AirgapAI Technical Architecture & Security Overview
Comprehensive technical documentation for enterprise architects evaluating AirgapAI's air-gapped AI/LLM solution.
Document Revision History
| Rev# | Date | Purpose | Pages Revised |
|---|---|---|---|
| 4 | 2025-9-9 | Added details on Federal Government use | Section 3 |
| 3 | 2025-7-8 | Added additional security enhancements | All |
| 2 | 2025-5-21 | Enhanced Details around Upcoming Blockify Options | All |
| 1 | 2024-12-10 | First Full Release | All |
Executive Summary
AirgapAI offers a highly secure, air-gapped AI/LLM solution designed for organizations with stringent security and compliance requirements. This platform operates entirely locally on endpoints, ensuring all data processing, storage, and queries remain on the device with no outbound connections or cloud dependencies.
Uncompromised Data Security
All user data and processes are isolated and stored locally, eliminating data exfiltration risks and meeting the highest security standards (e.g., USAF, DOD, Intelligence).
Air-Gapped by Design
The system is engineered for complete isolation, making it ideal for handling classified data and operating in sensitive, high-security environments, or in remote field locations with limited connectivity.
Flexible Deployment & Management
Supports various device types (x86, ARM, VDI), offering fully local installation with centralized or decentralized updates. User profiles and data can be managed via familiar methods like Intune or network shares.
Granular Access Control
Patented "Blockify" data pipeline enables fine-grained permissioning and exclusion-based access to "IdeaBlocks," ensuring data is accessible only to authorized users.
Hardware Agnostic & Scalable
Supports discrete GPUs and NPUs, allowing for performance scaling and compatibility with diverse hardware and model sizes.
Single-Tenancy
Each device operates as a single-user, single-tenant environment, further isolating application stacks and preventing cross-device data transmission.
Controlled Updates
Model updates require human/admin approval, providing an additional layer of security control.
No Telemetry or Keylogging
Ensures complete privacy and prevents any outbound data transmission.
System Technical Architecture
1.1 Devices, Models, and Deployment
Device Support
- Workstations, AI PCs (Laptops and Desktops), and other endpoints
- Supports x86 (Intel, AMD) and ARM (Snapdragon); silicon-agnostic
- Virtual Desktop Infrastructure (VDI) with GPUs only
Deployment Model
- Fully local (air-gapped) installation, keeping all data/processes on endpoint
- No network connectivity required by default; designed for strict, high-security settings
- Centralized/Decentralized Updates:
- Delivered via Intune/other image management, network file share, or as part of new device image
- User profiles configured via JSON files saved in %APPDATA%
- Automatic updates, if enabled, can be served from Iternal systems (PoC phase only) or a simple customer-hosted server
Hardware Acceleration & Model Sizing
- Supports discrete GPUs and NPUs for performance scaling
- Model sizes can vary (1B, 7B, 8B, 22B, etc.) to match hardware capabilities
- Parallel inference execution support available
Data Ingestion & Delivery
- Datasets deployed as JSONL files via device imaging, network shares, or direct upload
1.2 Data Flow and Management
Blockify Data Pipeline
- Patented pipeline ingests, deduplicates, and distills data (marketing, tech docs, PDFs, PPTs) into modular "IdeaBlocks"
- Enhanced LLM accuracy with fine-grained permissioning
- 100% Local on client device Blockify available in August 2025 (license included)
- Datacenter Blockify LLM available today (separate license)
Ingestion Workflows
- Can occur on-prem (data center w/ GPU) or locally (device)
- Supports both central and device-based ingestion/updates
- Updates via network shares, Intune/similar, or direct upload
1.3 Application Stack
Technology
- JavaScript / React / Electron, cross-platform, user-focused workflows
Solution Packaging
- AirgapAI, Blockify, and tools are available as pre-built SKUs
Model Support
- Customer selected Public Open Source LLMs (Llama, Gemma, Mistral, etc.) and custom quantized models based on standard architectures
- Users can customize workflows before pushing to local devices via JSON config files and your standard file push process via MS Intune
1.4 Storage Architecture
Local Storage
- All user data is locally stored in user profiles in a levels.js database (chats, settings, benchmark data) or as files (models, corpuses)
- No off-device transmission
- Levels.js is an open-source technology, so IT can "pull data" from the device but that would be a completely separate process from our application
Data Format
- JSON files in user-controlled directories define user configuration
- Update your Blockified Dataset and app at any frequency to ensure data accuracy
Security Architecture & Controls
2.1 Air-Gapped Security Model
Complete Isolation
- AirgapAI runs 100% locally; all processing, storage, and queries stay on the device
- No outbound connections, "call-home," or cloud dependencies required for installation, updates, or maintenance
- An optional customer-deployed management service is under design to enable easier administration
- Compliant with highest security environments (USAF, DOD, Intelligence, CBP, etc.)
- Suitable for classified data (defense strategies, nuclear protocols)
Administration
- Device-level management and per-IdeaBlocks defined permissioning
- Customer can build their own dataset filtering to enable exclusion-based access via AND / OR filters (i.e. Venn-diagram) matrix before provisioning IdeaBlocks to the device
Enterprise Deployments
- Data Security is driven by device protocols / encryption / userlogins, etc
LAN Mode (Future Capability Forthcoming)
- "LAN mode" allows connection to GPU servers for secure, enterprise compute sharing, maintaining endpoint data integrity
2.2 Device and User Security
User Isolation
- Each device is single-user, single-tenant (no multi-user mode)
- No cross-device data transmission
- Every user on a shared system has their own datasets and no access to other users data
Imaging and Provisioning
- Integrates into existing secure imaging pipeline: extraction, validation, pre-shipment hardening as a simple installable .EXE application
- All installable EXEs are digitally signed with certificates stored/managed on FIPS 140-2 compliant HSM
Controlled Updates
- All updates, either of the application itself or its components (Models, Corpuses) require human/admin approval
2.3 Selected Security-Related Features
- No network-based integration: Entirely local
- Granular permissioning: Add tags to each IdeaBlock for granular compliance
- On-device model support: No model/data egress
- No telemetry/keylogging: No outbound data of any kind
Federal Readiness: Regulatory Alignment, Scope, and Controls for U.S. Government Use
Summary
AirgapAI is engineered for high-security, air-gapped operation that keeps all data processing and storage within the customer's accredited environment, eliminating reliance on vendor-operated services.
This architecture simplifies scoping for FAR/DFARS/CMMC, enables straightforward RMF/ATO integration, avoids FedRAMP unless optional cloud components are introduced, and supports federal hardening, logging, supply chain, and accessibility requirements without data egress from government control.
Purpose
This addendum describes how AirgapAI supports U.S. Federal Government cybersecurity, compliance, and accreditation needs. It clarifies data-handling scope, cloud applicability, and alignment with key regulations and frameworks (FAR/DFARS/CMMC, NIST, FedRAMP, RMF/ATO, and related federal requirements).
Scope and Data Boundary
- Deployment model: AirgapAI is 100% downloadable software that installs and operates locally on customer-controlled endpoints (workstations, laptops, VDI). There are no vendor-operated services, compute, or remote access required.
- Data handling: All processing, storage, and queries occur on the device. The software does not transmit telemetry, "call home," or require cloud connectivity. By default, no vendor-operated information system processes Federal Contract Information (FCI), Controlled Unclassified Information (CUI), or Covered Defense Information (CDI).
- Optional components: Customers may update software and models via their own tooling (e.g., Intune, network shares, offline media). Any optional centralized components can be customer-hosted inside their accreditation boundary.
Regulatory Alignment and Applicability
| Regulation | Applicability & Support |
|---|---|
| FAR 52.204-21 | Applies to contractor internal systems that process, store, or transmit FCI. In the default AirgapAI deployment, no vendor system handles FCI; the customer's endpoints are the in-scope systems. AirgapAI's offline, single-tenant endpoint design supports customer compliance by eliminating nonessential connectivity and simplifying boundary definition and segmentation. |
| DFARS 252.204-7012 | Applies when contractor internal systems handle CDI/CUI. In the default deployment, no vendor system handles CDI/CUI; the customer's accredited environment is in scope. If any cloud service is used to store/process CUI, DFARS 7012 requires FedRAMP Moderate equivalent. AirgapAI does not require cloud services. |
| DFARS 252.204-7019/7020 | Requires an SPRS score for each contractor system that will handle CUI/CDI. In the default deployment, the customer's environment is the assessed system. No vendor system requires an SPRS score unless it processes CUI. |
| CMMC | CMMC Level 1 (FCI) and Level 2 (CUI) apply to the contractor systems handling those data types. AirgapAI's offline architecture allows customers to confine scope to a tightly segmented enclave or endpoint baseline, reducing inheritance and assessment complexity. |
| NIST RMF/FISMA/SP 800-53 | AirgapAI can be incorporated into a system authorization (ATO) as a component on endpoints within the customer's boundary. The application supports customer implementation of AC, AU, CM, SC, and SI control families by operating offline, honoring OS security baselines, supporting signed updates, and enabling centralized provisioning without external dependencies. |
| FedRAMP | FedRAMP applies to cloud services. AirgapAI, when deployed as local software with no vendor-hosted components, is not a cloud service and does not require FedRAMP authorization. |
| DoD CC SRG | If used by DoD on cloud infrastructure, the deployment must align to the required Impact Level (IL2/IL4/IL5/IL6) consistent with the data category. AirgapAI does not require cloud hosting. |
STIGs and Platform Hardening
- AirgapAI installs and runs on customer-hosted, STIG-hardened operating systems and images without requiring persistent elevated privileges or external network access
- All controls are inherited from the underlying OS and enterprise stack
Cryptography and Integrity
- No proprietary cryptographic algorithms are introduced. Where cryptography is used, the application leverages platform-native, FIPS 140-2/140-3 validated modules made available by the operating system when configured for FIPS mode
- All release and update packages are digitally signed. Signing keys are held in HSMs compliant with FIPS 140-2 or higher
- Customers may rely on host-based full-disk encryption (e.g., BitLocker, FileVault) and OS credential protection to protect at-rest data. The application does not transmit or escrow keys
Logging, Monitoring, and Incident Response
- AirgapAI produces local logs suitable for ingestion by agency EDR/SIEM tools; it does not exfiltrate logs. Customers may define log retention and forwarding per OMB M-21-31 and agency policy
- The vendor maintains an internal incident response program. If any future deployment were to handle FCI/CUI/CDI on vendor systems, incident reporting would follow contractual requirements (e.g., DFARS 7012 timelines and DIBNet submission). In the default offline model, no vendor-held federal data exists
CUI, ITAR, and Export Controls
- The offline, single-tenant design prevents data egress and is suitable for processing CUI within a customer-accredited environment. Data remains in customer custody at all times
- For ITAR/EAR-controlled data, customers remain the exporters of record; AirgapAI does not transmit or store data outside the customer's boundary. Customers should apply applicable device, user, and media controls consistent with their accreditation
- CUI marking, handling, and sanitization procedures remain customer responsibilities within their authorized environment
Accessibility, Privacy, and Records
- Section 508: Software follows 508 Compliant best practices in a commercially reasonable manner
- Privacy: No telemetry, analytics, or keylogging. The application does not collect personal data or transmit any user content
- Records: The application stores data only on the local endpoint under customer control. Retention and disposition are governed by the customer's records schedules
Conclusion
This architecture delivers a robust, air-gapped, and highly secure platform for AI/LLM-enabled workflows. It supports granular, IdeaBlock-level permissioning, advanced compliance needs, and modular/custom deployment models (endpoint, edge, private cloud). With airgapped 100% local and full data/process isolation, the system is engineered for the most demanding classified, regulatory, or privacy-centric organizations.
Frequently Asked Questions
AirgapAI supports workstations, AI PCs (laptops and desktops), and other endpoints. It is silicon-agnostic, supporting x86 (Intel, AMD) and ARM (Snapdragon) architectures. Virtual Desktop Infrastructure (VDI) with GPUs is also supported.
AirgapAI uses fully local (air-gapped) installation, keeping all data and processes on the endpoint. No network connectivity is required by default. Updates can be delivered via Intune/other image management, network file share, or as part of new device images. User profiles are configured via JSON files saved in %APPDATA%.
Blockify is a patented pipeline that ingests, deduplicates, and distills data (marketing, tech docs, PDFs, PPTs) into modular "IdeaBlocks." It provides enhanced LLM accuracy with fine-grained permissioning. 100% local on-device Blockify is available, with Datacenter Blockify LLM also available under separate license.
All user data is locally stored in user profiles using a levels.js database (chats, settings, benchmark data) or as files (models, corpuses). There is no off-device transmission. JSON files in user-controlled directories define user configuration.
AirgapAI is designed to be compliant with the highest security environments including USAF, DOD, Intelligence, and CBP. It supports FAR/DFARS/CMMC requirements, NIST RMF, FISMA, and SP 800-53 controls. The architecture is suitable for classified data handling.
No. FedRAMP applies to cloud services. AirgapAI, when deployed as local software with no vendor-hosted components, is not a cloud service and does not require FedRAMP authorization. If used by DoD on cloud infrastructure, deployment must align to the required Impact Level (IL2/IL4/IL5/IL6).
No proprietary cryptographic algorithms are introduced. The application leverages platform-native, FIPS 140-2/140-3 validated modules when configured for FIPS mode. All release and update packages are digitally signed with keys held in HSMs compliant with FIPS 140-2 or higher.
Ready to Evaluate AirgapAI?
Get a detailed technical briefing tailored to your enterprise architecture requirements.