Dashboard

The Dashboard is the view Command Center opens on and the one you return to between tasks. It is laid out in the order you actually ask its questions: what just happened (Live), what does my system contain (Infrastructure), what is scheduled (Today's schedule), what can I dispatch right now (Video Analysis), what am I steering toward (Mission Control), what runs on a cadence (your pinned routines), and — last, because it is what you consult when something is wrong rather than what you scan first — is everything running (Platform Health).

The header shows the workspace root the app is operating on — if that path is wrong, fix it in Settings ▸ Project before anything else.

Live — the activity feed

The wide panel at the top left streams recent events: file creations and edits, routine runs, daemon state transitions, mission updates, and session starts. It is how you notice agent work landing without watching a terminal.

• Click a file event to reveal that file in Documents. Shift-click (or the ⧉ icon on hover) opens it in a new Documents window instead.
• Log opens the full Activity Log — the durable, filterable history behind this box, and the only way into that view.
• Clear empties the box only. Nothing is deleted; the history stays in the Activity Log until its retention window removes it.

Infrastructure — what the system contains

The panel beside Live shows live counts of the working parts in your workspace: agents, protocols, active missions, MCP servers, skills, and report documents. Each row is a shortcut — click it to jump to the owning view, except the Missions row, which opens the missions log directly. The MCP Servers row carries a health badge summarizing how many servers are currently reachable.

The Missions count is deliberately narrower than the others: it counts the missions currently active in the missions log, not every mission on record. Everything else counts what is on disk.

If every count shows as unknown, the workspace root is not resolving — check Settings ▸ Project.

Today's schedule — the Calendar at a glance

The tall card on the right is one day of the Calendar, drawn at the week grid's own scale: an hour axis, entries placed at their instant and sized by the length they were given. It scrolls rather than compressing, and opens on the hours that matter — now, if you are looking at today; the first entry otherwise.

• Step days with the arrows, or Today to come back. Once you step away, the card stays on the day you chose instead of following the clock.
• New entry starts an entry at the next half hour of the shown day; you finish it in the Calendar.
• Click an entry to open its occurrence detail — status, history, and the report the run produced.
• A warning pill appears in the header only when something needs attention, counting the failed, missed, misconfigured and overdue occurrences and taking you to the Calendar. It stays absent when there is nothing wrong — a permanently lit "0 problems" badge trains the eye to ignore the spot where a real problem will appear.
• The footer counts what is scheduled for the day, how many are running, and how many sketches are saved as drafts. Sketches are intent rather than schedule, so they are counted here but never drawn.

Two deliberate limits are worth knowing. The card reports that Calendar dispatch is paused and navigates you to where that is changed — it never toggles it, because a live master switch for all scheduled firing sitting on a glanceable card makes one misclick able to silently disarm the machine. And an entry spanning more than two weeks is not surfaced here; the footer says so rather than letting the omission read as an empty day. The Calendar itself is the complete surface.

Video Analysis — the express lane

Below Live and Infrastructure sits the quick lane for Observer's video-analysis pipeline: paste a video URL and process it (download, transcript, keyframes), analyze the newest processed item in one click, or chain the two so analysis starts as soon as the download is ready. In-flight download and analyzer pills, the latest status, and a link to the finished report all show inline.

It adds nothing the Observer view cannot do — Observer ▸ Video Analysis stays the full surface, with bulk selection, model and output overrides, and drill-down. This is the express lane for the one case you hit most often.

Mission Control — the strategic files

The full-width band below surfaces your standing strategic files — the context, directives, and missions log every agent session reads — plus any notes you keep alongside them. You can read and edit them inline without leaving the Dashboard; the same files are the home surface of the Missions Board. See missions and the fleet for what these files mean and how agents consume them.

Routines — your pinned set

The next section lists your favorited routines only, by design — the full catalog lives in the Routines view, and you star the ones you want on the Dashboard from there. Each row is the same live control you get in the Routines view: run and stop, the schedule control, model and status, and the star itself.

A badge next to the section heading shows whether the routines daemon is online. When it is offline the section explains itself: start the daemon from the Routines view or from Platform Health below. When no routine is starred yet, the empty state points you to the Routines view to pin some.

See routines and automation for the model behind scheduled agent work.

Platform Health — daemon control

The last section is the control surface for the background services the platform runs on. One row per service, re-probed every 10 seconds, wrapping into as many columns as the window width allows:

Every row above ArcadeDB is app-managed, unconditionally, and offers:

• Start / Stop — starts or stops that one service. The rest of the app keeps running; views that depend on a stopped service simply show offline.
• Autostart toggle — whether the service launches with the app. This is a hot setting; it persists to the runtime config (see the configuration reference) and applies from the next launch.
• Click the row to jump to the view that owns the service.

ArcadeDB's controls depend on how you run it. On the provisioned-local backend the app owns the server it started, so the row gets the same Start, Stop, and autostart. Pointed at an external or containerized instance, the row reports its version and status only — the app did not start that server and will not stop it. If you are hunting a Start button that is not there, that is why.

Shared Agent Memory has one honest caveat. Its port can also be served by a process an agent session owns. Start adopts such a holder rather than spawning a second one, and Stop refuses to end a process it does not own, reporting that the service is still responding and may be an external instance. Both outcomes are correct: the row will not silently end someone's session.

When a view elsewhere in the app reports "daemon offline", this panel is where you start it. If a service refuses to come up, run Doctor — it probes the same services with fix actions attached.

What you may not see

Two panels are tied to what your subscription includes, and they are hidden rather than shown locked — a paid feature rendering as a broken one is worse than an absent one. Today's schedule appears with the Calendar, and Video Analysis with Observer; both are Pro surfaces. See Tiers: Pro & Ultra. The rest of the Dashboard is part of the shell and always renders.

Related pages

• Activity Log — the durable history behind the Live feed
• Calendar — the full surface behind the day card
• Doctor — health checks with one-click fixes
• Settings — where the wiring behind this screen is configured
• The operator model — why the Dashboard reads the way it does