The Device Page
This page is a reference for what every row means on the integration's main page in Home Assistant: Settings, Devices and Services, Device Sentinel.
Enable Signals, Enable Last Seen, Enable Battery. Three buttons, one per diagnostic kind. Each scans your entity registry for entities of that kind that an integration shipped turned off, and turns them on so Device Sentinel can learn from them. This only affects watched devices, and entities you disabled yourself are left alone. They are split into three buttons so you can enable exactly what you want:
- Enable Last Seen: Turns on last-contact entities. This is protocol truth and the most reliable clock a device can offer, strengthening freeze detection.
- Enable Battery: Turns on the battery-percentage sensors that the discharge history records.
- Enable Signals: Turns on the link-quality entities the signal floor learns from.
Maintenance Mode. Press this before you work on your hardware. For the next ten minutes, adjustable under Advanced, any device that recovers is treated as your fix rather than learned as its own behavior, so a battery swap or a re-pair never teaches Device Sentinel that a long silence is normal for that device. Press it again to end the window early; it closes itself otherwise. Detection is untouched: a device that fails during your window is still reported. See Maintenance Mode.
Regenerate Reports. Judges every device and rewrites everything it writes: the three Markdown files under config/device_sentinel (the telemetry, the classification and the silence episodes), and both HTML pages under www/device_sentinel (the working daily brief and the signal dwell chart). Press this when you have just fixed a device and want the reports and the dashboard cards to reflect it now rather than waiting for the next scheduled write. The fresh timestamp at the top of each confirms the run.
Problem List. Device Sentinel's own todo list, one row per current finding. The list is live and fills from real detections across the families. Checking a row acknowledges it; a finding that recovers removes its own row automatically. Items you add to this list yourself are left untouched. See The Problem List for reading and using it.
Status. Reads Watching, Learning, or Problem. Learning shows only until the first device establishes its rhythm; Watching shows from then on; Problem means Device Sentinel's own storage is unhealthy. The setup count and install date ride as attributes, and so do three counts, one per enable button: signal_awaiting_enable, last_seen_awaiting_enable, and battery_awaiting_enable. Each is how many entities that button would turn on right now, so a dashboard can show a button only while pressing it would do something, and a count that appears later usually means an integration update shipped new entities switched off. When a number surprises you, the answer is in the log: every press records exactly which entities it enabled.
Maintenance: Ends. A timestamp holding the end of the open maintenance window, and unknown while no window is open. Put it on a dashboard for a countdown, or use it in a conditional card so your maintenance controls appear only while the mode is on. See Maintenance Mode.
Devices Watched. How many devices are being observed and learned. The split behind this number (total, set aside, deviceless entities) rides as attributes.
Devices Learned. How many devices have an established rhythm, meaning enough learned days to be judged for a freeze. This count is not expected to reach Devices Watched, because devices with no natural heartbeat, like buttons and remotes, never establish a rhythm and are never judged frozen, by design.
The remaining sensors come in two families of three, named so they group together in your entity list: every Signal, Battery, and Device entity sorts together.
The Tracked family counts what Device Sentinel watches in each domain, after that domain's exclusions. The Problems family counts what currently has a fault. Both families ship switched off: everything they count already reaches the Problem List, the dashboard card, and the daily brief, so their audience is somebody building a dashboard, who turns on exactly the counts a card needs from the entity's own settings. Both families carry their devices in their attributes.
- Signal: Tracked. How many devices have a learned signal floor and therefore a live danger line, minus the signal-excluded devices. The scale split (LQI and RSSI) and the still-learning count ride as attributes.
- Battery: Tracked. How many devices have an elected battery entity, minus the battery-excluded devices.
- Device: Tracked. How many devices have a learned rhythm and are eligible for freeze detection, minus the global device excludes.
- Signal: Rails. How many devices are reporting a signal stuck at its fill value, 255 or -128, for three days running. That reads as a perfect link and is the opposite: the device stopped filling the field in. This is the one signal fault solid enough to reach your phone.
- Signal: Weak. How many links are struggling right now, meaning they spent more of yesterday below their own learned line than your red threshold allows. This one is live: a device appears while it is struggling and drops off as soon as it recovers. A struggling link also holds a row on the Problem List while it lasts. Nothing about it reaches your phone yet; alerting waits on the judgment earning trust.
- Battery: Falling. How many cells are projected to reach empty inside your Days Till Empty setting. A different set from the one below and a different question: low is a level already crossed, falling is one about to be. The devices ride as attributes with how long each has left. See The Battery Report.
- Battery: Low. How many batteries are low right now. One sensor: the count is the state, and the attributes carry one row per device (name, entity, area, level, low-since, plus the thresholds in effect). The row shape deliberately matches the Battery Sentinel blueprint, so dashboards built for that blueprint keep working.
- Device: Frozen. How many devices are frozen, unavailable, or unknown right now. Each carries its category and how long it has been down in the attributes.
Service Devices Ignored. Devices with no hardware to watch, such as service entries, cloud connections, and integration containers. Device Sentinel classifies these out automatically. This is not the exclude list; you did not choose these and they need no action from you.
This one is switched off when you install, because it answers a question most people ask once if ever. Turn it on from the entity's own settings if you want it.
A fresh install shows what you read daily or act on in your first week: Status, Devices Watched, Devices Learned, Maintenance: Ends, the Problem List, the five buttons, and, where a coordinator was found, the bridge and broker sensors. The Tracked and Problems families and Service Devices Ignored are created but switched off; turn on any you want for a dashboard from the entity's own settings.
Whatever is enabled, the behavior is the same. Detection, notifications, the daily brief, and every report run from the readings themselves rather than from these sensors, so a switched-off count costs you the view and nothing else.
If you installed Device Sentinel before this became the default, your entities keep the state you already gave them. Home Assistant applies a default only when an entity is first registered.