What makes a coding agent private?

The strongest privacy boundary is keeping model inference and agent data on controlled hardware rather than sending repository context to an external model endpoint.

Is privacy mode the same as local inference?

No. A hosted privacy mode can limit storage or training, but the request still leaves the machine. Local inference removes that external request.

Can OpenJet run air-gapped?

Yes, after installation and model acquisition. Its air-gapped mode blocks non-loopback network access.

private AI coding agent

A private AI coding agent for sensitive repositories

Privacy mode and local inference solve different problems. Local inference removes the external model request from the data path entirely.

OpenJet keeps the model, repository tools, prompts, output, transcripts, and resume checkpoints on the machine. It is designed for developers whose privacy requirement is local processing, not only a hosted provider's retention setting.

Install OpenJet Read the docs

Why local wins

Inference stays local. Repository context is processed by an open model running on the workstation or controlled GPU host.
Telemetry is narrow. OpenJet excludes prompt text, tool output, file paths, model paths, and tool arguments from telemetry.
Actions require approval. Local does not mean unchecked: shell commands remain approval-gated.
Air-gapped mode is explicit. A network guard can block non-loopback access, including DNS, after required assets are installed.

Common deployment paths

Work on client source code that cannot be submitted to a model vendor.
Use AI assistance in regulated development environments.
Keep security review evidence inside a controlled network.
Give a small team a shared local GPU agent without external inference.

Frequently asked questions

What makes a coding agent private?: The strongest privacy boundary is keeping model inference and agent data on controlled hardware rather than sending repository context to an external model endpoint.
Is privacy mode the same as local inference?: No. A hosted privacy mode can limit storage or training, but the request still leaves the machine. Local inference removes that external request.
Can OpenJet run air-gapped?: Yes, after installation and model acquisition. Its air-gapped mode blocks non-loopback network access.

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.