# Air-Gapped AI Assistant: Requirements Checklist | iKB

> What an air-gapped AI assistant needs, what works and what does not without external networks, and a checklist for planning an offline AI deployment.

Source: https://kbmy.ai/guides/air-gapped-ai-assistant

[iKB](https://kbmy.ai/) / Air-gapped AI assistant guide

# Deployment guide Planning an air-gapped AI assistant.

Air-gapped means isolated from external networks. For banks, government agencies and critical infrastructure, that can be the only acceptable way to use AI on sensitive documents. This guide covers what has to run locally, what you give up, and what to check before you start.

[Request a technical review](https://kbmy.ai/contact)

What it means

## Everything runs inside your walls.

In an air-gapped AI deployment, the core services run on your own infrastructure with no connection to the internet: file processing, search, the database, the AI models and the answers. Nothing is sent to a cloud AI provider.

That is different from a private cloud or on-premise setup that still calls a hosted AI model or an external connector. Those setups keep the application private, but data can still leave the network. Review each endpoint and action before you decide.

iKB supports fully air-gapped deployment with local AI models. Core knowledge search and answering run locally. Features that need live external sources or hosted services, such as live web research or cloud connectors, are unavailable in that setup.

What changes offline

## What works, and what does not.

Features in an air-gapped iKB deployment Air-gapped Why Document search and answers Available Runs on local services and local models Source references Available Built from your local documents Access rules and Decision Records Available Stored and applied locally Local AI models Required No hosted model can be reached Live web research Unavailable Needs external network access Cloud connectors (for example Google Drive) Unavailable Need the external service

Requirements checklist

## Seven things to plan before you start.

- 01 ### Hardware for local models AI models run on your servers, so plan compute (usually GPUs) for the number of users and the models you choose. Size it with your vendor.
- 02 ### Model choice Pick local models that fit your hardware, languages and quality needs. Test them on your own questions before committing.
- 03 ### Getting documents in Decide how approved files reach the isolated network, and who checks them. Connectors that need external services will not work.
- 04 ### Updates and patches Agree how software and model updates are brought in safely, and how often.
- 05 ### Access and identity Decide who can use which topics, and how accounts are managed inside the network.
- 06 ### Monitoring and recovery Plan health checks, backups and recovery on your own infrastructure. iKB provides a command-line interface and health probes for operators.
- 07 ### Audit and retention Decide how long Decision Records are kept, who can export them, and how evidence leaves the network if auditors need it.

Straight answers

## Air-gap questions, answered.

**Is air-gapped the same as on-premise?**

No. On-premise means the software runs on your servers. Air-gapped also means those servers have no external network connection, so hosted AI models and cloud services cannot be used.

**Can an air-gapped assistant still give good answers?**

Yes, if the local models suit your documents and questions. Test answer quality on your own material during a pilot before rolling out.

**Can we start in the cloud and move to air-gapped later?**

Yes. Many teams pilot on hosted infrastructure with non-sensitive documents, then move to private or air-gapped deployment for production. Plan the move early, because some features depend on external services.

Keep reading

## Related pages.

[Compare all iKB deployment options](https://kbmy.ai/architecture/deployment) [Local and hosted model controls](https://kbmy.ai/product/model-controls) [Private deployment pricing](https://kbmy.ai/pricing)

## Planning an offline AI deployment?

Talk to our team about your network, hardware and security requirements. We will map what runs locally.

[Book a demo](https://kbmy.ai/contact)