Private by default means the architecture has to earn that claim.
This Node Addendum supplements the general AdMindra Privacy Policy. It explains the extra data flows created by a local AdMindra Node and by optional Mesh participation.
1. Local Private AI
When you use the local quick-test path, prompts and model inference are processed by services bound to your computer. AdMindra does not need a community-sharing connection for local inference. Local models and their files remain on the computer unless you deliberately remove or replace them.
2. Connecting a Node
When you connect a Node to an AdMindra workspace, the application stores a high-entropy Node credential in the operating system secure credential store. The server stores only the information required to authenticate, route and operate the Node connection. Disconnecting the Node removes the local Node credential.
3. Device and reliability telemetry
A connected Node can report product version, operating-system/platform type, available CPU/RAM/GPU capability, current resource load, selected model capabilities, runtime readiness, heartbeat state and sanitized reliability counters. This lets AdMindra choose compatible workers, avoid overloading a device and diagnose failures. Node is not designed to upload arbitrary personal files merely to produce this telemetry.
4. Personal Mesh jobs
If you send a private job to your own Node from AdMindra on another device, AdMindra routes the scoped job over encrypted outbound connections. Inputs and generated outputs may pass through your private AdMindra workspace or temporary media storage as needed to deliver the result. Private mode does not automatically convert the job into community compute.
5. Community compute
Community sharing is optional. When enabled, a Node receives only jobs that pass AdMindra's current eligibility and policy checks. The Node receives the scoped material needed for the leased job rather than unrestricted access to another user's workspace. Result data required to complete the job can be returned to AdMindra for verification and delivery.
6. Friends, family and network priority
Trusted sharing grants are account-side routing rules. They determine which eligible workloads may receive priority on spare capacity; they do not give a friend remote desktop access, general file-system access or unrestricted control of the Node.
7. Network security
AdMindra Node is designed around outbound HTTPS and localhost-only local AI services where practical. You should not need to expose inbound router ports for normal operation. Node never needs AdMindra provider master keys or Supabase service credentials on the client.
8. Your choices
You can keep local AI without community sharing, pause the worker, turn community sharing off, disconnect the workspace, or uninstall Node. The general AdMindra Privacy Policy continues to govern account, workspace, billing and support data outside the Node-specific flows described here.