# AIPermission

A local, single-user permission gateway that lets AI assistants operate developer-owned systems through scoped, credentialed connector actions instead of handing over raw credentials.

AIPermission is a local developer gateway that sits between an AI assistant and the systems it needs to operate. A developer registers connector targets and stores their credential profiles in an encrypted local database, then grants each AI token only specific target, profile, and action combinations, governed by Disabled, Prompt, or Always rules.

When an AI assistant requests a connector action with a reason, the gateway blocks it, asks the developer for approval, or runs it automatically depending on the current rule, and returns bounded results without exposing raw credential values. All built-in connectors share the same project, target, credential, permission, approval, history, and audit pipeline.

It is explicitly a local-only security boundary: it is meant to run on the developer's own machine with its Docker port bound to localhost, not as a remote multi-user service, LAN gateway, or team RBAC system. History and audit records are encrypted and local, and are deleted if the local database is replaced or its retention period expires.

Website: https://npmjs.com/package/@aipermission/mcp