Skip to content
AtheronLABS

You're visiting from the United States. Prices are shown in US dollars. Not right?

labs@atheron:~/industries/defence-aerospace/ai$ train agent --grounded

Air-gapped defence AI

A manuals assistant that runs inside your own network.

Restricted manuals and outside AI services do not mix. An open-source model on your own hardware, inside your own network, gives maintainers the same help without any data leaving. It answers and cites; the maintainer and the manual stay the authority.

// what matters here

What is different in this industry.

01

No path out

The model runs on a GPU server on an air-gapped network. No query, manual or answer reaches the internet.

02

Cites the section

Every answer names the manual and section it comes from, so a maintainer can check the source before acting.

03

Sees only what the user may see

Access follows the user's role, so a manual a person may not read is never used to answer them.

04

Updated by your process

New manuals and model updates come in on approved media, on the schedule your security team sets.

// example projects

Priced examples for this service.

Each one is hypothetical, labelled as such, and priced live from our rate card. Open any of them in the estimator and make it yours.

Example projectWeb appAIOpen-source model

Air-gapped manuals assistant

Hypothetical. Not a client, not a result.

// The problem

Maintainers search thousands of pages of restricted technical manuals by hand, and no outside AI service is allowed near them.

// What we would build

An assistant on an open-source model running on a GPU server inside an air-gapped network, answering from the organisation's own manuals and citing the section, with updates brought in on approved media. Nothing reaches the internet, in either direction.

// What is in it

  • Manuals and service bulletins indexed inside the air-gapped network
  • Answers that cite the manual section, for the maintainer to check
  • Access by role, so each user sees only the manuals they may see
  • An open-source model on a GPU server with no network path outside
  • Updates and new manuals loaded through your approved media process

// Stack

  • Llama or Mistral (open-source)
  • vLLM
  • PostgreSQL with pgvector
  • Next.js
  • Lenovo GPU server, air-gapped

// estimate

Build
≈ US$120,000 to US$183,000, delivered within 25 weeksCAD 171,200 to 260,200
Hosting
≈ US$176 a monthCAD 250 a month
Support
≈ US$2,500 a monthCAD 3,565 a month
Hardware
≈ US$47,800 to US$197,000CAD 68,000 to 280,800

Prices in your currency are estimates from today's Bank of Canada rate. All invoicing is in CAD or USD.

AI route
Open-source model
Model running cost
≈ US$176 a monthCAD 250 a month

Timeline by milestone

Discovery
2.9 to 3.4 weeks
Specification and evaluation plan
0.4 weeks
Design approved
2.4 to 3.4 weeks
Access rules reviewed with your designated official
1.4 to 2.4 weeks
Core features
1.4 to 2.4 weeks
Full build
0.9 to 1.4 weeks
Working pilot
3.9 to 6.9 weeks
Testing and fixes
0.4 to 0.9 weeks
Penetration test findings closed
0.9 to 1.9 weeks
Launch
0.9 to 1.4 weeks
Production
0.9 to 1.4 weeks

Outside our hands, and added to the calendar

Hardware delivery, after it is ordered
2 to 6 weeks
Independent penetration test and retest
2 to 4 weeks

How it is paid

Deposit 20%
CAD 34,180 to 51,980
Discovery 1.5%
CAD 2,563.50 to 3,898.50
Specification and evaluation plan 8%
CAD 13,672 to 20,792
Design approved 1.5%
CAD 2,563.50 to 3,898.50
Access rules reviewed with your designated official 7.7%
CAD 13,159.30 to 20,012.30
Core features 4.6%
CAD 7,861.40 to 11,955.40
Full build 3.1%
CAD 5,297.90 to 8,056.90
Working pilot 24.2%
CAD 41,357.80 to 62,895.80
Testing and fixes 1.5%
CAD 2,563.50 to 3,898.50
Penetration test findings closed 7.7%
CAD 13,159.30 to 20,012.30
Launch 1.5%
CAD 2,563.50 to 3,898.50
Production 8.7%
CAD 14,868.30 to 22,611.30
Holdback, 30 days after launch (10%)
CAD 17,090 to 25,990
GPU time for training and testing, at cost
CAD 300. Billed up front, at cost, outside the milestones
GPU server for AI, for the model
CAD 68,000 to 280,800. Billed up front, at cost, outside the milestones

// questions

Questions we get about this.

Why open-source for restricted manuals?

A frontier model is reached over the internet. An open-source model can run entirely on your hardware, which restricted data usually requires.

How is the manuals assistant tested offline?

The evaluation set is built inside the network from real maintainer questions, and run before every update.

Can it read drawings as well as text?

Text is the starting point. Drawings and diagrams can be scoped separately, depending on what your security team allows and what the model can read reliably.

// next

Not quite your project?

Tell us what you have in mind. We will come back to you with a range and the questions that would narrow it. Or book a call and talk it through.

Air-gapped AI assistant for defence and aerospace manuals | Atheron Network Labs