Air-Gapped Deployment

Air-Gapped Enterprise Chatbot

An enterprise chatbot is a company-wide AI assistant — a ChatGPT-class experience connected to internal knowledge, governed by role-based access, and safe for employees to use with real work data, operating on a network with no connection to the public internet — models, updates, and telemetry all move by controlled offline transfer, so the system functions fully inside a classified or isolated enclave.

typical cost advantage vs per-seat AI at 1,000+ users
100%of chats inside your perimeter
65+AI & governance topics it can be grounded in
0chats used for vendor model training
Built for regulated deployment Deploys on your infrastructureNo data leaves your boundaryFull audit trailSSO & role-based access
The air-gapped enterprise chatbot decision

Personnel in classified and OT environments do the same drafting, summarizing, and searching as everyone else — with no AI allowed. An air-gapped chatbot brings ChatGPT-class assistance into the enclave itself, grounded in in-enclave knowledge, so secure-environment staff stop being the last workforce without AI leverage.

An air-gapped chatbot is the most visible AI system you will deploy in a classified environment, and visibility cuts both ways. Done well, it is the tool that finally gives cleared staff the assistance their unclassified colleagues have had for years. Done badly, it is a slow, thin imitation that people try twice and abandon — and in an enclave there is no shadow alternative to fall back on, so the gap simply stays unfilled.

That raises the quality bar rather than lowering it. Outside the gap, an adequate internal tool competes with ChatGPT and loses gracefully. Inside, it competes with nothing, which means users judge it against their memory of public tools and their patience is short. Latency, streaming behaviour, and answer grounding matter more here than feature breadth.

The distinctive design problem is classification handling in conversation. A chat interface invites users to paste content, and in a multi-level environment the system must be unambiguous about the classification of what it holds, what it retrieves, and what it produces. Marking conventions, retention rules, and per-user scoping have to be decided before rollout, because retrofitting them into an existing conversation store is genuinely hard.

Air-Gapped by design

Why teams run their enterprise chatbot air-gapped

Built for defense, intelligence, critical-infrastructure and classified-environment teams.

01

Zero external connectivity, by design

An air-gapped enterprise chatbot makes no outbound calls — no license pings, no telemetry, no model API fallbacks. If a component phones home, it fails certification; the architecture must assume the internet does not exist.

02

Built for classified and SCIF environments

Defense, intelligence, and critical-infrastructure operators need AI capability where cloud AI is categorically prohibited. The enterprise chatbot runs entirely on enclave hardware and clears accreditation reviews because there is nothing external to assess.

03

Controlled update path

Models, embeddings, and software updates arrive as signed offline bundles through your cross-domain transfer process — the same discipline you already apply to any software entering the enclave.

What it does

Core capabilities of an enterprise enterprise chatbot

ChatGPT-class experience

Chat, documents, code, and images in one interface employees actually adopt — no capability downgrade versus consumer tools.

Grounded in company knowledge

Answers draw on your wikis, policies, and documents through private RAG, with citations.

Role-based governance

Who can use which models, tools, and knowledge bases is policy, enforced centrally with full audit.

Multi-model backend

Conversations route across local and permitted models by task, invisibly to users.

Architecture

What an air-gapped deployment changes

Everything ships as a self-contained bundle: container images, model weights, embedding models, and documentation must install from local media with no registry or CDN access.

Local models only: the enterprise chatbot serves open-weight models on enclave GPUs; there is no cloud fallback tier, so model selection and routing happen entirely inside the gap.

Audit evidence must be exportable on your terms — logs stay in the enclave and leave only through your controlled review process.

Reference stack

The air-gapped enterprise chatbot stack

Air-gapped chatbot, layer by layer — with the reason each choice holds up under this deployment mode.

Layer Typical choice Why, here
Chat interface Self-hosted web UI served from inside the enclave No CDN for assets, no external fonts, no analytics beacons — all common defaults that fail here and are easy to miss until first load.
Model serving Local open-weight model with streaming enabled Streaming tokens is a perception fix, not a throughput one, and it is the single biggest lever on whether an enclave chatbot feels usable.
Grounding Private retrieval over enclave document stores Ungrounded chat in a classified setting produces confident, unciteable answers — the fastest way to lose analyst trust permanently.
Identity & scoping Enclave SSO with per-user clearance-aware retrieval scope Two users asking the same question must get different grounding if their access differs. This has to be enforced at retrieval, not in the prompt.
Conversation store In-enclave database with classification labels and retention rules Conversation history inherits the classification of its content. Treat it as a records system, not a cache.
Capacity planning

Sizing an air-gapped enterprise chatbot

Profile Scale Hardware What actually binds
Team pilot 50–150 cleared users One 48 GB GPU serving a mid-size model Peak concurrency is bursty and tied to the working day; average utilisation badly understates what you need.
Site-wide 500–2,000 users on one enclave 2–4 GPUs with a small model absorbing routine turns Time-to-first-token, not tokens per second, is what users experience as speed. Optimise for it.
Multi-level Separate deployments per classification domain Independent stack per domain; no shared model or store The requirement is separation, so this is deliberately not one deployment with tenant flags.
Compliance drivers

Regulations that point to air-gapped

Classified handling

The enterprise chatbot operates inside SCIF and enclave boundaries; there is nothing to accredit outside them.

ITAR / export control

Technical data entering the enterprise chatbot never transits foreign-controlled infrastructure.

NIS2 / NERC CIP

Critical-infrastructure isolation is met structurally by the enterprise chatbot, not contractually.

Zero-trust postures

No third-party endpoints to allow-list — the enterprise chatbot’s attack surface is your own network.

Honest fit check

When air-gapped is the right call — and when it isn’t

Choose air-gapped when

  • The network the enterprise chatbot must serve is already isolated — classified programs, OT networks, offline research enclaves.
  • Policy prohibits any external AI API, including via proxy or private link.
  • You need AI capability in disconnected field or vessel environments with intermittent or no connectivity.

Consider another mode when

  • You can tolerate controlled outbound connectivity → an on-premises enterprise chatbot is simpler to operate and update.
  • Your requirement is legal jurisdiction rather than physical isolation → the sovereign variant fits; air-gapping an enterprise chatbot is stricter than most regulators ask.
Deployment modes compared

Enterprise Chatbot: Air-Gapped vs the alternatives

Deployment mode Typical owner What you gain — and give up
Air-Gapped (this page) CISO / Classified Program Lead Structural security no contract can match — in exchange for moving every model, index, and software update through an offline bundle process.
On-Premises CTO / Head of Infrastructure Maximum physical control and the strongest economics at steady volume — in exchange for owning the hardware, the capacity plan, and the upgrade cycle.
Self-Hosted Platform Engineering Lead Complete stack and model freedom with no usage meter — in exchange for your team owning operations, CVE response, and the upgrade cadence.
Private CISO / Data Protection Officer The fastest route to confidential AI — in exchange for a boundary defined by tenancy and contract rather than by a building you own.
Rollout

How to deploy an air-gapped enterprise chatbot

  1. 01

    Decide classification handling first

    Agree how conversations are marked, how long they are retained, and what happens to content pasted above the system's accreditation level. These rules shape the data model, so deciding them after launch means a migration.

  2. 02

    Bundle the full stack including front-end assets

    Package the UI, model, retrieval layer, and every static asset. Web front-ends are the most common source of accidental external calls — fonts, icon CDNs, and telemetry all need to be stripped and verified offline.

  3. 03

    Wire clearance-aware retrieval before opening access

    Bind retrieval scope to each user's identity and clearance, then test with accounts at different levels to confirm grounding actually differs. Prompt-level instructions are not an access control.

  4. 04

    Pilot with a group that has a real daily need

    Choose users whose work is document-heavy and whose feedback you can act on quickly. Broad early access to a rough system in an enclave burns the goodwill you need for the real rollout.

  5. 05

    Instrument quality locally and iterate

    Collect thumbs-up/down and unanswered-question logs inside the enclave and review them weekly. With no external analytics, a deliberate local feedback loop is the only signal you will get.

Failure modes

Where air-gapped enterprise chatbot projects fail

Front-end assets that reach outside

A web font, icon set, or error-reporting script embedded in the UI fails silently or blocks load inside the gap. It is nearly always discovered on the enclave, at the worst possible moment.

Clearance enforced in the prompt

Telling the model not to reveal certain material is not access control. If a passage was retrieved, it was disclosed to the model and can surface in an answer. Filter at retrieval.

Launching without grounding

A chatbot that cannot cite enclave documents produces plausible answers analysts cannot verify. Trust lost this way rarely comes back, even after retrieval is added later.

Buyer checklist

How to evaluate an air-gapped enterprise chatbot

1

Is the experience good enough that employees stop pasting data into public chatbots?

2

Does it answer from your internal knowledge with citations, not just general knowledge?

3

Can admins govern models, tools, and data access per role or department?

4

Where do conversation logs live, and who can read them?

5

What does it cost at full-company rollout versus per-seat cloud tools?

Air-gapped deployments trade update convenience for structural security; budget for the offline bundle process, but the enterprise chatbot itself prices like any fixed in-enclave infrastructure — no meters, no per-token exposure.

How VDF AI delivers it

An air-gapped enterprise chatbot, on the VDF AI platform

VDF AI Chat is the private enterprise chatbot: ChatGPT-class UX, private RAG grounding, role-based governance, and flat platform pricing instead of per-seat meters.

FAQ

Air-Gapped Enterprise Chatbot questions, answered

What is an air-gapped enterprise chatbot?

An enterprise chatbot is a company-wide AI assistant — a ChatGPT-class experience connected to internal knowledge, governed by role-based access, and safe for employees to use with real work data, operating on a network with no connection to the public internet — models, updates, and telemetry all move by controlled offline transfer, so the system functions fully inside a classified or isolated enclave.

Why do enterprises choose an air-gapped enterprise chatbot over a cloud service?

An air-gapped enterprise chatbot makes no outbound calls — no license pings, no telemetry, no model API fallbacks. If a component phones home, it fails certification; the architecture must assume the internet does not exist. Air-gapped deployments trade update convenience for structural security; budget for the offline bundle process, but the enterprise chatbot itself prices like any fixed in-enclave infrastructure — no meters, no per-token exposure.

Which regulations drive air-gapped enterprise chatbot adoption?

The most common drivers are Classified handling, ITAR / export control, NIS2 / NERC CIP, Zero-trust postures. Classified handling: The enterprise chatbot operates inside SCIF and enclave boundaries; there is nothing to accredit outside them.

Can VDF AI run as an air-gapped enterprise chatbot?

Yes. VDF AI Chat is the private enterprise chatbot: ChatGPT-class UX, private RAG grounding, role-based governance, and flat platform pricing instead of per-seat meters. VDF AI Chat installs entirely inside the enclave — interface, model, and retrieval — with retrieval scoped per user identity, so two analysts asking the same question are grounded only in what each is cleared to see.

Can you run a ChatGPT-style assistant on a classified network?

Yes. The interface, model, and retrieval layer all run from local images on enclave hardware with no external dependency. The engineering work is in removing the assumptions a normal web application makes — CDN assets, telemetry, licence checks — and in enforcing clearance-aware retrieval so different users are grounded in different documents.

How do you stop an air-gapped chatbot showing users documents they are not cleared for?

Scope retrieval to the user's identity and clearance so restricted passages are never retrieved in the first place. Filtering after generation, or instructing the model to withhold material, both fail: once a passage is in the model's context it has already been disclosed and can leak through paraphrase.

How many users can one air-gapped chatbot deployment support?

A single 48 GB GPU comfortably serves a team of roughly 50–150 with a mid-size model, because chat traffic is bursty and mostly short. Scaling to a site of a thousand or more usually means adding a small model to absorb routine turns rather than simply adding large-model capacity.

Does an air-gapped chatbot need RAG?

In practice, yes. Without retrieval the assistant can only draw on what the model memorised during training, which contains none of your enclave's material and cannot be cited. Grounding in local documents is what turns it from a writing aid into something analysts can actually verify and rely on.

On-Prem AI

Plan your on-prem AI deployment

Book an architecture call and we will scope a private, on-prem AI deployment for your environment — integrations, hardware, and governance included.

View the deployment roadmap