AI coding assistants and regulated work

The short version

Most compliance friction around AI coding assistants comes from one fact: the tool sends your code somewhere. Nearly every question a reviewer, auditor or review board asks is downstream of that.

Onyx Code runs the model on the machine, which removes the recipient from the picture for the inference step and turns most of those questions into questions about a laptop your organisation already manages. It does not make anybody compliant, and no editor can. What it changes is how many parties are involved and how much of the answer you can demonstrate rather than assert.

These pages describe mechanisms and name the decisions that remain yours. Each one says plainly where the limits are, including the things this product does not have.

The guides

• AI coding assistants and GDPR. The question is not whether AI is allowed. It is whether you have added a processor and a transfer.
• AI coding assistants and HIPAA. What a business associate relationship is, and how a tool that never transmits avoids creating one.
• SOC 2 and AI developer tools. The evidence requests an auditor actually makes, written as a list you can answer.
• Approving AI tooling on classified networks. The process: how a tool gets evaluated, transferred and approved where there is no route out.

The four questions underneath all of them

Different frameworks, different vocabulary, same underlying questions. If you only prepare four answers, prepare these.

• Who receives the data. In local mode, nobody. Inference runs on the workstation over loopback, and bring-your-own-key is opt-in and off by default.
• What the tool can see. Broader than people assume, because an assistant indexes and retrieves across the workspace rather than reading only the open file. .onyxnoai narrows it, and being committed to the repository it is reviewable like any other change.
• Whether the boundary is enforced or promised. Air-Gap Mode is enforced in three independent layers and logs what it blocked, which is a different kind of claim from a policy statement.
• Whether you can show what happened. The AI Activity Log and its PDF export are usually the artefact that ends the conversation.

What you can put in front of a reviewer

Whichever framework you are working under, the ending is the same: somebody wants an artefact.

Whatever the framework, a review comes down to the same two questions: what did the tool touch, and can you show it. Onyx Code answers both from the machine rather than from a vendor policy page.

• AI Activity Log. Every prompt, every file the AI read, every edit it proposed, with timestamps, recorded locally.
• Audit PDF export. The same record as a document you can hand to somebody who is never going to open your editor. This is usually the artefact that ends the conversation.
• .onyxnoai exclusion patterns. Gitignore-style paths no AI surface may read: not chat, not the index, not autocomplete, not agents. It lives in the repository, so it is reviewable in a pull request and it travels with the code.
• Prompt sanitizer. Every send and attachment is scanned locally for credentials, keys, tokens and certificates, with redaction offered before the prompt is built.
• Air-Gap Mode. A switch that blocks outbound traffic, enforced in three independent layers, with blocked attempts written to the log.
• No telemetry by default. Nothing to disable and nothing to negotiate.

All of the above are on the Free tier. Privacy and audit features are never behind the paywall, which also means a reviewer can evaluate every one of them before anybody spends money. The air-gapped development page covers the enforcement mechanisms in more depth, and the documentation gives the commands to verify them independently.