What's New

This manual documents QUBIQ 1.0.6. Capabilities that arrived during the 1.0 series carry a badge like this one New in 1.0.6 on the page that describes them, so you can tell at a glance whether something is available on the version you are running.

1.0.6

A licence you buy at 10am works at 10:01. Changed in 1.0.6 Every service re-reads and re-verifies the signed licence file the moment one is installed, across the whole fleet, within the same second. Before this, only two processes heard about it: a customer who added the Siemens S7 and Allen-Bradley modules to a running installation watched the supervisor start both workers — visible proof the purchase had landed — and then got MODULE_NOT_LICENSED when they created a connection, because every other service was still holding the entitlements it read at boot. The only cure was the restart the whole mechanism exists to avoid. → Licensing

Report by exception, for real. Changed in 1.0.6 Publish on change only has been on by default since 1.0.0 and did nothing for the ordinary tag: the gate was the tag's deadband, which defaults to 0 and is documented as a threshold for storage, so a tag with none — which is most of them — went to the broker at the full scan rate, and a boolean was never suppressed at all.

• A reading is now sent when it says something new: the value changed, or the quality did. Quality is part of the reading, not metadata about it — with Retain on, suppressing a Good→Bad transition at an unmoved value would leave the broker serving a Good-looking number to every late subscriber for as long as the process stood still.
• A heartbeat per publisher, off by default, re-sends the value already held after a chosen silence. RBE makes silence ambiguous — a process standing still and a dead gateway are both a topic that stopped — and a tag on a 250 ms poll with a 60-second heartbeat costs the broker one message a minute instead of 240.
• A tag can opt out on its own. Follow the broker / only when it changes / every reading, saved with the tag, because cadence is a property of the signal: a counter, a heartbeat or a batch step has to arrive at its scan rate wherever it is sent, and one tag can be in scope of several publishers at once.

This changes the message rate on upgrade

The first run after upgrading drops existing publishers from scan rate to change rate. That is the intent and a large reduction — but it is a real change for a subscriber that was using message arrival as a liveness signal, which is what the heartbeat is for. → Report by exception

Publishing a folder is a checkbox, not a verb. Changed in 1.0.6 The tree's Publishing menu now gives a folder and a multi-selection the same per-broker checkbox a tag already had, with the third state that was blocking it: ticked, unticked, and a dash for a folder the fence only reaches into. A dashed box hands off to the dialog rather than guessing — "publish the rest" and "stop the ones that are" are equally plausible readings of the same click, and that menu writes immediately. The dialog loses its Action radio, which was asking a second time for a decision already made. And the bug underneath: stopping a folder now stops the subtree, where before a publisher scoped to Line1/Pump and Line1/Valve made stopping Line1 report no change and leave both tags publishing. → MQTT publishing

A backup that is honest about media. New in 1.0.6 The File Manager's content and the uploads store now travel — under a new Include File Manager & uploaded files flag that is off by default, because those folders hold videos and 3D models and a nightly backup dragging 400 MB of them is one that stops being run. The file list always travels either way, so a view that points at a document still points at it after a restore, and row and file reunite by content the moment the folders are copied across.

The honest part: on any instance built since August, the File Manager's content was in no archive at all. It had stopped being an internal catalogue and become an ordinary folder, and the backup went on carrying the catalogue — which was empty, and "0 files" is indistinguishable from an installation that never used the feature. Every guard stayed green. → Media

A backup moves between Linux, Windows and macOS. New in 1.0.6 Names that one filesystem stores and another cannot are now reported: refused on import for a project, view, script or query name, warned about on export while you can still rename the thing, and refused at creation for the names that are keys. The three silent cases are the reason — a trailing dot, a case variant and two Unicode spellings of one word each collapse two files into one with no error at either end. Passphrases are normalised before the key is derived, so an accented one typed on a Mac and on Windows opens the same archive. → Moving between operating systems

Writes to a PLC stop queueing behind other people's problems. Changed in 1.0.6 An audit of the Siemens S7 and EtherNet/IP write path, which had had less attention than the read side. One dead panel used to stall writes to every other controller — and the caller was then told worker unavailable about a worker that was perfectly healthy. An operator's setpoint now goes in front of queued polling. A learned element size is discarded on reconnect, so a program download cannot leave a stale width behind and reopen the silent partial write the probe exists to close. A STRING write the CPU's PDU could only apply in halves is refused with the numbers and the remedy rather than torn. And a value whose meaning is not obvious — 2 or maybe written to a BOOL — is refused instead of quietly written as OFF. → Siemens S7 · Allen-Bradley

The script sandbox says what is actually installed. Changed in 1.0.6 The allow-list pre-approves around thirty third-party libraries — numpy, pandas, cv2 — and ships none of them installed. Allow-list membership is permission, not availability, and the two looked identical on the screen; the first sign of the difference was a ModuleNotFoundError at run time. Each entry now carries its own state, the list filters by it, and a missing one offers to install itself. → The allow-list

A Problems panel, and one font size across every code editor. New in 1.0.6 The Designer hosts six separate Monaco editors and their size was fixed, so anyone who turned it up to read a long script had to do it again in the next panel. There was also nowhere to see what the linter had found — the footer showed a bare count, which tells you there is a problem and then declines to say what. Both linters already produced the message and the location; now you can click it. → The editor

The 3D Viewer's camera. New in 1.0.6 Motion is eased by default, in both directions: an operator's rotate, pan and zoom glide to a stop instead of snapping, and a new pose — a target re-aimed by an event action, a bound tag or an edit — flies there rather than cutting, so nobody loses track of which way round the equipment is. enableRotate joins the gesture switches so a fixed overview cannot be tumbled into a useless view. The [x, y, z] properties now name their slots and fill the whole tuple in one click, where the grid used to show rows 0/1/2 and its + appended a bare 0 — producing a 1- or 4-element array the widget rejected, so the field appeared to do nothing. → 3D Viewer

AI: three more things a calculation can read. New in 1.0.6 Tag relationships, Manuals and Stored file join the input kinds for a specialist's compute body and a playbook's checks. A compute-only specialist runs no nested conversation and so holds no tools at all — its only data is that list — so until now a calculation over a commissioning record, a manual or a tag's stored links was not expressible, while a playbook could order the identical read in one step. A compute-only specialist is also now offered every kind: the dropdown followed the ticked skills, which that mode deliberately has none of, so Saved query never appeared and definitions the server accepts and runs could not be written in the editor. → Specialists

Also in the assistant: a compute body may be 40,000 bytes rather than 20,000 (a compute body is never sent to a model, so the old number was justified by a cost that does not exist); each turn is told which tools it actually holds, instead of inheriting a persona that describes every skill an installation might plug in; and Read files joins the other reads as a default, bounded by still needing Manage the library, which an operator cannot hold.

The AI node's defaults fit the work. Changed in 1.0.6 maxToolIterations 3 → 8 and the node timeout 15s → 60s. Three cannot finish the sequence the Database Explorer skill documents — list the databases, introspect for tables, introspect for columns, then query — so a node was cut off one call short and answered "I don't have access to any data source", which reads as a permission problem and is not one. → AI node

A Python node's timeout reaches its system.* calls. Changed in 1.0.6 A node given 600 seconds used to die on the first system.* call that ran past a fixed internal limit, whatever the node was configured for.

Also in 1.0.6: a Windows arm64 build exists and is produced on request, deliberately out of the default release until somebody has run a gateway on one — an ARM64 laptop installs the x64 build today, and this manual's platform table had been listing arm64 as though it were a download.

Fixes worth knowing about: a tag bound over Siemens S7, Modbus, Allen-Bradley, SNMP or TCP/UDP showed the raw connection UUID where its name belonged; a purged project left phantom pipelines in the SuperAdmin statistics until a restart, and a project created after boot was never watched at all; a failed request to the MQTT Publishing page rendered as "the MQTT Driver module is not licensed" — the worst available mistranslation of "I could not reach the server", because it sends somebody to look at their licence instead of at the error; unticking Retain, the loop guard or publish on change only while creating a publisher was silently ignored; deleting a specialist in the Pipeline workspace left its row in the tree; and a licensed service could silently fail to hear about connection lookups, script file reads or the system.* worker API because the bus had never been told to grant those subjects.

1.0.5

Siemens and Allen-Bradley controllers, spoken natively. New in 1.0.5 Two drivers, and deliberately two — a plant standardised on Siemens should not be carrying an Allen-Bradley licence.

• Siemens S7 reaches S7-300/400 and non-optimized S7-1200/1500 over classic S7comm, by absolute address: data blocks, flags, the process image, timers and counters. The page carries a worked tag for every memory area, because the offset arithmetic (MW4 is the word at byte 4, so the next one is MW6) is where the mistakes live.
• Allen-Bradley (EtherNet/IP) reads Logix controllers by their own tag names. Browse pulls the symbol table and the UDT templates off the controller and fills in both the address and the data type, so you do not have to know either.

Both confirm an address before you bind it: a Read now button shows the decoded value and the raw bytes side by side, which is the difference between "this address is wrong" and "this type is wrong" — two problems that look identical in the number alone.

A refusal no longer costs you the connection. New in 1.0.5 Both drivers multiplex every tag onto a single link. A CPU refusing one address — a data block that does not exist, an offset past the end — is now reported against that tag alone. Every other tag on the controller keeps updating, and the link stays up.

Expression tags. New in 1.0.5 Derived values in CEL, compiled and cached, evaluated server-side. They are fail-closed: a missing input is an error on the tag, never a silent zero — which matters the moment somebody alarms on the result.

A backup you can actually restore from. New in 1.0.5 The archive now carries the whole schema and the File Manager's bytes, and a restore can run at first boot — before commissioning, rather than after it. See Backup & restore.

Licensing. New in 1.0.5 Per-module entitlement, enforced at the service boundary rather than in the UI.

A backup will not be read by a build that would read it wrongly. New in 1.0.5 Archives carry a compatibility floor measured from what they actually contain, so an estate that never touched a newer feature keeps producing archives older builds accept. Two silent-corruption bugs went with it: a restore no longer turns disabled connections, disabled tags and suspended users back on. → Backup & restore

1.0.4

Errors are something a pipeline can act on. New in 1.0.4 Until now a failed node was logged and its branch stopped. Two features change that, and they compose:

• The Catch node receives every failure in its pipeline — or a chosen few — and hands it to whatever should report it: a Telegram message, a dead-letter table, a Link into one central error-handling pipeline. The message it emits carries not only the error but the payload that caused it, so a failure can be replayed rather than only announced.
• A per-node error output puts a second, red socket on any node. Wire it and that node's failure carries on down a branch of its own, so a cached value can be substituted and the rest of the flow still runs. Off by default; every saved pipeline renders exactly as before.

A lost administrator password is recoverable, and a console reset cannot be quiet. New in 1.0.4 qubiq admin reset-password at the server console prints a temporary password once, forces a change at next sign-in, and ends every session that account held. It is recorded three ways, one of which no one can dismiss. First-run setup is now genuinely one-shot: deleting the user table no longer reopens the public setup endpoint, which is the hole this replaces. → Console recovery

The database encryption key has an escape hatch. New in 1.0.4 qubiq db-key create prints a recovery key you write on paper. Without one, a changed service account or a rebuilt profile could leave app.db permanently unreadable — which needs no attacker, just an ordinary Tuesday. The key is also re-sealed on every boot so a service-account change stops being a data-loss event. → Console recovery

A script may not do what its caller could not. New in 1.0.4 Tag write levels now apply to system.tag.write, not just to a write from a screen; and a Gateway script started from a browser button runs with the caller's own capabilities rather than unclamped. The Script Sandbox page also says plainly, at the Install button, that a data library widens what scripts can reach — the sandbox stops mistakes, not a determined author. → The script sandbox

Handled faults have their own log level. New in 1.0.4 WARN sits between routine chatter and an outage, and is filterable. It also fixes a live bug: the log panel had been labelling every warn row RETRYING, telling its reader to expect an attempt that was never coming.

Two new API reference pages — every system.* and context.* member, with worked examples in both languages. → System API · Context API

An electrical symbol library. New in 1.0.4 529 electrical symbols across 24 categories — contacts, breakers, relays, logic gates, op-amps, MOSFETs, PLC ladder elements — join the 485 process symbols on the same canvas. 77 of them carry an operating state you can bind to a tag: a contact opens, a coil lights, a selector turns. Both states share a viewBox and terminal positions, so a bound line stays connected when the tag flips. → Drawing & symbols

The Breakpoint layout. New in 1.0.4 A fourth layout mode, for a screen whose phone version is a different screen rather than a narrower one. Each width tier gets its own authored panel and the view renders exactly one; a tier can mirror another, so a new view costs two screens rather than three. Switches at the project's own breakpoints, measures the container rather than the window (so they nest), and carries a dead-band for the flicker that happens when two tiers disagree about a scrollbar. → Breakpoint layout

The Timer node. New in 1.0.4 The pipeline runtime can now put time between messages — pulse a tag and put it back, hold a burst until it settles, delay every message, count down to a deadline, repeat on a cadence, or raise an alarm when a message that should have arrived did not. One independent timer per topic, controllable from the flow itself. → Timer node

Search that answers three questions. New in 1.0.4 A ⌘K / Ctrl+K palette across every project; faceted tag filters in the namespace (is:disabled, no:binding, is:unused, has:history, is:published, source:expression, …); where is this used for a tag or a connection; and an impact warning before a rename or delete that names what will break. → Search & impact

Telegram sends a position. New in 1.0.4 A pin, a named venue, a live location this platform then moves as a vehicle moves, or a button asking a recipient to share theirs. → Telegram node

Numeric readouts keep their decimals. New in 1.0.4 The Gauge, Tank, Moving Analog Indicator and HMI Trend take a decimals property instead of rounding a value away. A tag reading 23.998 is no longer displayed as 24 with no way to get the fraction back. → Gauge

Zoom and fit controls on the pipeline canvas. New in 1.0.4 Undo, redo, zoom in and out, reset to 100% and zoom-to-fit, with their keyboard equivalents shown on each.

The Connected column carries a date. New in 1.0.4 A LiveView screen on a wall stays up for days, and a bare 14:03 on such a row reads as a few minutes ago when it was Tuesday. The full instant is in the cell's tooltip.

Also in 1.0.4: the trial is recorded outside the data directory, so deleting that folder no longer buys a second one; and the pipeline-level POST /api/pipelines/{id}/execute route was removed — it answered 500 on every call it ever received, and nothing used it.

Fixes worth knowing about: node status badges never rendered in a production build; the runtime's STALE indicator was measuring clock skew rather than staleness; Python results were being refused by the bus, and child processes looked for the interpreter in the wrong directory.

1.0.3

A configuration fix that matters on every production install: every service opened a different app.db. Also: the import browser no longer collapses, and the storage folder picker no longer flickers or turns unreadable on hover.

1.0.2

AI Specialists can compute. New in 1.0.2 A specialist may carry a Python compute body, run end to end behind every fence, and declare its inputs once — so a figure is calculated exactly rather than worked out in a model's head. A specialist can also be saved as a draft: testable by its author, withheld from the assistant. → Answering with code

The trial is 21 days, not 7. New in 1.0.2 A week can vanish waiting for PLC access or a downtime window. → Licensing

On Windows, double-clicking the executable no longer flashes a console, and an uninstalled copy refuses to run.

1.0.1

P&ID line types. New in 1.0.1 ISA-5.1 tells signal lines apart by a repeated glyph rather than by colour — pneumatic hashes, data-link circles, a capillary X, a sonic wave. A line now carries its meaning as a property. → Drawing & symbols

Fittings that ride the pipe. New in 1.0.1 A flange, orifice plate, strainer or spectacle blind sits on a run at a normalised position along it, and stays there through a drag, a bend or a change of routing mode. → Drawing & symbols

A multi-value Progress bar. New in 1.0.1 One track, several coloured segments, each separately bindable, in absolute or proportional mode. → Progress

Storage folders are browsed, not typed. New in 1.0.1 A server-side folder picker with OS-aware suggestions, so a path is chosen rather than guessed at.

The status panel shows every volume QUBIQ writes to, not just one. New in 1.0.1

Open-source notices. New in 1.0.1 A third-party licence inventory, and the page that shows it, at /notices.

Also in 1.0.1: multi-touch pinch and pan on the Designer and pipeline canvases; a rebuilt Windows installer, and the same application identity on macOS and Linux; and licence status in the app status bar.

The Breakpoint Container widget was withdrawn from the component library in this release. The idea returned in 1.0.4 as a layout mode rather than a widget you drag from the palette — its containers are structural, so they are not in the component reference.

1.0.0

The first release. The parts of it this manual gained pages for:

MQTT publishing. New in 1.0.0 Mirror namespace tags out to an external broker — classic MQTT or Sparkplug B, with scope and exclusion fences, four layers of loop protection, and a choice between live and durably-buffered delivery. → MQTT publishing

Per-module licensing. New in 1.0.0 Offline, hardware-bound activation; a modules page that never hides what you do not own; and a lapse that degrades visibly rather than stopping a plant. → Licensing & modules

The Switch node. New in 1.0.0 The first router in the pipeline runtime, and the only node with more than one output. → Switch node

A Color Picker widget, and menu components that can run an action instead of navigating. → Color Picker · Horizontal Menu

AI playbooks, specialists and named endpoints. New in 1.0.0 An answer procedure written down once so every phrasing of a question gets the same complete answer; saved sub-assistants the assistant hands one sub-question to; and an admin-authored menu of lookups into an ERP or a CMMS that the model can select from but never compose. → Playbooks · Specialists · External systems · Skills & fences

---

The platform at a glance

What QUBIQ does, and where each part is documented.

Platform

Ships as one file. A single executable contains every service. Starting it with no arguments boots the supervisor, which runs one process per service under its own QBQ-* name. → Architecture

Zero-port internal bus. The message bus binds to loopback on a random port. No broker port is exposed, authenticated or firewalled, and no service other than the gateways opens a port. → The gateway

Encryption at rest. The configuration database is encrypted with a key bound to the machine it runs on. JWT_SECRET and ENCRYPTION_KEY are required, with no insecure defaults. → Security

In-app setup wizard. There is no default account and no create-superadmin step; the first super admin is created through the interface, once. → First-run setup

Connectivity

Protocol coverage — OPC-UA, MQTT with Sparkplug B, Modbus, SNMP (v1/v2c/v3 with vendor MIBs and traps), raw TCP/UDP, SQL, MongoDB, QuestDB and media servers — plus SMTP, Telegram and WhatsApp for notification and a Qdrant vector store for the assistant. → Connections

Publishing out, not just reading in: tags mirrored to an external broker in classic MQTT or Sparkplug B, at the change rate rather than the scan rate, with a per-publisher heartbeat and a per-tag cadence. → MQTT publishing

Passive liveness. Connection health comes from the protocol libraries' own callbacks rather than from polling. LIVENESS_MODE selects passive (default), hybrid or active. → Connections

OPC-UA node quarantine. A repeatedly-failing node is quarantined so one bad address cannot degrade a whole subscription, with a release path when the PLC program is fixed. → OPC-UA

Isolated MIB compilation. Vendor SNMP MIBs are compiled in a separate child process, so a malformed MIB cannot crash the gateway. → SNMP

Data model

User-defined types with parameters, nesting, reconciliation planning, per-instance overrides and member-context capture/revert. → User-defined types

Portable namespace import/export. Documents carry no database identifiers and reference connections by name, so a model moves between environments cleanly. → Bulk import & export

Tag write-security levels. A 0–3 hierarchy enforced server-side, independent of the writable flag. → Tags

Reference tracking. Where a tag is used, what a rename would break, and what nothing references at all. → Search & impact

Historian

History epochs. Changing the historian is non-destructive: reads resolve across epochs and stitch the timeline. Range statistics are recombined exactly rather than averaged. → Historian

Physical store ownership. Ownership is recorded inside the QuestDB store, so a cloned VM or a restored backup cannot silently corrupt a live historian. → QuestDB

Self-healing. Schema-drift detection with rebuild, write-ahead-log supervision that distinguishes recoverable stalls from poison segments, and on-demand rollup rebuild. → Historian configuration

Live-tail merge. Recent in-memory samples are merged into live chart queries, so charts look live despite batched storage.

Reliability

Store and forward for all outgoing data writes: a durable per-connection buffer, ordered ack-after-flush drain, an ordering latch, transient-versus-permanent failure classification, buffer pressure alarms and an operator quarantine. → Store and forward

Pipeline error handling — a Catch node that reports any node's failure with the payload that caused it, and a per-node error output that lets the flow recover in place. → Errors

Alarms

ISA-18.2 lifecycle — NORM / UNACK / ACK / RTNUN, with latching, require-ack, on/off delays, absolute and percentage deadbands, and bad-quality alarming. → Configuring alarms

Three distinct suppression kinds — timed shelving with automatic release, indefinite suppress-by-design, and out-of-service — so the reason an alarm is silent is always recorded. → Operating alarms

Performance metrics — rate, floods, bad actors, priority distribution and standing alarms.

Visualization

53 widgets, 9 drawing tools, 1048 symbols — 485 process (P&ID) and 529 electrical, plus 3D model viewing, PDF, video, maps, scheduling and a full form engine. → Components

Vector drawing with orthogonal, isometric and curved routing, endpoint binding to equipment, ISA-5.1 line types, in-line fittings and stroke animation for process mimics. → Drawing & symbols

Four layout modes — coordinate, grid, flex and breakpoint — chosen per view. Breakpoint gives each width tier its own authored screen rather than reflowing one. → Views and layout · Breakpoint layout

Optimistic locking on views, scripts and pipelines: a conflicting concurrent save is rejected rather than silently overwriting.

Streamed query results in LiveView, so a large result set renders as it arrives rather than after it has all landed.

Pipelines

23 node types across eight categories — sources, routing, protocol I/O, databases, transforms, AI, notification, REST and error handling. → Node reference

Scripting

One API surface in two languages. system.* is identical, camelCase, in Python and JavaScript. → System API reference

Gateway scripts with default-deny @gateway exports, static AST-based discovery, per-entry-point timeouts and idempotency keys. → Gateway scripts

Managed transactions with isolation levels and a server-side TTL, so a browser disconnect cannot strand an open transaction.

Capability scopes gating every system call, fail-safe by default — a newly-added API is gated rather than accidentally open. → The script sandbox

Administrator-installed Python packages, isolated in a managed directory with a connectivity check for air-gapped networks.

Security

Two independent sign-in systems. Engineers and plant operators sign in separately, against different user stores. Neither system accepts the other's sign-in. → LiveView authentication

Tamper-evident audit journal in an external SQL database, hash-chained, durable-buffered, with a verification endpoint — aligned to IEC 62443-3-3. → Audit journal

Scoped API keys, hashed at rest, with fail-closed REST routes, IP allow-lists, rate limits and payload caps. → API keys

Sandboxed user SQLite connections, confined to a resolved directory with traversal and internal-database denial.

Console recovery without a back door. A lost administrator password and a lost database key are both recoverable by somebody with the server's data directory, and by nobody else. Every reset is audited and surfaced where an administrator will see it. → Console recovery

AI assistant

Propose-and-approve. The assistant drafts views, pipelines, alarms and queries as proposals a human approves. Actions are audited under both the assistant and the approving user. → AI assistant

Skills, realms and fences. What an assistant may do is the intersection of its skills, where it runs, what the administrator switched on, and what the person asking already holds — and each instance carries its own fences over tags, databases, external systems and specialists. → Skills & fences

Playbooks and specialists — a procedure written down once, and saved sub-assistants with their own fences and budgets. → Playbooks · Specialists

Per-connection access and read-only credentials, so the assistant can explore a production database without being able to change it.

Administration

Per-module, offline licensing with a modules page that locks rather than hides, and a licence that reaches every running service without a restart. → Licensing & modules

Backup and restore with whole-archive passphrase encryption, secret re-keying between machines, opt-in media, and a filename pre-flight that makes an archive portable between Linux, Windows and macOS. → Backup & restore

Project lifecycle — enable, disable, trash with a 30-day retention window, restore and purge. → Projects

Service control from the interface, with the supervisor honouring a deliberate stop. → Services

---

Upgrading

1. Take a backup — schema migrations are not reversible.
2. Stop the service.
3. Replace the binary.
4. Start it. The supervisor migrates the schema before spawning children.
5. Check the startup report, then verify connections and pipelines.

Coming from 1.0.5, two changes are worth checking after the first start:

• MQTT publishers drop to the change rate. Intended, and a large reduction — but if a subscriber was treating message arrival as a liveness signal, give the publisher a heartbeat.
• Exports no longer carry File Manager or uploaded files by default, so the next backup will be dramatically smaller than the last. Tick Include File Manager & uploaded files for a full disaster-recovery copy, or back those two folders up with your file-backup tool. → Media

Keep .env. Losing ENCRYPTION_KEY makes every stored credential unrecoverable, and losing this machine's database key store makes app.db unreadable — so run qubiq db-key create and file the code somewhere off the machine. → Console recovery

Check your licence covers the version

A licence names the versions it covers. Upgrading past that ceiling reports version not covered — not unlicensed — and the remedy is to install the version you own or extend the licence. → Licensing