Smart SSH puts an approval gate in front of every AI-suggested command

Field notes

The interface is organized around a command-policy loop: a natural-language request produces a proposed command, rules assign an outcome, and the operator approves, edits, or rejects it. Examples include blocking rm -rf patterns, evaluating chained commands piece by piece, and masking AWS-style credentials before output reaches an AI provider. The page also describes server-scoped rules, OS-keychain credentials, provider selection, and Apache 2.0 licensing.