Docs/Help

Safety

What the Gray AI agent does and does not protect you from: it runs shell commands with your privileges and has no command guard.

Source: SECURITY.md

Gray executes shell commands from the model. There is no command guard and no approval prompt: the model's bash runs what it writes, with your user's privileges. There is no container or VM isolation, so run Gray in a container or VM for untrusted work.

Want approvals? The permissions plugin adds a guard on tool calls and fails closed when nobody answers.

REPL sessions keep raw transcripts at 0600 under ~/.gray/sessions for exact resume, including any secret that crossed a tool call. gray -p print mode scrubs secrets before persisting. Plan backups, snapshots and disk access accordingly.

Plugins are scanned at install time and must be granted the capabilities they declare, but consent is an audit trail, not a sandbox. Report vulnerabilities through SECURITY.md.

Join the Discord

Chat with the people building and running Gray. Show what you made with it.

Join Discord →

Need help?

Open an issue on GitHub, run /feedback from the REPL, or ask in Discord.

Open an issue →