air-gapped AI coding agent
Run a coding agent inside an air-gapped network
Air-gapped AI requires more than a privacy promise: the model, runtime, dependencies, repository context, and operational workflow all need an offline path.
OpenJet can operate after the runtime and model assets are staged locally. Its air-gapped mode blocks non-loopback network access, including DNS, while repository tools and approved shell execution continue on the machine.
Why local wins
- No hosted inference dependency. The open model runs inside the disconnected environment.
- Network guard. Air-gapped mode blocks non-loopback connections and DNS rather than relying only on configuration convention.
- Stage assets deliberately. Model weights, packages, and runtime dependencies can be acquired and transferred before the network boundary is closed.
- Keep the agent useful. File tools, session state, context handling, and approved commands remain available without the internet.
Common deployment paths
- Develop software on disconnected research or defense networks.
- Review sensitive code where outbound access is prohibited.
- Maintain field systems during extended network loss.
- Evaluate local coding models under sovereign-compute requirements.
Frequently asked questions
- Can an AI coding agent run fully air-gapped?
- Yes, when all model weights, runtime components, and dependencies are staged inside the environment before outbound access is removed.
- How does OpenJet enforce offline operation?
- Its air-gapped mode installs a hard network guard that blocks non-loopback access, including DNS.
- Does the agent still run shell commands offline?
- Yes. Local shell tools remain available under the normal approval requirement.
Explore OpenJet
Embed a local model with the Python SDK, run Qwen3.8-27B as a local coding agent, compare local coding-agent alternatives, read the docs, or go back to the homepage.