Running an AI coding agent air-gapped. What it takes, and how Kryex does it.
An AI coding agent that edits code, queries databases and runs commands can work in an air-gapped environment — but only if the agent, the server that authorizes its actions, its models and its audit trail all live inside your boundary.
Updated 16 October 2026
Most coding agents assume the cloud.
Most coding agents send your code to a hosted model, keep credentials in local configuration files, and record what happened — if at all — after it happened. For any team whose code can't leave the building, that rules them out.
An agent that runs air-gapped has to meet six requirements. Here they are, with how Kryex Code and Kryex Admin meet each one.
Six requirements for an air-gapped coding agent.
Every component inside your network
The agent, the server that authorizes its actions and the admin console all have to run on infrastructure you control, with no dependency on a vendor's cloud.
Kryex Code, Kryex Server and Kryex Admin all deploy in your VPC or fully air-gapped. Nothing requires Kryex Labs' servers to be reachable.
Models that run where your code is
An air-gapped agent can only call models inside the boundary — self-hosted open models or your company's private endpoint.
Enterprise runs only the models you connect: local models through Ollama or LM Studio, or any OpenAI-compatible endpoint inside your walls.
Credentials that never reach a laptop
Database passwords and API tokens on developer machines are the first thing a security review flags.
Credentials for governed systems are encrypted on Kryex Server, sealed per organization. The desktop only sends which tool and which arguments.
Authorization before execution
Logging what an agent did is not enough when it can write to production. The decision has to happen before the action runs.
Every database statement, MCP tool call, browser and web action is checked against the person's live grant on the server first. Default is deny.
Control over commands on every machine
Shell access is where agents can do the most damage, and policy has to hold even when a laptop is offline.
Admins set Allow, Ask or Block per command category, can turn shell commands off entirely, and can lock the policy on devices, even offline.
An audit trail you own
Regulated teams need evidence of every decision, stored in their own environment.
Every governed decision is written to a hash-chained, append-only audit log in your deployment, with a live view, filters and CSV export.
Common questions.
Can an AI coding agent work without internet access?
Yes, if the agent, its authorization server and its models all run inside your network. Kryex supports this: in an air-gapped Enterprise deployment it runs only the models you connect, such as local models through Ollama or LM Studio or your own OpenAI-compatible endpoint.
What is the difference between self-hosted and air-gapped?
Self-hosted means the software runs on infrastructure you control, which may still reach the internet. Air-gapped means that environment has no connection to outside networks at all. Kryex Enterprise supports both: deployed in your VPC, or fully air-gapped.
Which models can an air-gapped coding agent use?
Only models that run inside the boundary — open models served locally, or a private endpoint your company hosts. Kryex Code works with Ollama, LM Studio and any OpenAI-compatible endpoint.
How do you audit an AI agent in an air-gapped environment?
The audit log has to live inside the boundary too. In Kryex, Kryex Server writes every governed decision to a hash-chained, append-only log in your own deployment, and admins review and export it from Kryex Admin.