An unused Android tablet became a dedicated display for a home server and a rented online server, with service status, resource readings and limited lighting controls in one place.
The tablet's unsupported operating system determined the design. A supported Linux host collects the data and sends small updates over USB. Credentials stay off the tablet, and its Wi-Fi and Bluetooth remain disabled. Separate Linux accounts handle observation, USB delivery and approved lighting commands.

Testing checked more than appearance. Interrupting updates had to produce a visible warning, preserve the last readings and recover without restarting the app. Other checks covered restricted account access, confirmation that the tablet displayed the correct update, and stored command history that prevents duplicate execution across service restarts.

The work combined Python/API integration, native Android development, Linux permissions and failure diagnosis. My role covered scope, interface review, hardware checks and approved deployment steps; Hermes and AI coding assistants supported implementation, tests and code review.
The dashboard is deployed with lighting enabled and the superseded owner-level token revoked. Android remains unsupported; reboot recovery and unattended charging safety were outside acceptance.
Technical notes
The overview above describes the finished display. The implementation below covers how information reaches it, how access is separated and what was checked before acceptance.
Keeping collection off the tablet
The tablet is a Hisense Sero 7 Pro running Android 4.2.1. A native Java app supports that older platform without relying on a modern browser. Python services on supported Linux hosts handle integrations and validation. The tablet has no Internet permission and keeps Wi-Fi and Bluetooth disabled.
The data path is:
- The hosted server sends authenticated resource observations to a receiver at Home.
- A selected read-only export separates display data from the receiver's private database.
- Home adds local resource readings, Home Assistant observations, Kuma service checks and passive presence from its existing Tailscale client.
- The USB service delivers a size-limited snapshot to the native app.
Resource telemetry covers Home and the hosted server. Presence for other devices is a separate signal: appearing online does not establish application health or administrative readiness.
Separating permissions
Three dedicated Linux service accounts divide the work. The USB account handles the tablet connection but cannot read Home Assistant credentials. The observation account uses a read-only Home Assistant identity. The lighting broker uses a separate non-administrator identity and owns the command ledger.
Installed source and policy are protected from modification by those runtime accounts. The separation limits access between components; it does not make the old Android installation supported or claim that the entire network is isolated.
Handling lighting requests
The tablet requests only predefined lighting actions. The host checks the allowed action, request age and rate limit before dispatch. Confirmation reduces accidental touches. SQLite stores command identifiers and outcomes so repeated delivery can return an existing result rather than issue the action again.
The cutover preserved all 31 original command records. Acceptance added one confirmed harmless ping, then reused its evidence instead of creating more requests to complete later checks. The superseded owner-level HUD token was revoked only after replacement acceptance. An independent request confirmed that the old token was rejected while both replacement credentials still authenticated.
Checking freshness and recovery
Each observation retains its original acquisition time and expiry. A new USB transfer must not make an old measurement appear current. The native app acknowledges the snapshot sequence it rendered, so delivery alone is not treated as display confirmation.
During acceptance, a verifier sometimes read a new frame before its acknowledgement arrived. A bounded wait for the exact matching pair resolved that timing race. Sequence, freshness and permission checks remained in place.
Actual-device tests checked visible stale readings and recovery without relaunching the app. Other checks covered service-account access restrictions, broker restart, replay handling and command-history preservation. Lighting was enabled after fresh, ready snapshots and matching renderer acknowledgements passed; final closeout did not switch any lights.
Delivered scope
The dashboard is deployed with lighting controls enabled. Existing applications and the Cloudflare tunnel remained unchanged during cutover. Services are configured to start at boot, but a reboot acceptance test was not performed. Unattended charging safety and measured power savings were not established.
Hermes and AI coding assistants supported implementation, regression tests and separate code reviews. Scope decisions, interface review, hardware checks and administrator-approved deployment steps remained part of the work.