Current

2026-10-22

Security and continuity: strong password policy, immutable audit log, security event alerts, one-year alarm history, daily automatic backup and configuration version history. Each can be switched on or off; only Root can change them.

New

Settings → Security: strong password policy (at least 10 characters; upper case, lower case, digit and special character) enforced on every password create, change and reset path.
Immutable audit and event logs: records cannot be deleted by hand and are kept at least 365 days; only expired records are purged automatically.
Security event alerts: lockout after repeated failed logins, user permission or role changes, security setting changes and rejected log deletions are sent to administrators by mobile push and e-mail.
Settings → Backup: daily automatic backup (NAS, external disk or local folder; number of copies to keep) and configuration version history with one-click rollback.
Audible alarm: a sound in the browser when a new alarm raises, switchable per user.
Server-side history export: all raw rows of the selected tags as CSV, Excel or XML, with no row cap.

Improvement

Alarm history in a new database: one year by default with no record limit (configurable); date range filter and CSV in the history tab. Existing history is migrated on first start.
Only Root can change the Security, Retention and Backup settings; they cannot be granted to any role or group.

3.5.3

2026-10-04

↓ Download

Device Library: Envest’s ready devices install from the web interface in one step, with channel, device, tags, calculated tags and screens.

New

Device Library (book icon in the project tree header and the Connectivity view): ready devices from the website are listed with categories, search and a product page; several can be selected and installed together. Only install details such as IP, port, unit ID, FINS nodes and station parameters are asked; tags, calculated tags and screens (with settings screens) come ready.
Conflict checks before installing: same device name, a channel name used by another driver and calculated tag names; a free station name is suggested automatically.
Import from file (.mhpkg.json): on servers without internet access, a package downloaded from the website or edited with an AI assistant installs the same way.

Improvement

MergenHub Mobile (with the app update): an “Add device” wizard in phone and server mode — device library, saved templates, file/QR or manual; Settings › About. In phone mode a device installs with a summary screen and tabbed settings screens made for phones.

Fix

Ready station screens (Korubin templates): on OCP110 and AquaCnt well/booster main screens the last settings button overflowed its panel; all buttons are now on one row and the panel fits the booster and tank main screens.
Omron OCP110/OCS110 settings screens no longer show empty boxes for parameters the panel does not have; empty rows and columns were removed from the safety tables.
Installing from the library creates one top-level screen per device (settings screens below it); the station name on the screen follows the new device name. Short names in the SCADA screen list no longer look indented like sub-screens.
In MergenHub Mobile, a server screen opened in the design editor when the phone was held in landscape or on a tablet; the in-app view is now always the operator view.
System notifications: restarting the scan engine by hand (for example to renew the trial window) wrongly sent “service failed and was restarted automatically” notifications to phones; now only an automatic recovery of a crashed service is reported.

3.5.2

2026-09-16

↓ Download

The server now watches its own health: resource pressure, service crashes and unexpected shutdowns reach authorised users’ phones as push notifications.

New

System health notifications: when a CPU, memory, disk or thread threshold is crossed (sustained, not short spikes), when a service crashes and is restarted, and when the server comes back after a power cut, crash or forced stop, authorised users (Runtime control or Settings) with mobile access and notifications enabled get a push notification; recovery is reported too. The same alert is not repeated within 30 minutes.
All system alerts are written to the Event Log with the source “System Monitor” (Sistem İzleyici).
A “Send mobile push” option next to the warning thresholds on the Server Resources panel (on by default).

Improvement

MergenHub Mobile (with the app update): server resources and service status screen; chat with the AI assistant when it is enabled on the server; live values while picking tags in the screen editor; automatic HTTPS/HTTP detection when adding a server by domain name.

3.5.1

2026-09-04

↓ Download

MergenHub Mobile now shows SCADA screens exactly as the web interface does (Koru1000 backgrounds, symbols and sub-screen buttons included). Mobile push joins email and SMS under Notifications, sessions can be closed from the web, and a ready configuration file is generated for reverse proxies (Apache, nginx, IIS, Caddy).

New

Server screens in the mobile app look exactly as in the web interface: the app opens the server’s own SCADA viewer; backgrounds, symbols, animations and sub-screen buttons work. The connection uses a single-use, 60-second ticket; actions are logged in the audit trail with the “mobile” source.
Settings › Notifications › “Mobile push”: push setting, devices receiving push and, for unregistered devices, the reason (e.g. notification permission not granted on the phone), per-device test, manual notification to one user or everyone.
“Sign out” in the active users list: closes a mobile device’s session or all web sessions of a user in one click. “Close my other sessions”: closes all your sessions except this browser (other browsers, the desktop app, phones).
“Ready proxy config” for publishing on the web: enter your domain and get an Apache, nginx, IIS (web.config) or Caddy file; copy or download it. A checklist shows what to set on the MergenHub side.

Improvement

Writing “https://” in the listen address made the desktop app unable to connect. The field is now chosen as “Who can connect” and “HTTP port”; old settings are corrected automatically. HTTPS is served on its own port.
“Allow HTTP from the network” now works: when off and HTTPS is on, HTTP is usable only on the server itself and everyone on the network uses HTTPS.
Removing and re-adding a server on the phone no longer piles up old device sessions; a new sign-in from the same device closes the old one.

3.5.0

2026-08-17

↓ Download

MergenHub Mobile (iOS and Android) connects to the server from a phone: the user signs in once with user name and password, the server issues a session token for that device, and the password is not stored on the phone. Administrators manage mobile access and device sessions on Settings › Web, REST API & Mobile; alarms arrive on the phone as push notifications. The mobile app needs no extra license.

New

MergenHub Mobile (iOS and Android): connect to the server from a phone. The user signs in once with user name and password; the server issues a session token for that device and the password is not stored on the phone. Permissions and channel/device access lists are the same as for the web user.
Settings › “Web, REST API & Mobile” page: allow mobile sign-in, mobile notifications for alarms and session lifetime; active users in the last 15 minutes (source web, mobile or API; IP and client); mobile device sessions (user, device, platform, app version, last seen, IP) with one-click sign-out; send a test notification. These settings are changed only from the web interface; the mobile app cannot change them.
Per-user mobile access and mobile notification switches. Turning off a user’s mobile access also closes their open mobile sessions; web access is not affected.
Push notifications to the phone for alarms (Expo Push). New “Mobile notification” channel in the alarm definition; notifications go to the devices of users who have the Alarm permission and access to the tag, and who have mobile notifications turned on. The server needs outbound HTTPS to exp.host.
New “Source” column in the audit log: shows whether an action came from the web interface, the mobile app, an API client or an internal service.
String tags in the REST API: stringValue field in the values response; the write endpoint accepts a number, text or true/false. Writing text is supported by the Modbus (40001.10H/L), MQTT and OPC UA client drivers.

Improvement

Security: the license status response shows the full license key only to Admin and Root users.
MergenHub Mobile needs no extra license; the server’s existing license applies. Existing REST API clients are not affected by these changes.

3.4.7

2026-07-30

↓ Download

Each driver now limits the number of channels it can open and the number of devices in one channel; the values match the industry-standard server’s driver limits and every one was exercised at the limit with simulators. The Siemens S7 driver moved to non-blocking I/O: many S7 connections now open in 3 seconds instead of 48.

New

Per-driver channel limits and per-channel device limits (e.g. Modbus TCP 1,024 channels / 8,192 devices, IEC 61850 256 / 256, MQTT 60 / 1,024, OMRON FINS 512 / 32). The limit is per driver, not per project. It is checked only when adding (UI, REST API, AI assistant, copy, import); when reached, the operation is rejected and the project is unchanged. A saved project is never rejected or trimmed when it loads.
New “Channel and Device Limits” section under User Guide › Connectivity; each protocol section shows its own limit, and the project transfer and REST API sections explain import rejection and the 400 response.

Improvement

The Siemens S7 driver no longer blocks threads: 1,024 S7 connections open in 3 seconds instead of 48, and 1,024 S7 devices in one channel refresh every 0.7 s instead of every 19 s. Telegrams, timeouts and error texts are unchanged.
S7 no longer opens an extra “ping” connection before connecting, so CPUs/CPs with few sessions do not lose one. The send timeout is now honoured (a send could previously hang indefinitely).

3.4.6

2026-07-18

↓ Download

Sahadan bildirilen dört kusur: kurulum sihirbazında kutular ve seçenekler yarım görünüyordu, kurulum bitiminde hiçbir şeye tıklanamıyordu, indirilen paketin doğrulanması dosyayı bulamıyordu ve uygulama penceresi masaüstünden ayrışmıyordu.

Fix

Kurulum sihirbazında kutular ve seçenekler yarım görünüyordu: açıklama metinleri yüksek DPI'da daha çok satıra sarınca sayfa taşıyordu; ayrıca Servis Başlatma sayfasında seçenek listesi sayfanın dibine kadar uzatıldığı için altındaki gecikme kutusu görünür alanın dışında kalıyordu. Sihirbaz büyütüldü, yeniden boyutlandırılabilir yapıldı ve liste altındaki alana yer açacak şekilde kısaltıldı.
Kurulum bitiminde hiçbir şeye tıklanamıyor, Windows hata sesi geliyordu (her kurulumda değil). Kurulum sonu bilgilendirmeleri ayrı bir pencere olarak açılıyor, alt süreçler ön plana geçtiğinde sihirbazın arkasında kalıyordu; sihirbaz o pencere yüzünden kilitli olduğu için her tıklama hata sesi çıkarıyordu. Bu bilgiler artık son ekranın metni.
"İndirilen paketi doğrula" dosyayı bulamıyordu ("Dosya bulunamadı: C:WINDOWSsystem32configsystemprofileDownloads..."). Doğrulama Windows servisinde çalıştığı için İndirilenler klasörü olarak sistem hesabının profilini görüyordu. Paket artık tüm kullanıcıların İndirilenler ve Masaüstü klasörlerinde aranıyor; bulunamazsa nerelere bakıldığı yazılıyor ve kutuya klasör yolu da yazılabiliyor.
Uygulama penceresi masaüstünden ayrışmıyordu: köşeler keskin, kenar belirsizdi. Uygulama kendi üst çubuğunu kullandığı için Windows 11 pencereyi kendiliğinden yuvarlamıyordu; köşe tercihi artık açıkça bildiriliyor. Windows 10 ve öncesinde davranış değişmez.

3.4.5

2026-06-30

↓ Download

Four field-reported interface defects: on small screens dialog windows were clipped at the top and bottom, top-bar icons overlapped each other, and the channel/device detail window did not open on double-click in Firefox. The “publishing on the web” section was also rewritten: a step-by-step guide for domain, self-signed, PFX and reverse proxy.

New

Publishing guide: three scenario cards on Settings › Server (local network only · published directly · behind a reverse proxy) plus a “Step-by-step publishing guide” button.
New “Publishing on the Web” section in User Guide › Installation & Deployment: whether a domain is needed, what a self-signed certificate does and does not do, when a PFX is required and how to produce one (win-acme, the OpenSSL command, exporting from Windows), the eight steps of direct publishing, a security checklist before going public, nginx · Apache · IIS reverse proxy samples and a common-problems table.

Improvement

A one-line distinction in the certificate section: self-signed only encrypts (browsers warn; it is for LAN/test), a PFX is a real certificate (for public hosting); the intermediate chain must be inside the PFX.
When the top bar runs out of room the “View mode” selector first drops its labels and then moves to the bottom of the sidebar; no button is hidden.

Fix

Dialog windows broke on small screens: when a window was taller than the screen its top and bottom fell outside the viewport and, because the backdrop did not scroll, the title and the Save button were unreachable. The content now scrolls inside the window and the title and buttons stay visible.
Top-bar icons overlapped: when space ran out the icon buttons shrank below their box and spilled over (at 820 px the alarm bell dropped to 19 px). Icons are no longer squeezed.
The channel/device detail window did not open on double-click in Firefox (it worked in Chrome). Double-click is now detected by time and position rather than on the row element itself.

3.4.4

2026-06-12

↓ Download

Modbus TCP socket close now defaults to FIN (same as Kepware); RST close became a device option. The field report “device behind a modem stops answering once MergenHub starts” was fully resolved in the field test. The false “EXPIRED — scanning stopped” row on the License page of licensed installations is fixed.

New

Device > Ethernet > “Socket Close Mode”: FIN (default, like Kepware) / RST. Single-session GSM/radio RTUs that keep a half-open session for minutes after FIN can select RST to restore the previous behaviour.

Improvement

Nothing else changed: timing, reconnect and block logic are identical to 3.4.3.

Fix

Connection-fragile Modbus TCP devices behind a modem/gateway: when MergenHub closed the socket with RST the modem considered the session still open and did not answer the next connection. Field test (2026-09-16): the FIN close resolved it at the root.
On licensed installations the License page showed “EXPIRED — scanning stopped” after 2 hours (visual only; scanning never stopped). The trial-window row is now shown only while actually in trial mode.

3.4.3

2026-05-31

↓ Download

Closes the second silent-assistant symptom with local models (LM Studio, Qwen3): when the model wrote its tool call into the reasoning channel the server returned empty content. Calls found in the reasoning channel are now turned into real tool calls; if content stays empty, the reasoning text is shown as the answer.

Improvement

Tool calls in the reasoning channel are recovered too: if the content has no tag, the reasoning text is scanned; the remaining reasoning stays reasoning. (Known serving-layer defect in LM Studio / llama.cpp / vLLM with Qwen3.x — a model that skips thinking writes everything into the think block, and the server’s tool parser only looks at content.)
On local models, empty content + non-empty reasoning → the reasoning becomes the answer; no empty bubble. Not applied to cloud providers.
“Says it, doesn’t do it” and “fabricated report” guards (the Cline / Roo Code “no tool used” approach): if the model says “now I’ll build the screen” and stops without a tool call, or claims “created / added” when no changing tool ran, the assistant nudges it into action and continues (at most twice per turn). Interim notes reach the model as user messages (local templates do not understand a mid-conversation system message).
Action rules in the system prompt: macros only on explicit request/approval; “done” only from tool results; live function addresses on the Simulator (SINE/RAMP/RANDOM…). A function name given in a type/function field becomes the address; a tag left on a static register gets a warning.
Value box (display) fits narrow boxes: the label drops, the font shrinks (30 px boxes used to show clipped numbers). Default tool budget 24→32.

Fix

The “let me load the skill first” silence and the raw <tool_call> tag left on screen, still seen after 3.4.2 (reasoning channel + visible-tool filter).
The well-and-tank scenario was verified end to end from an empty project with a local 9B model: 2 channels/devices, live SINE tags, well-pump-valve-flowmeter-tank line, live values, no macros.

3.4.2

2026-05-13

↓ Download

AI providers reworked against their real APIs. LM Studio, Ollama and similar local servers now work out of the box: the address is completed automatically, models are listed with their capabilities, load/unload from memory and thinking control are available, and tool calls a local model writes as plain text are recovered. Tool stumbles found in an end-to-end live run (a well-and-tank screen with Qwen 9B) were fixed.

New

LM Studio model management: per model, whether it is loaded in memory, context window, thinking modes, tool-calling and vision capability, file size; load into memory (with a 32K context) / unload from the list. Actions are written to the audit log.
Thinking control on local models: on Ollama, LM Studio, vLLM, llama.cpp and the like the “Thinking” selector is now effective; with “Off”, Qwen3/DeepSeek-R1 derivatives no longer think at length on every answer.
Plain-text tool-call recovery: calls a local model writes as raw template tags (Qwen XML, Hermes JSON, Llama python_tag, Mistral [TOOL_CALLS]) are turned into real tool calls; the raw tag never reaches the screen. When the tool budget runs out, a “work left unfinished” note is shown.

Improvement

Base URL completion: typing “localhost:1234” or “http://192.168.1.10:1234” adds the scheme and /v1 automatically; the test result shows the address actually used. Addresses that already have a path (Google, Cohere, Perplexity) are left alone.
Incompatible responses become clear errors: a 200 that is not OpenAI-shaped is reported with a hint instead of an empty list or a silent turn; a context overflow (n_keep ≥ n_ctx) is not retried and says what to do.
Assistant tools read a local model’s arguments leniently: numbers/quoted strings, a tags/list array, type/read_only synonyms and type names (Float, Word, Boolean…) are accepted; the error message shows the expected shape.
A freshly created device’s tag no longer reads “Bad” on first read: read_tag_values wakes the device, requests an immediate read and waits (up to 2 s) for a fresh sample.
All cloud providers verified with key-less probes (OpenAI, Anthropic, OpenRouter, Google, Mistral, DeepSeek, Groq, xAI, Together, Fireworks, Cerebras, Cohere); static model catalogue for Perplexity (its API has no /models endpoint).

Fix

The assistant did not work at all with LM Studio (empty model list, silence to “hello”, save error): missing /v1, rejection of scheme-less addresses and silent acceptance of empty responses — all three closed.
With thinking set to “Off” on LM Studio the model still thought; the value sent was also rejected by the server (the valid value is “none”).
When a local model wrote its tool call as plain text the assistant said “let me check” and went silent.

3.4.1

2026-04-25

↓ Download

Seven field-reported defects in the 3.4.0 macro editing experience: Ctrl+wheel no longer zooms the page, double-click no longer selects text, the tag feeding a JS-driven element is shown, a part can be highlighted from the tag list, the code editor formats one-line code, appearance fields accept input, and the tag browser opens on the current tag.

Improvement

Element mode now finds the tag behind a JS-driven part: getElementById/querySelector usage is traced, the feeding tag is shown under "Data of this element" and can be bound as data-mh-tag with one click.
Code editor: "Format" (splits one-line code into lines without changing meaning) and a "Wrap" option.
Target button in the tag list: the parts using that tag flash on the canvas.
The tag browser opens with the current tag’s branch expanded and the row highlighted.

Fix

Ctrl+wheel over the macro panel or editor zoomed the browser page; zoom now goes to the canvas.
Double-clicking a macro selected text (blue selection); text selection is off in pick mode and re-enabled for in-place editing.
Appearance fields (colour, font size, alignment, padding) could not be typed into: two focus-loss roots fixed — panel re-render and the size message from the sandbox.

3.4.0

2026-04-13

↓ Download

The macro (your own HTML object on a screen) stopped being a code box and became part of the screen: an SDK that reaches every system value, on-canvas element editing, and a layout engine that knows its box. Code fields are now a syntax-highlighted, full-screen, context-aware editor.

New

Macro SDK: a macro can now read active alarms, the event and audit logs, system resources (CPU/RAM/disk), connected OPC UA clients, device status and a device’s read/write diagnostics — 18 system data sources in total. The code stays sandboxed and never touches the network: the application asks the question, with your own session and only against read endpoints, so a macro can never see what you cannot. These screens need no field tags.
Element mode: double-click a macro to go inside. The part under the cursor is outlined and named, the part you click opens in the right panel, and the breadcrumb selects its container.
On-canvas editing: a small toolbar appears above the selected part — edit text in place (double-click), bold, font size, move up/down, duplicate, delete, bind a tag. Edits are written into the HTML in place, your code is not reformatted, and Ctrl+Z undoes them.
Blocks: row, column, card, value, status badge, heading, text and table blocks let you build a macro from parts. A newly added part scrolls into view and flashes briefly.
Ready-made system panels: alarm list, event log, system resources, OPC UA clients, device status and read/write log templates, one click each.
Code editor: syntax-highlighted, line-numbered and full-screen for HTML/CSS/JS. Typing MH.query offers the data list, assigning the result offers that type’s fields, and data-mh-tag="" completes real tags. No external dependency — it works the same on an air-gapped site.
SDK & examples panel: parameters, returned fields and a working example for every data type, with an "insert into JS" button. A Macro chapter was added to the User Guide.

Improvement

Layout engine: the macro box has a fixed size and long content used to be clipped. There are now four modes — Fit box (content is scaled, never clipped; default for new macros), Flow, Fill height, and Box follows content (the screen page grows too when needed). Scaling never touches your source.
Template buttons no longer overwrite existing content; they append, and the behaviour is switchable in the panel.
MH.query now unwraps list responses itself (it returns the array directly, with root fields such as clientCount/total/enabled attached) — in a live trial the wrapper was missed and produced a panel that claimed "no events".

Fix

The macro error box now lists the ids present in the document for "null" errors, so a missing id is obvious at a glance.
The "Add tag" button only added an empty row and did nothing on a second click; it now opens the tag browser and highlights the added row.
An escape bug in the sandbox script silently disabled the safety net that rescues selectors written without "#" — fixed.

3.3.1

2026-03-26

↓ Download

Field-reported defects: the device TCP port now really reaches the wire, the submersible pump body no longer spins, and the ready-made well scene uses the new drawings. On the assistant side, reading a screen shrank by 75% and tool search was added — token cost dropped noticeably.

New

Tool search (find_tools): the assistant describes the job in its own words and only the matching tools enter the conversation. Opening a tool group used to pull in 20-30 tool definitions at once; tools from external MCP servers now go through the same search.
Part targeting in symbol animation: rotation can be aimed at the impeller/rotor inside a symbol.
Connector directory: adding an external system no longer means filling in a blank address row. There is a searchable, categorised directory — Figma (screen mock-ups and symbols), Notion (plant wiki, commissioning procedures), Jira/Confluence and Linear (fault tickets, work orders), Asana (maintenance work), Canva (panel labels, report covers), another MergenHub (a second site) and your own MCP server (ERP, database bridge). Cards are permanent: clicking one connects without asking which method, and “disconnect” only clears the credentials. Every address in the directory was verified live before release.
Connect with OAuth: providers such as Figma, Notion, Jira and Linear do not issue long-lived API keys; you now authorise in the browser (PKCE + discovery + silent refresh) and access is refreshed automatically. Token authentication remains for your own servers.
Macros: where the built-in objects are not enough you can place your own HTML/CSS/JS on a screen (a site-specific table, a custom indicator, a small calculator) — and the assistant can write one when needed. The code runs sandboxed, cannot reach the application session and downloads nothing from outside, so it behaves identically on an air-gapped plant network. It reads only the tags it declares, and writing goes through the normal permission and approval path.

Improvement

Reading a screen shrank by 75%: the raw drawing source of every symbol was being sent along and stayed in the conversation history, so it was resent every turn. It is trimmed now; when the assistant genuinely needs the source it can request a single object in full.
Reasoning is on by default and the assistant step budget went from 8 to 14, so multi-step setups no longer stop half way with “I will do this next”.
Creating or editing a device now returns the address, port and scan settings as actually stored, and the assistant cannot report success without comparing them against what you asked for.
The skills UI was reworked: search and filtering, starting from a template (site standard, device type, report format), Markdown write/preview, duplicate, and .md import/export.

Fix

On Windows Server the application stopped at “WebView2 runtime not found”: the Microsoft Edge WebView2 component the interface runs in ships with Windows 11 and recent Windows 10 but not with Windows Server, and the installer never installed it. The offline installer for the runtime is now inside the package, so it also installs on air-gapped servers; where it is already present nothing is touched. (This is why the setup file grew.)
The device TCP port never reached the wire: it was written to a key the driver does not read, so the connection silently used 502. It is now written to the driver’s own key.
When no port was given the driver default settled in silently; this is now reported explicitly.
The submersible pump spun around its own axis: symbols with no rotatable part no longer get a rotation, and running state is shown by colour instead.
The ready-made well scene still used the old drawings (an upright valve); it now goes through the new scene engine — horizontal flanged valve, shared ground line, and a connection that re-routes itself when you move an object.
The application closing after pressing F5 on a SCADA screen: the viewer component could crash while reloading a heavy screen, leaving a dead window. The screen now reloads itself, and when it cannot recover it says what happened and what to do.

3.3.0

2026-03-08

↓ Download

Project templates and cross-industry screen design: package a solution you built and install it at the next site from a single dialog, and let the assistant build screens from water/wastewater to electrical single-line diagrams. Plus multiple SMS providers, account-bound profile preferences, and canvas and scheduling improvements in the automation editor.

New

Project templates: package a solution (channel + device + tag structure + SCADA screen) in one click. Site-specific values such as IP, port, channel and device names become parameters asked at install time, so the same solution is deployed elsewhere from a single dialog.
Templates in the New Screen dialog: cards grouped by category, search, and each template’s own logo, vendor and version. Koru1000 templates sit alongside with their own card.
Template files (.mhtpl): export a template to share it, import someone else’s. An imported template never overwrites yours; built-in templates are read-only and editing one creates your own copy.
Installing shows a preview first — what will be created, with nothing changed yet. A template never deletes your existing project; on a name clash the default is to rename.
The AI assistant understands templates: it installs a matching template instead of drawing from scratch, asks for missing parameters instead of inventing them, and offers to package a solution once it is finished.
SMS providers: İleti Merkezi, Verimor and Twilio join NetGSM; the “Other” option remains a free-form template for any REST gateway.
My profile: every user manages their own email/phone and notification preferences (in-app / email / SMS), and sees their role, group, permissions and last login.
Scheduling wizard in the automation editor: set up “every X minutes”, “daily at HH:MM”, “weekdays”, “selected days of the week” or “day X of the month” without writing a cron expression, with a plain-language summary and the next five run times.
Cross-industry screen design: the assistant no longer computes coordinates — it lists the equipment in order and scale, placement, ground line and connecting run are derived automatically. The catalogue holds 50 equipment types and 25 ready-made stations: water/wastewater, chemical, boiler room, cooling, compressed air and ventilation, silo and bulk material, filling and packaging, surface finishing and paint, food processing, and electrical single-line diagrams.
Sub-agent: long research jobs such as scanning the symbol library or taking inventory now run in a separate context and return only a summary to the main conversation. It is read-only (it cannot change anything) and is presented for approval up front with the task, the step budget and a cost notice.
Appearance and preferences in the profile: display name, interface language, theme and start page are stored on your account. They used to live in the browser, so they were lost when you signed in from another computer.
New "During work hours" pattern in the schedule wizard: run every N minutes throughout a shift, night shifts (22:00→05:59) included, without writing cron.

Improvement

Templates carry no secrets: passwords/keys, server settings, alarm history and logged data never enter the package — a template is made to be shared.
NetGSM now defaults to the XML (POST) API, removing the truncation/encoding issues long messages and Turkish characters hit on the GET API (GET remains selectable).
SMS numbers are normalised to 90XXXXXXXXXX, and provider error codes are translated into actionable messages instead of raw codes.
User management now has a Contact button on every row, so an administrator can add or correct a user’s email/phone later.
Installing and packaging templates sit behind the same permission as full project import; a resource-restricted user cannot install templates.
Live connections: move a screen object and the pipe or cable re-routes itself; endpoints can be dragged on the canvas and carry arrow, coupling or flange ends plus flow animation. The run is drawn behind the objects.
Symbol selection is now visual: the assistant sees the candidates side by side in one image (drawing, name, horizontal/vertical orientation) and picks from that — which is why valves no longer stand upright on a horizontal run.
Equipment sits on a shared ground line, so tanks are no longer half-buried and valves no longer float. The run passes exactly through the inlet of the tank at the end of the line.
In the automation canvas the run panel starts collapsed for a flow that has never run, leaving more room for the diagram; it opens by itself when you press Run.

Fix

The automation canvas now pans 1:1 — the drawing moves exactly as far as the mouse, at any zoom level. Wheel zoom keeps the point under the cursor fixed, and fit/100% buttons plus two-finger pinch zoom were added.
The TCP port you entered when adding a device could silently fall back to the default (502). There is now a dedicated port field, and when the port arrives from more than one place with different values it is no longer guessed — you are asked which one is correct.
The schedule summary spelled out expressions such as "every 5 minutes between 08 and 18" by listing the hours one by one, which made it look as if nothing ran in between. It now reads as an hour window plus a frequency.
In the screen preview two symbols sharing a gradient id could overwrite each other, making a tank appear gold or streaked green.

3.2.1

2026-02-24

↓ Download

Automation workflows and an in-app AI assistant: build rules by drag and drop, run the system in plain language — with permissions, approvals and audit built in.

New

Automation module: build trigger → condition/calculation → action flows by drag and drop. Schedule (interval or cron), tag value (threshold, deadband, dwell, cooldown), alarm and webhook triggers; write-tag, e-mail, SMS, HTTP request, event log and alarm acknowledge actions. Flows are evaluated on the server every second.
Formula editor: click-to-insert symbol chips, live syntax checking, the tag’s current value and a trial result. Runs are traced step by step; when a flow cannot finish you see the step it stopped at and the error.
In-app AI assistant: reads the project, creates devices, tags, alarms, screens and workflows, and opens the matching page. Cloud (OpenAI, Anthropic, OpenRouter…) or local models (Ollama, LM Studio) — with local models no data leaves the machine.
Assistant setup wizard: enter the key and models, context lengths and prices are fetched automatically; tokens and cost are shown per conversation.
Permissions and audit: the assistant can never exceed your own permissions; every change goes through an approval card, no data reaches a cloud provider before the data-processing consent is given, and every approval is written to an immutable compliance ledger.
Two optional capabilities, off by default: running Python/PowerShell in a sandbox (separate low-privilege account, no access to the data folder, network allowlist, emergency stop) and — with permission — seeing the current screen.

Improvement

New capabilities: MCP bridge, assistant script execution (granted only by the system administrator) and assistant/MCP security administration.
When the assistant changes something the open page refreshes itself; approval cards show the entity name instead of its id.
Desktop app: window buttons no longer overlap the assistant panel, narrow-window layout and untranslated property names fixed.
Assistant conversations now get a short model-written title instead of a truncated first sentence.
Adding a device or channel now asks for the driver's REQUIRED fields: IP for Modbus TCP, master and outstation addresses for DNP3, common address for IEC 104, rack/slot for S7, COM port and baud for serial lines, broker for MQTT, endpoint for OPC UA. The assistant asks instead of inventing them.
New devices default to the tag-defined scan rate at 1000 ms; existing devices are unaffected.

Fix

After restarting the runtime, the remaining trial window refreshes without reloading the page.

3.2.0

2026-02-06

↓ Download

Automation workflows and an in-app AI assistant: build rules by drag and drop, run the system in plain language — with permissions, approvals and audit built in.

New

Automation module: build trigger → condition/calculation → action flows by drag and drop. Schedule (interval or cron), tag value (threshold, deadband, dwell, cooldown), alarm and webhook triggers; write-tag, e-mail, SMS, HTTP request, event log and alarm acknowledge actions. Flows are evaluated on the server every second.
Formula editor: click-to-insert symbol chips, live syntax checking, the tag’s current value and a trial result. Runs are traced step by step; when a flow cannot finish you see the step it stopped at and the error.
In-app AI assistant: reads the project, creates devices, tags, alarms, screens and workflows, and opens the matching page. Cloud (OpenAI, Anthropic, OpenRouter…) or local models (Ollama, LM Studio) — with local models no data leaves the machine.
Assistant setup wizard: enter the key and models, context lengths and prices are fetched automatically; tokens and cost are shown per conversation.
Permissions and audit: the assistant can never exceed your own permissions; every change goes through an approval card, no data reaches a cloud provider before the data-processing consent is given, and every approval is written to an immutable compliance ledger.
Two optional capabilities, off by default: running Python/PowerShell in a sandbox (separate low-privilege account, no access to the data folder, network allowlist, emergency stop) and — with permission — seeing the current screen.

Improvement

New capabilities: MCP bridge, assistant script execution (granted only by the system administrator) and assistant/MCP security administration.
When the assistant changes something the open page refreshes itself; approval cards show the entity name instead of its id.
Desktop app: window buttons no longer overlap the assistant panel, narrow-window layout and untranslated property names fixed.

3.1.30

2026-01-19

↓ Download

Missing glyphs fixed; the MCP server now appears in AI clients with its name and icon.

Improvement

The MCP server now reports its product name and icon to AI clients (ChatGPT, Claude, LM Studio); previously it was listed under its raw code name with a placeholder icon.

Fix

A few icons rendered as an empty box or question mark (Users › "Unrestricted" mode, OPC UA client "Scan Network" and "Subscribe to all"). Every icon name in the application was audited; all are valid now.
The MCP server reported a hardcoded version "1.0.0" to clients; it now reports the actual application version.

3.1.29

2026-01-07

↓ Download

Update check after an upgrade fixed; the application now starts itself when setup finishes.

Improvement

Setup now offers "Start the MergenHub application" on the final page, ticked by default; previously only the system tray was started.
Licence reminder when uninstalling: nothing to do if you reinstall on the same machine (the licence is never deleted); if you are moving to another machine you are told to release it first.

Fix

After upgrading, "Check for updates" still reported the previous version. The result is cached for 24 hours and that cache survives an upgrade; the cache is now invalidated when the installed version changes, so the portal is queried immediately. The "Installed" value is read from the running build.

3.1.28

2025-12-20

↓ Download

Maintenance & Support gate: a version can be installed only if it was released on or before your maintenance end date. If maintenance lapses you can still download every version it covered, indefinitely.

New

Version archive in the customer panel: each release is marked "Included" or "Needs renewal", and the newest version you can install is shown at the top.
Maintenance & Support can now be added or extended with one click from the licence own page — for every package.

Improvement

Setup never starts for a version your licence does not cover: it shows both dates and what to do, leaving your running system and licence untouched.
"Verify / refresh from portal" is always available in Settings › License. After renewing maintenance you pull the updated document here; with no internet, download it from the portal and use "Import updated document".
If a build newer than your maintenance end is already running, the system is not stopped: it warns persistently and only stops honouring the licence on a restart 30 days later.

3.1.27

2025-12-02

↓ Download

Reveal the license key in the app, resize the window from its edges, and keep the license through a clean install.

New

Show/hide the license key in Settings › License and copy it with one click.

Improvement

The app window can now be resized from every edge and corner.
Maintenance & Support can now be added or extended with one click from the licence's own page (for every package).
Rich text editor when creating a support request: paste screenshots directly, drag-and-drop attachments, formatting.

Fix

A clean install no longer removes your license — the activated license document and machine key are preserved and the license stays active.

3.1.23

2025-11-20

↓ Download

Desktop window fix: clicking a top-bar button no longer shrinks the window to the center.

Fix

Clicking a status chip/button in the top bar no longer restores a maximized window to a small centered size: dragging now begins only after the pointer moves past a threshold, so a plain click never moves the window.

3.1.22

2025-11-02

↓ Download

The desktop app is now a true native window: the OS title bar is gone and controls are built into the app.

New

Seamless window: the duplicate 'MergenHub' title is gone; minimize/maximize/close are built into the app's top bar (theme-aware), with drag and double-click-to-maximize from the bar.
'Open in browser' moved from a floating button to the top-right cluster (no longer overlaps content on SCADA etc.) and can be turned off from Settings > Web & REST API (locked/kiosk use).

Improvement

Maximizing no longer overflows under the taskbar (constrained to the work area).

3.1.21

2025-10-15

↓ Download

Anonymous, opt-in telemetry (trial mode only) and small UI improvements.

New

Anonymous telemetry: only when UNLICENSED (trial) and ONLY with your consent, the app may send anonymous data in the background (version, approximate tag count, OS, a reproducible anonymous install id) to help improve the product. No personal/hardware info, MAC, or your project data is sent; telemetry stops entirely once a license is entered; consent can be withdrawn from Settings > License.
Flags on the sign-in language switch (TR/EN).

Improvement

Telemetry runs in the background on the existing daily update check; if the portal is unreachable the app is unaffected.

3.1.20

2025-10-03

↓ Download

Desktop app (chromeless WebView2 window) and unlimited trial: until a license is entered, ALL features and UNLIMITED tags are available within a 2-hour runtime window.

New

Desktop app: the local UI now opens in its own window without a browser bar. Desktop and Start-menu shortcuts are added.
One-click 'Open in browser' from inside the shell (bottom-right button + right-click menu + Ctrl+B); the Administration tray opens it 'as an app' or 'in the browser' separately.
Redesigned sign-in screen (brand colors, TR/EN, show password) with a 'Remember me' option.
The native title bar follows the UI theme (light/dark).

Improvement

Trial mode is now UNLIMITED tags: the 500-tag cap for unlicensed runtime was removed; the only limit is the 2-hour runtime window.
The update check gives instant feedback ('checking...', result).
Clean install now also removes old leftover files.

3.1.19

2025-09-15

↓ Download

Single installer: licensing and trial mode. Until a license key is entered, all features are available with 500 tags and a 2-hour runtime window.

New

Licensing: online activation (key) or offline activation (.mhreq request file -> portal -> .mhlic document) from Settings > License; release / move a license to another machine.
Packages differ only by tag count (500 / 5,000 / 25,000 / 50,000 / unlimited); every feature is available in every package.
Trial mode: unlicensed installs get 500 tags and a 2-hour runtime window per runtime start; on expiry a top-bar notice offers a one-click new window.
Daily update check with a "new version available" balloon (per-version "do not show again").

Improvement

The BETA package and the regular package were merged into a single installer. In-place upgrades keep existing settings.
The setup wizard shows a licensing/trial notice.

3.1.18

2025-08-28

↓ Download

The SCADA flow animation now works on every object type and on all Envest symbols.

Improvement

Pipe and polyline-pipe objects honour the flow rule directly (no need to also enable the "Flow animation" checkbox).
The animation preview (⚡) now plays every rule assigned to an object in the design view.
Flow direction can now be set to right/left/up/down; the pipe stripe reaches the ends.
Direction and dash/gap settings of an animation rule were lost after saving; they are now kept.
Pipe link ends and bend points can be dragged on the canvas (double-click to add/remove bends); ends snap to the object shape.
Choosing "Pipe (draw)" from the palette opens the drawing tool directly.

Critical fix

A "Flow" animation assigned to Envest pipe, elbow, valve and pump symbols never ran; it now flows inside the symbol shape.
On other objects the flow stripe ignored colour/speed/direction and was invisible on the light theme; it now runs with the chosen colour and speed, clipped to the object shape.

3.1.17

2025-08-16

↓ Download

Device error state is now consistent everywhere: a device shown red in the tree is reported as faulted to OPC UA clients too.

Improvement

Short connection drops still do not make the status flag flap — the decision is based on the time of the last valid data rather than the instantaneous connection state.

Critical fix

An unreachable device could report "no error" over OPC UA while the interface showed it as faulted; both sides now agree.

3.1.16

2025-07-29

↓ Download

Device status (_System) keeps publishing regularly, including throughout a communication outage.

Improvement

Device status values are republished at regular intervals even when unchanged, so a client recovers on its own if it misses a notification.

Critical fix

When communication was lost, device status could stay stuck at its previous value in the OPC UA client; the change now reliably reaches the client.

3.1.15

2025-07-11

↓ Download

OPC UA values no longer drop to zero when communication is lost; the last valid value is held.

New

New OPC UA project property: "Keep last value when communication is lost" (on by default).

Critical fix

When communication was lost, OPC UA clients received zero as the value; the quality is now reported as "Bad" while the last valid value is held, so screens and trends no longer zero out.

3.1.14

2025-06-29

↓ Download

Immediate, accurate device status when communication is lost, plus time/type in the tag table.

Fix

In the tag table, system rows left the Time column empty and showed the same Type on every row.
The blue info frame in the screen editor could spill over the panels in some situations.

Critical fix

When a device address (IP) was changed to an unreachable one, the error flag could stay clear for minutes; it is now raised on the first failed connection.

3.1.13

2025-06-11

↓ Download

Illustrator-style selection in the screen editor, an animation list, a richer REST client and an expanded manual.

New

Animation list: every animation on the screen (template bindings included) in one panel; clicking a row locates and highlights the object on canvas.
Data Bridge REST client gains authentication options (Basic, Bearer, API key), the PATCH method, an example-driven help section and a "send test request" button.
The manual gains a "Koru1000 Screen Templates" chapter and coverage of the new features.

Improvement

Selection inside drawings is now stepwise: double-click enters one level, clicking outside steps back out (Illustrator flow).
Tag fields gain a one-click "clear selection" button; long tag paths show in full on hover.
The installer can reveal the admin password, and a generated strong password is copied to the clipboard with one click.

Fix

The selection frame could linger across page switches and window/panel movements, spilling over the panels; it now stays aligned to the drawing area at all times.
Single click and right-click on drawing parts could stop responding after certain edits.

3.1.9

2025-05-24

↓ Download

Device health indicator accuracy, Turkish name lookups and screen editor fixes.

Fix

Channel and device names containing the Turkish I/i letters could cause "device not found" errors in some operations.
Subscribing to the health tag alone did not start scanning the device; monitoring now polls the device as soon as it begins.
In the screen editor, the selection frame could linger over the panels after panning or zooming.

Critical fix

The device health indicator (_NoError) could report "no error" on some protocols even when communication was down; it now reflects actual data flow.

3.1.8

2025-05-12

↓ Download

Selection frames spilled over the panels in the screen editor.

Fix

When a part of an object was selected, its frame spilled over the left palette and right property panel as the object neared the edge; it is now clipped to the drawing area.
The part name label also stays inside the drawing area: it moves below the object when there is no room above, and disappears when the selection is cleared.

3.1.7

2025-04-24

↓ Download

The AI (MCP) connection can now use the whole system; three permission holes closed.

New

18 new tools for the AI connection: service start/stop and crash recovery, audit log, settings, project export and import, tag discovery from a device, user and permission management.
New permission groups: Service Control, Audit Log, Settings, Users - each grantable on its own.
Privilege-escalation guard on the new user management tools: the AI cannot grant any permission its token owner lacks; such a request is refused, not silently trimmed.
Project import previews by default: nothing changes unless you explicitly confirm.

Fix

Disabling a user did not fully cut their AI access: a few tools that need no permission kept working.
Two tools (set animation, place on pipe) belonged to no permission group and could never be called.
In the well and booster tag pools the open/closed status always read 0 (bad scaling).
The screen wizard now warns which calculation a station parameter left at 0 will zero out.

3.1.6

2025-04-06

↓ Download

Pipe stretching and layer order fixes.

New

Pipes now have a direction option: Automatic, Horizontal or Vertical.
An explicit Unrestricted / Restricted switch for OPC UA tag access: a restriction can be undone in one click.
Switching to Unrestricted keeps your rules; switching back to Restricted brings them back unchanged.
Bulk access: tick channels and devices, then set a level, reset to inherit, or delete every rule at once.
Apply one level to every device in a channel from the channel row.

Fix

Empty margins at pipe ends when stretching are gone; a pipe now fills its box exactly.
A horizontal pipe stretched vertically no longer collapses to a thin line.
Bring to front and send to back now work on pipe links as well.
Bringing a part inside a symbol to the front did not work.
Changing a part color no longer corrupts the symbol gradients.

3.1.5

2025-03-25

↓ Download

Project export/import, and a data-loss fix when importing a single device.

New

The whole project can be exported and imported as one file: channels, devices, tags, virtual tags, modules and settings.
Merge and replace modes on import, with a preview and conflict handling (skip, rename, overwrite).
Passwords and keys are not exported by default.

Fix

Importing a single channel or device no longer wipes the existing project.
Some files silently appeared empty on import.
Top navigation and narrow-screen layout on the website were fixed.

3.1.4

2025-03-07

↓ Download

Scaling, calculated-tag visibility and scan-mode fixes; manual tank height from the screen.

New

Values without a device tag, such as tank height, can be entered manually on the screen or bound to a virtual tag.

Fix

Scaling settings were lost in template tag pools; scaled tags now report correct values.
The scan mode chosen while adding a device was not applied, causing values to drop in and out.
The calculated tag list appeared empty after installing a template.
Some device settings were saved but never applied.

3.1.2

2025-02-17

↓ Download

Ready-made screen templates, a complete OPC UA server profile and per-user tag authorisation.

New

Koru1000 Screen Templates: ready screens for well, tank, booster, DMA and PRV; device, tag pool and calculated values are created in one step.
The same screen can be used with different devices (CP110, AquaCnt, Omron).
A template can be installed as a screen only, then bound to your existing tags.
Per-user tag authorisation on the OPC UA server: read and read/write can be granted separately.
Historical data access, server methods and alarm publishing over OPC UA.

Improvement

Ready-made logging trigger scenarios (for example, log while a motor runs).
Buffered data status is now visible when the target database is unreachable.
Live preview for table format and column names.
Screen templates are an optional component during setup.

Fix

Virtual tag data type selection was corrected; decimal values are no longer rounded.
Certificate selection in OPC UA security settings was fixed.
Object selection, animation and flow behaviour in the screen editor were improved.

3.0.9

2025-02-05

↓ Download

A broad quality pass across alarms, Data Bridge, Modbus and security.

Improvement

Modbus device settings were simplified and documented.

Fix

Alarm state tracking and the handling of invalid measurement values were corrected.
Notification (e-mail and SMS) failures are now visible.
Data transfer during connection outages was improved.

Security

Authorisation hardening applied; upgrading is recommended.
Sensitive information is less exposed and access restrictions are applied consistently.

3.0.8

2025-01-18

↓ Download

Stability and calculation accuracy for Advanced Tags.

Improvement

Formula validation was tightened.

Fix

Advanced tags now run continuously.
Handling of invalid calculation results and number formats was corrected.

Critical fix

Resilience against faulty formulas was increased.
Persistence of totalizer and counter values was improved.

3.0.7

2024-12-31

↓ Download

Reliable data logging and broad database compatibility.

Improvement

MySQL and MariaDB version compatibility was extended.

Fix

Logging errors that occurred at certain times of day were resolved.
Issues with custom column and table names were resolved.
Timestamp precision was improved.

Security

Database operations were hardened.

3.0.6

2024-12-19

↓ Download

The data-logging settings screen was redesigned.

New

Trigger settings moved to a card layout, each summarised in a single sentence.
Day selection and column mapping screens were simplified.

Improvement

Verified in light and dark themes; your existing settings are preserved exactly.

3.0.5

2024-12-01

↓ Download

Critical fix for the appearance of SCADA symbols.

Improvement

Saved screens repair themselves the next time they open — nothing to redraw.

Critical fix

Shape and colour distortion seen on some symbols was resolved.

3.0.4

2024-11-13

↓ Download

IEC 61850 connection resilience.

Improvement

An extra safety net was added for situations where no data is returned.

Critical fix

Connections now recover automatically — no service restart required.

3.0.3

2024-11-01

↓ Download

Critical fix for symbol appearance on SCADA screens.

Improvement

Saved screens repair themselves; screen data is untouched.

Critical fix

Adding a new object no longer distorts the objects already on screen.

3.0.2

2024-10-14

↓ Download

Driver resilience and the complete data-logging feature set.

New

Multiple triggers, column mapping and table modes in data logging.

Improvement

Every field driver was tested for recovery after an outage without a restart.
MQTT multi-broker behaviour was improved.

Critical fix

IEC 61850 connection stability was improved.

3.0.1

2024-09-26

↓ Download

Communication status indicator and connection settings.

Improvement

OPC UA Client and IEC 61850 connection settings are applied in full.

Fix

The communication status indicator now reflects the actual data flow.

3.0.0

2024-09-14

↓ Download

The first official product release of MergenHub.

New

Nine field drivers: Modbus, Siemens S7, OMRON, MQTT, DNP3, IEC 60870-5-104, IEC 61850 and OPC UA Client.
OPC UA Server with multiple endpoints and security policies.
Built-in SCADA/HMI editor, Historian, Data Bridge and ISA-18.2 alarms.
User, group and resource level authorisation, plus an MCP/AI server.

Improvement

Verified scale: 450,000+ tags; single-step installation through the Windows setup wizard.

2.4.8

2024-08-27

↓ Download

Closing release of the 2.4.x series.

Improvement

The "Cumulative Difference" feature was renamed "Delta Counter".

2.4.7

2024-08-09

↓ Download

Database writes and alarm delay.

New

Alarm tolerance (on-delay) time and network scanning for database server discovery.

Critical fix

An error on the database write path was fixed.

2.4.4 – 2.4.6

2024-07-28

↓ Download

Historian database discovery.

New

Server scanning, database listing and table creation from inside the application.

Improvement

Live validation for every database type, with plain-language error messages.

2.4.3

2024-07-10

↓ Download

SCADA runtime fixes.

Improvement

The conveyor symbol was redesigned.

Fix

An input field issue in run mode was fixed.
Alarm clean-up was improved.

2.4.1 – 2.4.2

2024-06-22

↓ Download

Fixes for data corruption caused by scaling.

Critical fix

Copying of scaling settings and the conversion applied on writes were corrected.

2.3.x

2024-06-10

↓ Download

The OPC UA Client driver was added.

New

Encrypted sessions to remote OPC UA servers, endpoint discovery and selective tree import.

2.2.0

2024-05-23

↓ Download

Communication continuity fix.

Critical fix

A communication outage seen on some devices after tag changes was fixed.

What changed, version by version

Where do I get older versions?