Fleet overview
Containers, groups mining, fleet power and active alarms. One card per container with status, power, hash rate per algorithm, indoor temperature and a 2-hour sparkline. Refreshes every 10 seconds.
Home/CubeApp
Every miner. Every signal. One portal.
Included with every container · edge + cloud
CubeApp is two halves. An industrial edge server inside the container polls every miner, sensor, PDU and camera and runs the container on its own: group rotation, grid signal, safety lockout. A cloud portal shows it all to your team, live, from anywhere, and keeps the history.
10 s
Live refresh
0
Open inbound ports
30 d
Buffer on the edge
2FA
On every account
01 · Why
Built for containers that stand far away.
Cheap power is rarely where people are. CubeApp solves the four problems that come with that.
Distance
One portal, every value
Every reading, every camera and every switch of every container in one place, on desktop and on the phone. Problems float to the top.
Outages
Runs without internet
The link to a remote site is the weakest part. The edge server keeps running the container on its own, buffers every reading locally and backfills the cloud when the link returns. Nothing is lost, nothing stops.
Safety
A fixed priority order
Safety shutdown, PLC logic and the E-stop always win over manual control, the grid signal and rotation. Software cannot bypass it. AI advises, AI never controls.
Accountability
Everything logged
Who, when, why, previous to new state. Every client sees only their own containers, read-only, in a branded portal with exports for their own reporting.
02 · How it works
Edge, sync, cloud.
Control happens on site. The cloud watches, stores and notifies.
01
Edge
The agent on the edge server polls miners, PLC, sensors, PDUs, network and cameras. It applies the grid signal and rotates miner groups itself, deterministically, on site. Every reading lands in a local buffer first.
02
Sync
Outbound HTTPS only, no open inbound ports. Each container authenticates with its own token, pushes telemetry every 10 seconds, polls for commands and replays its buffer after any gap. UTC on the wire.
03
Cloud
The portal stores telemetry in a time-series database, evaluates alarms, notifies by e-mail and Telegram and renders live dashboards, the camera wall, history, reports and the copilot. Operators control, clients read.
Priority order · highest first
- Critical safety shutdown
- PLC safety logic
- Manual E-stop
- Manual control
- Grid signal
- Rotation
- Normal operation
03 · Autonomous on site
The container runs itself.
The orchestrator is deterministic, unit-tested code on the edge server. No AI in the loop, no dependency on the cloud.
Signal
Three grid-signal forms
Binary, stepped, or target power in kW translated into groups. Always clamped to the groups that exist, every translation explained in plain words in the log.
Rotation
Fair by design
Groups with the lowest accumulated runtime start first, the highest shed first. Deterministic tie-breaks, visible in the history.
Anti-cycling
No flapping
Minimum run and rest times, a deadband on the signal, staggered starts. A start counts only when 90 % of the group's miners run.
Safety
Safety over manual over grid
Any critical alarm puts the container in safety lockout: no group starts, shedding stays allowed, manual starts are refused with the reason. Shutting down stays the PLC's job.
Manual
Pin, release, lock out
Pin a group on or off above the signal, release it back to rotation, lock a group out for maintenance. Pins and lockouts survive a restart.
Offline
Keeps going without internet
Polling, rotation, alarms and lockouts keep running on cached thresholds. Readings buffer for 30 days and drain in order when the link is back.
Monitor-only mode: a site can run with control switched off. Every state-changing command is then refused on the edge server itself, not by dashboard discipline, and the portal shows it.
04 · The portal
Screen by screen.
Every screen composes itself from the modules the container has. What a container does not have renders no tab and no empty card. All times stored in UTC, shown in your zone.
Container · Live
Power, hash rate, groups and miners with an issue line. Camera wall with the latest frame per camera, the security-system state, an airflow schematic from intake to exhaust, rack temperatures and the strip that explains why these groups are running.
Groups & manual control
One card per group with state, runtime, starts and a miner tile grid. Start, stop, release and lockout behind a confirmation with a reason that lands in the log. Drag-and-drop editor to move miners between groups.
Miners
Every machine in one table with search and filters. The miner drawer shows hash, temperature, boards alive, watts, efficiency, pool, the switch port it hangs on and a 24-hour sparkline.
Grid
The current signal and its translation into groups, the signal history with the orchestrator's response and the per-group decision log: which rule decided each state.
History
One hour to twelve months, compare with the previous period, power, hash per algorithm, temperatures, airflow and the group-runtime distribution that makes rotation fairness visible.
Alarms
Fleet-wide list with filters, alarm detail with the measurement around the episode and the threshold line, the notification delivery log per recipient, and mute with a mandatory reason.
Tower
The always-dark wall display: fleet power and hash with baseline deltas, one tile per container, a live event ticker. A cockpit per container with gauges, a heat-coloured miner matrix and camera tiles.
Also in the portal
- 01SensorsClimate per probe, wind, airflow, door, smoke, electrical, air conditioning, rack sensors
- 02NetworkSite tunnel, WAN latency and loss, a virtual switch per port, Wi-Fi roll-up, VLANs. Read-only
- 03PDUsVirtual PDU strip per outlet, phase balance, consumption history up to 12 months
- 04ActivityAlarms, events, commands, audit entries, notifications and exports in one feed
- 05Reports & exportsTelemetry, alarms and audit log as xlsx or csv, 24 h to 30 d, mailed when ready
- 06DiagnosticsPer-module health, edge hardware, platform jobs, replay from a timestamp
- 07CopilotStaff only. Chat over live data, read-only by construction, every answer lists its sources
- 08Profile & accessTelegram linking, personal notification matrix, 2FA recovery codes, sessions and trusted devices
In your pocket
The same portal on the phone.
Installable web app, usable down to 360 px, hit targets you can actually tap. Nothing is cached on purpose: a control room must never see stale data.
05 · Integrations
Every device in the container, one agent.
Read-only by default. Only miner groups are ever commanded, and PDU outlets where a site explicitly opts in. Everything else is observed, never touched. A device that does not answer shows as unavailable, never as a zero.
- 01ASIC minersLuxOS / CGMiner API: state, hash rate, temperatures, boards, PSU watts, pool. Sleep and wake per group
- 02GPU rigsHiveOS API or SRBMiner local API: per-GPU model, temperature, fan, watts. Read-only
- 03Grid signalBinary, stepped or target-power signals, translated into the number of groups to run
- 04PLC sensorsBeckhoff read-out: temperatures, humidity, wind, door, smoke, airflow, kW, V, A, Hz, kWh. Safety logic stays in the PLC
- 05LoadbankDEIF backup loadbank: status, takeover, kW, per-phase values, fault flags
- 06Site networkTP-Link Omada API or SNMPv3: switches, access points, router, WAN, VLANs, port map with PoE
- 07CamerasRTSP stills every 45 s, ONVIF events: people, intrusion, line-cross, tamper, motion. Latest frame only, no footage retention
- 08Security hubAJAX hub over SIA DC-09 (AES-128): fire, intrusion, tamper, mains, batteries, arm and disarm with user
- 09Rack PDUsSNMPv3: per outlet on/off, current and energy, per phase V, A, W, kWh. Outlet switching where a site opts in
- 10Air conditioning & drivesModbus TCP with a configurable register map: power, mode, setpoint, room temperature, fault code; frequency inverters run state, Hz, A, kW. Read-only
- 11Rack sensorsShelly probes: per-rack temperatures, relay contact
- 12The edge server itselfDisk, RAM, CPU and temperature with alarms; nightly encrypted backup off-site; WireGuard or OpenVPN to the site
06 · Security
Secure by construction.
Hidden buttons are never the boundary. Every rule below is enforced on the server or on the edge, not in the browser.
Accounts
Two-factor on every account
Mandatory TOTP, enforced on first login. Invite-only, no public registration, rate limits on every authentication route.
Authorisation
Checked on the server
Every action is authorised server-side. A container outside your scope does not exist for you, not even as a 403.
Edge
One token per container
A bearer token per edge server, only its hash stored, rotate and revoke from the portal, optional source pinning. Outbound only: no inbound ports on the container.
Transport
HTTPS only
Automatic certificates, HSTS and hardening headers. Plain HTTP is refused, also by the edge agent.
Secrets
Write-only keys
Never in git. Platform keys are write-only in the interface and change only with a fresh TOTP code. Edge credentials never travel in telemetry.
Records
Encrypted backups, permanent audit
Nightly encrypted database and edge backups off-site, verified restorable. Alarms, events, commands and audit entries are never pruned.
07 · Alarms & notifications
Alarms that survive an internet loss.
Threshold and severity of every alarm are configuration per container, from templates, with hysteresis and repeat suppression. Evaluated on the edge, so alarming keeps working when the link is down.
- 01Dashboard, e-mail, TelegramWho hears what is a matrix per container, severity and channel, with a coverage view that marks gaps
- 02Episodes, not floodsRaise and recover fold into one episode; an alarm-storm budget replaces a flood with one notice
- 03Mute with a reasonPresets from a week to six months, a mandatory reason, and an automatic lift when the alarm escalates
- 04Platform alarmsContainer unreachable, buffer high, clock drift, failed critical delivery, backup failed or stale
08 · The AI principle
AI advises. AI never controls.
There is no AI in the control path. The orchestrator is deterministic code on the edge server. The AI layer does two things, both visibly marked and both read-only.
- 01Incident summary on demandA person presses Analyze. The text stays in the interface and never rides a mail or Telegram message
- 02Copilot for staffChat over live data, scoped to what the user may already see. No write tool exists, so none can be misused
- 03BudgetedEvery call is cost-tracked against a monthly budget that switches the AI layer off when reached
- 04Notifications never waitMail and Telegram carry the measured facts and a link, immediately
Request a quote
Two business days.
Tell us what you run and where. You get a priced proposal, not a call-back.
