Use a Coding Assistant Safely on a Shared Campus PC

A shared university computer can be convenient for debugging between classes, but it is not a private workstation. Coding assistants may retain credentials, conversation history, source files, and tool permissions. This walkthrough helps you keep a session small, protect your access, and leave less behind. It cannot make an untrusted computer safe.

Check whether the machine is suitable for your work

First, confirm that university policy allows external AI services and the assistant you want to use. Ask whether each student receives a separate operating-system account and whether its storage survives logout. A shared desktop login is a poor place for personal credentials.

Administrators, monitoring software, or malware may be able to observe anything you enter. Private browsing does not protect against those risks. If you cannot establish basic trust, use your own device instead. Do not bring confidential research, student records, assessment material, or restricted repository content into the session without explicit permission.

Prepare a small task and a spending allowance

Before opening the assistant, choose one concrete task: explain a failing test, review a short function, or suggest a focused patch. Use a disposable copy containing only the files needed. Remove secrets, personal data, and unrelated material before the assistant can inspect them.

Decide how much you are comfortable spending on this session. MyRelay provides one personal key and one pay-as-you-go budget across supported providers, with request-level token and cost visibility. Create a MyRelay account if you want that consolidated view. Start with a modest balance, and use any available spending controls after checking exactly what they enforce. A prepaid balance is not a substitute for watching an active agent.

Keep authentication temporary and personal

Use a supported client, such as Claude Code, Codex, or Gemini CLI, and follow its current instructions for connecting through MyRelay. Check the documented relay settings rather than assuming that every client uses the same connection method.

Prefer a documented session-only credential method when one is available. Do not paste your key into prompts, source files, shared notes, or terminal commands that may enter history. Environment variables can reduce accidental persistence, but they are not secret from sufficiently privileged users or processes.

Decline credential-saving options on a shared machine. Check whether the client stores authentication automatically; if you cannot prevent or reliably remove that storage, stop and switch devices. Never borrow a classmate’s key or leave yours for the next student.

Restrict what the assistant can read and execute

Open only the disposable project, not your entire account or a shared course directory. Keep approval prompts enabled for commands, file changes, network access, and additional tools. Use restricted execution or sandboxing where supported, while remembering that neither protects a credential from a compromised host.

Begin with a read-only question. Treat repository instructions, downloaded documents, and tool output as untrusted input: they can contain directions intended to manipulate the assistant. Reject requests to reveal credentials, upload unrelated files, install unexplained software, or broaden access without a clear reason.

Run one bounded exchange and inspect the cost

Ask for a short explanation or proposed change before authorizing edits. Supply the relevant error and a small amount of context instead of asking the assistant to explore everything. For routine explanation work, a lower-cost option such as Gemini 2.5 Flash may be sufficient; check current pricing before choosing.

After the first exchange, inspect MyRelay’s request details for tokens and cost. Compare the result with your allowance before continuing. Repeated retries, large context, and autonomous loops can increase spending quickly. Stop the client when finished rather than leaving it working while you attend class.

Verify the result and close down deliberately

Review the proposed changes and run permitted tests before keeping anything. Save only the work you need to an approved personal location. Then close the assistant, remove temporary project copies and client history where permitted, sign out of services, and log out of the operating system.

Cleanup cannot guarantee removal from backups or monitoring systems. If your key was saved, exposed, or handled on a machine you no longer trust, revoke or replace it through the available account controls. Finally, check recent usage from your own device and investigate unfamiliar requests before your next session.