step triage Network — intake, classification, redaction, routing, review — operates without a single external call.
WHAT YOU NEED TO START
Prerequisites for a pilot
Environment
- Air-gapped or sovereign cloud
- On-prem GPU for SLMs
- pgvector-capable Postgres
- Hardened image registry
Policy
- Marking and classification standards
- Redaction rules per domain
- Routing matrix for cleared reviewers
- Retention schedule
People
- Cleared reviewers per domain
- One model governance lead
- One security engineer
- One mission owner
REFERENCE ARCHITECTURE
Documents move, not data
OCR · metadata extraction
Local SLM
Domain rules · marking conventions
Domain-scoped policy
PLAYBOOK · STEP BY STEP
Stand up an air-gapped triage line
Deploy VDF AI fully on-prem
Air-gapped, customer-hosted SLMs, no mandatory cloud dependency. Use VDF AI Compliance to define the security envelope.
OCR and metadata extraction
Use the built-in ocr MCP tool plus custom extractors to normalize scanned and digital documents.
Author classification and redaction agents
Each agent uses a tight, deterministic system prompt and outputs structured JSON: marking, confidence, justification.
Enforce domain policy
Domains scope which agents and models can see which document. AgentsHub enforces role-based access at routing time.
Human-in-the-loop review
Low-confidence cases route to cleared reviewers. Every model decision is replayable.
OUTCOMES
Throughput up, sovereignty intact
data leaves the perimeter — air-gapped by design.
throughput on routine classifications.
decisions replayable for inspector general or audit reviews.
SEEMR REFERENCE
Routing inside the boundary
SEEMR's energy and capability modes are doubly valuable when every model runs locally. SEEMR matches model size to task difficulty without leaving the network.
FREQUENTLY ASKED QUESTIONS
What teams ask before shipping this playbook
Can VDF AI run completely air-gapped?
Yes. Bundle distribution is supported via tarball; once installed, no outbound network calls are required.
Which models work locally?
Open-source models served via Ollama, vLLM, or your in-house runtime. Customers also bring fine-tuned models trained on cleared data.
How are sensitive markings preserved?
Every chunk and tool call carries marking metadata. Agents enforce the marking through structured outputs; the audit log captures both input and output markings.
What happens at low confidence?
Low-confidence classifications route to cleared human reviewers. The review action becomes a signal SEEMR uses for the next case.
Is there cross-domain risk?
No. Domain isolation in AgentsHub prevents an agent or tool from straddling marking boundaries. Cross-domain handoffs require explicit escalation.
How long does a sovereign deployment take?
Eight to sixteen weeks depending on accreditation overhead. The platform install itself is days.
RELATED PLAYBOOKS
Continue with related VDF AI patterns
GET IN TOUCH
You Have Questions
Tell us what you’re trying to achieve—governed AI Networks, enterprise RAG, deep integrations, or on‑premise deployment. We’ll help you map the right architecture, security posture, and rollout path. If you’re moving beyond AI pilots and need scalable, auditable execution, reach out—our team is ready to help.