node-red-contrib-knx-ultimate 6.4.1 → 7.0.0

package/CHANGELOG.md

@@ -6,50 +6,53 @@
 
 # CHANGELOG
 
-**Version 6.4.1** - September 2026<br/>
+**Version 7.0.0** - September 2026<br/>
 
-- **KNX AI — safe Cerebrum startup delivery**: corrected the RedBot Telegram envelope used by startup and other proactive messages after a Node-RED restart. Its synthetic chat context now fulfils RedBot's synchronous context contract instead of returning `undefined`, preventing `chat-platform` from raising an uncaught `sourceCode` error and terminating Node-RED. Saved RedBot adapter code is hardened at runtime as well. Schedule persistence now passes the generated Markdown filename to the atomic writer under the correct `filePath` key, removing the accompanying undefined-path warning.<br/>
-- **KNX AI — Cerebrum (BETA)**: renamed the Web Assistant area to Cerebrum and introduced a passive home-automation intelligence layer that studies KNX telegrams, useful Node-RED flow events and the registered HUE, Matter and Home Assistant adapters. Cerebrum maintains bounded current-state and observation context locally, learns only repeated patterns spanning distinct days and asks the occupant for confirmation or correction through the Assistant output before a habit can become predictive. Confirmed habits remain subject to the exact ETS/DPT permissions: group addresses marked read-only are never written.<br/>
+- **Major release — focused KNX package**: KNX Ultimate 7 starts the new package boundary. Conversational AI, supervised home learning, shared memory and multi-platform chat now belong to the standalone `node-red-contrib-cerebrum-ultimate`; this package remains the optional KNX/ETS integration. Existing deployed flows that contain the hidden legacy runtime types continue to load, but new instances can no longer be added from the palette.<br/>
+- **Cerebrum Ultimate documentation handoff**: removed the obsolete bundled-assistant guides, samples, dashboard pages and public labels. The README, multilingual homepages, wiki navigation, localized legacy-node help and tutorial roadmap now point to the standalone `node-red-contrib-cerebrum-ultimate` package. The hidden legacy runtime identifiers remain registered only so existing flows continue to load.<br/>
+- **Cerebrum standalone extraction**: the historical `knxUltimateAI` and `knxUltimateAIHomeAssistant` types remain registered so existing deployed flows continue to load and can be edited, but they are now hidden from the Node-RED palette. New Cerebrum installations move to the independent `node-red-contrib-cerebrum-ultimate` package, where KNX Ultimate is an optional integration alongside Home Assistant, HUE, Matter and other adapters.<br/>
+- **Cerebrum Ultimate — safe Cerebrum startup delivery**: corrected the RedBot Telegram envelope used by startup and other proactive messages after a Node-RED restart. Its synthetic chat context now fulfils RedBot's synchronous context contract instead of returning `undefined`, preventing `chat-platform` from raising an uncaught `sourceCode` error and terminating Node-RED. Saved RedBot adapter code is hardened at runtime as well. Schedule persistence now passes the generated Markdown filename to the atomic writer under the correct `filePath` key, removing the accompanying undefined-path warning.<br/>
+- **Cerebrum Ultimate — Cerebrum (BETA)**: renamed the Web Assistant area to Cerebrum and introduced a passive home-automation intelligence layer that studies KNX telegrams, useful Node-RED flow events and the registered HUE, Matter and Home Assistant adapters. Cerebrum maintains bounded current-state and observation context locally, learns only repeated patterns spanning distinct days and asks the occupant for confirmation or correction through the Assistant output before a habit can become predictive. Confirmed habits remain subject to the exact ETS/DPT permissions: group addresses marked read-only are never written.<br/>
 - **Cerebrum — Home Assistant adapter**: added the `Cerebrum Home Assistant` node and the shared runtime registry. When Node-RED is running as a Home Assistant add-on, Setup Doctor detects whether the required `ha-api` round trip is present and tells the user to add it when missing. Entity states, events and service capabilities are exposed to Cerebrum through bounded, vendor-neutral records without retaining raw messages, secrets or binary payloads.<br/>
-- **Cerebrum — supervised startup**: every KNX AI node now asks its configured model to generate a short startup reassurance on the Assistant output. The message carries `msg.boot = true` for simple flow filtering and falls back to a localized deterministic notice if the provider is unavailable.<br/>
-- **KNX AI Web — Cerebrum workspace and readable memory**: reorganized the Web navigation into Cerebrum submenus, removed the separate KNX AI node selector and bound the workspace directly to the node that opened it. Cerebrum Memory now offers an editable authoritative JSON view and a read-only simplified view; AI Chat Learning offers the equivalent editable native `.knxctx` view and localized simplified view. Copy follows the visible representation, while save, restore, download and concurrency checks continue to protect the complete authoritative files.<br/>
+- **Cerebrum — supervised startup**: every Cerebrum Ultimate node now asks its configured model to generate a short startup reassurance on the Assistant output. The message carries `msg.boot = true` for simple flow filtering and falls back to a localized deterministic notice if the provider is unavailable.<br/>
+- **Cerebrum Ultimate Web — Cerebrum workspace and readable memory**: reorganized the Web navigation into Cerebrum submenus, removed the separate Cerebrum Ultimate node selector and bound the workspace directly to the node that opened it. Cerebrum Memory now offers an editable authoritative JSON view and a read-only simplified view; AI Chat Learning offers the equivalent editable native `.knxctx` view and localized simplified view. Copy follows the visible representation, while save, restore, download and concurrency checks continue to protect the complete authoritative files.<br/>
 - **Cerebrum — complete import/export**: the Settings page now contains only import/export and its versioned backup includes AI configuration, chat learning, home memory, schedules and the readable schedules companion file, with validation and rollback-safe restoration.<br/>
 
 **Version 6.3.32** - August 2026<br/>
 
-- **KNX AI**: AI-first chat with complete cloud ETS context, selectable local windows and guided retrieval; selected ETS access is now the sole read/write authority, with no neutral role.<br/>
+- **Cerebrum Ultimate**: AI-first chat with complete cloud ETS context, selectable local windows and guided retrieval; selected ETS access is now the sole read/write authority, with no neutral role.<br/>
 
 **Version 6.3.31** - August 2026<br/>
 
-- **KNX AI — adaptive prompts for 8K local models**: the conversational prompt now reserves response space from the active model window before every request. For an 8K model it uses a compact tool contract and automatically bounds the newest conversation turns, KNX/adapter events, home memory, Web results, schedules, Function source, retrieved ETS objects and camera metadata. The complete selected ETS catalog remains local and available through deterministic retrieval; it is not copied wholesale into the prompt. The prompt-debug file records the final bounded system/user text actually sent to the model.<br/>
-- **KNX AI — model-guided local ETS retrieval**: the conversational model now receives `catalogActions` for deterministic searches over the complete selected ETS catalog held inside the node instead of receiving the full catalog in every prompt. Bounded multi-pass retrieval supports ranked search across exact GAs, ETS names, aliases, hierarchy, areas, semantics, DPTs and value labels; accent-insensitive and typo-tolerant matches; exact-address lookup; area discovery/browsing; and related command/status discovery. No socket or second classifier is used: LM Studio returns structured requests, KNX AI executes them locally and calls the same model again with a deduplicated subset of at most 24 objects. Commands and GA-role learning are accepted only for retrieved objects and still pass the complete local ETS/DPT/access validation. The final prompt debug file exposes the retrieval summaries and exact retrieved subset.<br/>
-- **KNX AI — complete ETS with deduplicated operational context**: removed the 4K/8K/16K/unlimited context-size selector and its persisted runtime property while keeping the complete selected ETS catalog locally searchable from every conversation. The implicit KNX and adapter archive range is now 20 minutes instead of 24 hours; an explicit user time range still takes precedence. Packaged help, README, wiki, examples and changelog text are no longer embedded in prompts and can instead be retrieved from public GitHub through the request-driven Web tool. Retrieved ETS metadata, archive rows and the current request each have one canonical prompt location; duplicated archive summaries, ETS-derived area lists, per-GA live snapshots and the always-on Node-RED project inventory were removed. Complete Function source is supplied only for explicit Function-code review requests. A node-specific, locally protected `knxai-last-chat-prompt-<node-id>.txt` debug file is overwritten before every conversational model call with the exact system/user prompt text and its estimated size, excluding API keys and HTTP headers; its path appears in the editor context card. Ollama receives the model's full declared window as `num_ctx`; KNX AI never retries an oversized request by compacting its input.<br/>
-- **KNX AI — explicit ETS catalog access and read-only policy**: added the same filterable group-address selector used by IoT Bridge’s MQTT profile, with Select all/none and bulk or per-row Read only controls. Every selected ETS object is available to conversational retrieval, while Flow Builder keeps its complete selected design inventory. Excluded objects are absent from AI tools, while read-only objects remain searchable and readable but are rejected locally for every write even if their inferred or learned role is command. Complete ETS names, hierarchy paths and duplicate-address aliases are preserved in the local catalog and retrieved object records. There is intentionally no backward-compatibility migration: an existing node without a newly saved explicit ETS selection exposes no group addresses until it is opened, configured, saved and deployed.<br/>
-- **KNX AI — request-driven Web access with clarification-first tool use**: removed the fixed proactive Web polling option, interval selector, background timer, synthetic recipient routing and related Setup Doctor state. During a normal conversation the model now decides semantically whether the current request needs fresh public information; when a missing goal, scope, place, time window or desired outcome would materially change the answer or query, it must ask one concise clarification and emit no tool action until the user answers. Explicit one-time or recurring Web checks remain available through the user-authorized natural-language scheduler and share the existing hourly budget, bounded research loop, verified citations and security controls. Editor strings, help, wiki documentation and tests were aligned in EN, IT, DE, FR, ES and zh-CN.<br/>
+- **Cerebrum Ultimate — adaptive prompts for 8K local models**: the conversational prompt now reserves response space from the active model window before every request. For an 8K model it uses a compact tool contract and automatically bounds the newest conversation turns, KNX/adapter events, home memory, Web results, schedules, Function source, retrieved ETS objects and camera metadata. The complete selected ETS catalog remains local and available through deterministic retrieval; it is not copied wholesale into the prompt. The prompt-debug file records the final bounded system/user text actually sent to the model.<br/>
+- **Cerebrum Ultimate — model-guided local ETS retrieval**: the conversational model now receives `catalogActions` for deterministic searches over the complete selected ETS catalog held inside the node instead of receiving the full catalog in every prompt. Bounded multi-pass retrieval supports ranked search across exact GAs, ETS names, aliases, hierarchy, areas, semantics, DPTs and value labels; accent-insensitive and typo-tolerant matches; exact-address lookup; area discovery/browsing; and related command/status discovery. No socket or second classifier is used: LM Studio returns structured requests, Cerebrum Ultimate executes them locally and calls the same model again with a deduplicated subset of at most 24 objects. Commands and GA-role learning are accepted only for retrieved objects and still pass the complete local ETS/DPT/access validation. The final prompt debug file exposes the retrieval summaries and exact retrieved subset.<br/>
+- **Cerebrum Ultimate — complete ETS with deduplicated operational context**: removed the 4K/8K/16K/unlimited context-size selector and its persisted runtime property while keeping the complete selected ETS catalog locally searchable from every conversation. The implicit KNX and adapter archive range is now 20 minutes instead of 24 hours; an explicit user time range still takes precedence. Packaged help, README, wiki, examples and changelog text are no longer embedded in prompts and can instead be retrieved from public GitHub through the request-driven Web tool. Retrieved ETS metadata, archive rows and the current request each have one canonical prompt location; duplicated archive summaries, ETS-derived area lists, per-GA live snapshots and the always-on Node-RED project inventory were removed. Complete Function source is supplied only for explicit Function-code review requests. A node-specific, locally protected `knxai-last-chat-prompt-<node-id>.txt` debug file is overwritten before every conversational model call with the exact system/user prompt text and its estimated size, excluding API keys and HTTP headers; its path appears in the editor context card. Ollama receives the model's full declared window as `num_ctx`; Cerebrum Ultimate never retries an oversized request by compacting its input.<br/>
+- **Cerebrum Ultimate — explicit ETS catalog access and read-only policy**: added the same filterable group-address selector used by IoT Bridge’s MQTT profile, with Select all/none and bulk or per-row Read only controls. Every selected ETS object is available to conversational retrieval, while Flow Builder keeps its complete selected design inventory. Excluded objects are absent from AI tools, while read-only objects remain searchable and readable but are rejected locally for every write even if their inferred or learned role is command. Complete ETS names, hierarchy paths and duplicate-address aliases are preserved in the local catalog and retrieved object records. There is intentionally no backward-compatibility migration: an existing node without a newly saved explicit ETS selection exposes no group addresses until it is opened, configured, saved and deployed.<br/>
+- **Cerebrum Ultimate — request-driven Web access with clarification-first tool use**: removed the fixed proactive Web polling option, interval selector, background timer, synthetic recipient routing and related Setup Doctor state. During a normal conversation the model now decides semantically whether the current request needs fresh public information; when a missing goal, scope, place, time window or desired outcome would materially change the answer or query, it must ask one concise clarification and emit no tool action until the user answers. Explicit one-time or recurring Web checks remain available through the user-authorized natural-language scheduler and share the existing hourly budget, bounded research loop, verified citations and security controls. Editor strings, help, wiki documentation and tests were aligned in EN, IT, DE, FR, ES and zh-CN.<br/>
 
 **Version 6.3.30** - August 2026<br/>
 
-- **KNX AI — two-way RedBot Telegram voice chat**: the `RedBot / node-red-contrib-chatbot (Telegram)` preset now accepts RedBot's native inbound `audio` payload and its already-downloaded OGG/Opus `Buffer`, applies the same 20 MB and five-minute safeguards without downloading the file a second time, and transcribes it through the selected OpenAI-compatible provider. Successful requests receive a native RedBot `audio` reply with the localized AI-generated-voice disclosure and text caption; KNX confirmations deliberately remain RedBot `inline-buttons` text messages because its Telegram voice sender cannot attach that keyboard to `sendVoice`. Existing saved RedBot input/output mappings are upgraded at runtime. The separate `windkh/node-red-contrib-telegrambot` voice contract and its tests remain unchanged.<br/>
+- **Cerebrum Ultimate — two-way RedBot Telegram voice chat**: the `RedBot / node-red-contrib-chatbot (Telegram)` preset now accepts RedBot's native inbound `audio` payload and its already-downloaded OGG/Opus `Buffer`, applies the same 20 MB and five-minute safeguards without downloading the file a second time, and transcribes it through the selected OpenAI-compatible provider. Successful requests receive a native RedBot `audio` reply with the localized AI-generated-voice disclosure and text caption; KNX confirmations deliberately remain RedBot `inline-buttons` text messages because its Telegram voice sender cannot attach that keyboard to `sendVoice`. Existing saved RedBot input/output mappings are upgraded at runtime. The separate `windkh/node-red-contrib-telegrambot` voice contract and its tests remain unchanged.<br/>
 
 **Version 6.3.29** - August 2026<br/>
 
-- **KNX AI — patient provider-neutral model requests**: every LLM chat request now receives the same 30-minute minimum deadline, regardless of provider, model or model manager, while longer legacy configured values remain honored. The model transport carries that deadline through both streaming and non-streaming responses instead of inheriting a shorter HTTP-library header timeout; fast responses are returned immediately, model-response timeouts remain distinct from unreachable-server errors, and Telegram audio operations retain their separate two-minute safeguard.<br/>
-- **KNX AI — provider-neutral reasoning control and responsive chat**: added a localized **Reasoning effort** selector with provider/model default, `none`, `minimal`, `low`, `medium`, `high`, `xhigh` and `max` choices. The default sends no preference; explicit supported values are mapped by request protocol to Chat Completions `reasoning_effort`, Anthropic `output_config.effort` or Ollama `think`, without model-name allowlists, and a rejected preference is removed while preserving the rest of the request. OpenAI-compatible and Ollama chats now stream responses for every configured endpoint, keep reasoning traces separate from the structured user reply and retry without streaming when an endpoint rejects it. Abort and HTTP transport timeouts are reported as model-response timeouts instead of as false connection failures; genuine unreachable-server errors retain their provider-specific setup guidance. Context-error compaction remains in force. Editor help and wiki documentation were aligned in EN, IT, DE, FR, ES and zh-CN.<br/>
-- **KNX AI — optional model-sized local context**: added a localized **No KNX AI limit** choice to the local-model chat-context selector. The existing 4K, 8K and 16K budgets are unchanged and 16K remains the backward-compatible default; when the new choice is explicit, KNX AI uses the active/reported Ollama or Bionic LM Studio context window instead of imposing its own token cap, while the model's physical limit and the existing context-error fallback remain authoritative.<br/>
-- **KNX AI editor — quieter Setup Doctor placement**: moved Setup Doctor to the bottom of the **Chat and home** page, below the available-chat-context overview, and made it an accessible collapsed-by-default section so routine configuration stays compact while every diagnostic and refresh action remains available on demand.<br/>
-- **KNX AI — semantic persistent chat scheduling and fixed AI Education**: added model-selected `scheduleActions` for creating, listing and cancelling one-time or recurring reminders, monitors and future home commands from normal human-language chat, without keywords, trigger phrases or rigid intent classifiers. Each node persists authoritative schedule state in JSON and a generated readable Markdown view across restarts; due tasks reuse both the model's execution instruction and the exact original user request while retaining Web budgets, detected-camera limits, explicit TTS Ultimate wiring, ETS/DPT validation and configured KNX confirmation. Empty structured replies from smaller local models receive one general semantic recovery pass instead of surfacing the old “No response text” placeholder. Scheduled camera results are delivered back to sidebar chat, and cancelling a plan also invalidates any pending camera work or KNX confirmation belonging to it. `reset` clears all schedules for the node, while `clear_chat` cancels only the current session's schedules. AI Education remains a fixed, user-edited node property stored in the Node-RED flow: the model reads it as authoritative guidance but never writes it; learned chat memory stores facts and preferences requested in chat, while scheduler storage holds plans and reminders. The previously introduced legacy `knxai-education-<node-id>.md` file is no longer read or written; existing copies are left untouched and ignored. Localized editor strings, help and wiki pages were aligned in EN, IT, DE, FR, ES and zh-CN.<br/>
+- **Cerebrum Ultimate — patient provider-neutral model requests**: every LLM chat request now receives the same 30-minute minimum deadline, regardless of provider, model or model manager, while longer legacy configured values remain honored. The model transport carries that deadline through both streaming and non-streaming responses instead of inheriting a shorter HTTP-library header timeout; fast responses are returned immediately, model-response timeouts remain distinct from unreachable-server errors, and Telegram audio operations retain their separate two-minute safeguard.<br/>
+- **Cerebrum Ultimate — provider-neutral reasoning control and responsive chat**: added a localized **Reasoning effort** selector with provider/model default, `none`, `minimal`, `low`, `medium`, `high`, `xhigh` and `max` choices. The default sends no preference; explicit supported values are mapped by request protocol to Chat Completions `reasoning_effort`, Anthropic `output_config.effort` or Ollama `think`, without model-name allowlists, and a rejected preference is removed while preserving the rest of the request. OpenAI-compatible and Ollama chats now stream responses for every configured endpoint, keep reasoning traces separate from the structured user reply and retry without streaming when an endpoint rejects it. Abort and HTTP transport timeouts are reported as model-response timeouts instead of as false connection failures; genuine unreachable-server errors retain their provider-specific setup guidance. Context-error compaction remains in force. Editor help and wiki documentation were aligned in EN, IT, DE, FR, ES and zh-CN.<br/>
+- **Cerebrum Ultimate — optional model-sized local context**: added a localized **No Cerebrum Ultimate limit** choice to the local-model chat-context selector. The existing 4K, 8K and 16K budgets are unchanged and 16K remains the backward-compatible default; when the new choice is explicit, Cerebrum Ultimate uses the active/reported Ollama or Bionic LM Studio context window instead of imposing its own token cap, while the model's physical limit and the existing context-error fallback remain authoritative.<br/>
+- **Cerebrum Ultimate editor — quieter Setup Doctor placement**: moved Setup Doctor to the bottom of the **Chat and home** page, below the available-chat-context overview, and made it an accessible collapsed-by-default section so routine configuration stays compact while every diagnostic and refresh action remains available on demand.<br/>
+- **Cerebrum Ultimate — semantic persistent chat scheduling and fixed AI Education**: added model-selected `scheduleActions` for creating, listing and cancelling one-time or recurring reminders, monitors and future home commands from normal human-language chat, without keywords, trigger phrases or rigid intent classifiers. Each node persists authoritative schedule state in JSON and a generated readable Markdown view across restarts; due tasks reuse both the model's execution instruction and the exact original user request while retaining Web budgets, detected-camera limits, explicit TTS Ultimate wiring, ETS/DPT validation and configured KNX confirmation. Empty structured replies from smaller local models receive one general semantic recovery pass instead of surfacing the old “No response text” placeholder. Scheduled camera results are delivered back to sidebar chat, and cancelling a plan also invalidates any pending camera work or KNX confirmation belonging to it. `reset` clears all schedules for the node, while `clear_chat` cancels only the current session's schedules. AI Education remains a fixed, user-edited node property stored in the Node-RED flow: the model reads it as authoritative guidance but never writes it; learned chat memory stores facts and preferences requested in chat, while scheduler storage holds plans and reminders. The previously introduced legacy `knxai-education-<node-id>.md` file is no longer read or written; existing copies are left untouched and ignored. Localized editor strings, help and wiki pages were aligned in EN, IT, DE, FR, ES and zh-CN.<br/>
 
 **Version 6.3.27** - August 2026<br/>
 
-- KNX AI: enabled some things by default.
+- Cerebrum Ultimate: enabled some things by default.
 
 **Version 6.3.26** - August 2026<br/>
 
-- **KNX AI — opt-in Web Intelligence with verified sources**: added a provider-independent general Web tool that the conversational model selects semantically from the current request, without topic-specific logic, keywords or intent classifiers. Web access is off by default; proactive Web checks require a separate opt-in, explicit user-authored AI Education, a destination learned from at least one normal chat request and a configurable 5, 10, 15, 30, 60 or 180-minute minimum interval. Setup Doctor reports Web/proactive readiness, missing Education or recipient and the live hourly budget. Interactive and proactive requests share a rolling hourly call budget, with at most three Web operations in total per user turn or proactive cycle. Web-backed answers carry runtime-validated numbered citations (`[S1]`, …) with sanitized URLs and available publication/retrieval times; external content remains untrusted data and cannot override assistant rules or permissions. Only bounded public HTTPS resources are accepted, with private/local/link-local/cloud-metadata targets, unsafe redirects, authenticated browsing and cookies blocked. Queries and the server's public IP are disclosed to external sites or the search service, while KNX/ETS data, camera content, chat identifiers, learned memory and credentials are never added automatically. Once verified Web results are available, the model can compose other enabled tools when authorized by the current chat or AI Education; existing tool availability, KNX permissions, ETS/DPT validation and configured write confirmation remain unchanged. Editor help and wiki documentation were aligned in EN, IT, DE, FR, ES and zh-CN.<br/>
-- **KNX AI — clearer chat wiring terminology**: renamed the editor labels for chat input/output pins, message adapters and automatically detected compatible nodes in all six shipped languages, including the related empty/loading/error states and localized help/wiki references.<br/>
-- **KNX AI — automatic Setup Doctor and personalized safe first run**: added an editor and Web UI preflight that checks the selected gateway and ETS import, AI/provider/model/API-key setup, provider reachability, chat and KNX-operation wiring, detected cameras and optional TTS Ultimate wiring. The provider check calls only its model-list endpoint, never sends a chat request and consumes no inference tokens; unused camera and TTS integrations do not reduce core readiness. Inventory wording now distinguishes the exact number of unique KNX group-address signals from estimated logical functions and deliberately avoids an unreliable physical-device count. Sending `/start` or `/help` produces a deterministic localized welcome on output 3 with installation-specific statistics and up to three safe suggestions, without calling the LLM, reading or writing KNX, or generating TTS; the Telegram preset renders those suggestions as an explicit reply keyboard. After explicit selection, a starter suggestion may perform exact KNX reads when needed, while KNX writes and routines, camera actions, TTS, persistent-memory changes and GA-role learning are suppressed. Help, wiki and the direct Telegram example were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
-- **KNX AI — explicit TTS Ultimate flow output**: replaced the internal **TTS Ultimate node for announcements** selector and direct node injection with a dedicated fifth output labelled **TTS Ultimate**. Each model-selected announcement now leaves the node through normal Node-RED wiring with the exact spoken text in `msg.payload`, `msg.topic = "knx_ai_announcement"` and `msg.knxAi` metadata for type, source node, session and reason; normal fan-out or Link Out/Link In can route it to one or more `ttsultimate` nodes. Output positions 1–4 remain unchanged. Existing flows that relied on the saved `ttsUltimateNodeId` selection must physically connect output 5 after upgrading; the old selection is no longer used. Help, wiki pages and importable KNX AI examples were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
-- **KNX AI — two-way Telegram voice chat**: the `windkh/node-red-contrib-telegrambot` preset now accepts native OGG/Opus voice messages, downloads them through the receiver-provided Telegram link with same-origin, timeout, 20 MB and five-minute safeguards, redacts the token-bearing URL, and transcribes them through OpenAI-compatible `/audio/transcriptions`. Successful voice requests receive native Telegram OGG/Opus replies synthesized by `/audio/speech`, including a localized AI-generated-voice disclosure, the text caption and confirmation keyboard; failed or oversized synthesis falls back to the complete text answer. Voice processing is intentionally plug-and-play only with the selected **OpenAI-compatible** chat provider: it reuses that provider's endpoint and API key with built-in `gpt-4o-mini-transcribe`, `gpt-4o-mini-tts` and `alloy` defaults. The separate Telegram voice-service section, credential and model fields were removed. Anthropic, Ollama and Bionic LM Studio remain available for text chat; a voice request with one of them returns a localized provider-selection instruction before audio download. Existing saved preset mappings are upgraded at runtime, the direct Telegram example now ships with the preset selected, and help/wiki documentation covers provider and privacy requirements in EN, IT, DE, FR, ES and zh-CN.<br/>
+- **Cerebrum Ultimate — opt-in Web Intelligence with verified sources**: added a provider-independent general Web tool that the conversational model selects semantically from the current request, without topic-specific logic, keywords or intent classifiers. Web access is off by default; proactive Web checks require a separate opt-in, explicit user-authored AI Education, a destination learned from at least one normal chat request and a configurable 5, 10, 15, 30, 60 or 180-minute minimum interval. Setup Doctor reports Web/proactive readiness, missing Education or recipient and the live hourly budget. Interactive and proactive requests share a rolling hourly call budget, with at most three Web operations in total per user turn or proactive cycle. Web-backed answers carry runtime-validated numbered citations (`[S1]`, …) with sanitized URLs and available publication/retrieval times; external content remains untrusted data and cannot override assistant rules or permissions. Only bounded public HTTPS resources are accepted, with private/local/link-local/cloud-metadata targets, unsafe redirects, authenticated browsing and cookies blocked. Queries and the server's public IP are disclosed to external sites or the search service, while KNX/ETS data, camera content, chat identifiers, learned memory and credentials are never added automatically. Once verified Web results are available, the model can compose other enabled tools when authorized by the current chat or AI Education; existing tool availability, KNX permissions, ETS/DPT validation and configured write confirmation remain unchanged. Editor help and wiki documentation were aligned in EN, IT, DE, FR, ES and zh-CN.<br/>
+- **Cerebrum Ultimate — clearer chat wiring terminology**: renamed the editor labels for chat input/output pins, message adapters and automatically detected compatible nodes in all six shipped languages, including the related empty/loading/error states and localized help/wiki references.<br/>
+- **Cerebrum Ultimate — automatic Setup Doctor and personalized safe first run**: added an editor and Web UI preflight that checks the selected gateway and ETS import, AI/provider/model/API-key setup, provider reachability, chat and KNX-operation wiring, detected cameras and optional TTS Ultimate wiring. The provider check calls only its model-list endpoint, never sends a chat request and consumes no inference tokens; unused camera and TTS integrations do not reduce core readiness. Inventory wording now distinguishes the exact number of unique KNX group-address signals from estimated logical functions and deliberately avoids an unreliable physical-device count. Sending `/start` or `/help` produces a deterministic localized welcome on output 3 with installation-specific statistics and up to three safe suggestions, without calling the LLM, reading or writing KNX, or generating TTS; the Telegram preset renders those suggestions as an explicit reply keyboard. After explicit selection, a starter suggestion may perform exact KNX reads when needed, while KNX writes and routines, camera actions, TTS, persistent-memory changes and GA-role learning are suppressed. Help, wiki and the direct Telegram example were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
+- **Cerebrum Ultimate — explicit TTS Ultimate flow output**: replaced the internal **TTS Ultimate node for announcements** selector and direct node injection with a dedicated fifth output labelled **TTS Ultimate**. Each model-selected announcement now leaves the node through normal Node-RED wiring with the exact spoken text in `msg.payload`, `msg.topic = "knx_ai_announcement"` and `msg.knxAi` metadata for type, source node, session and reason; normal fan-out or Link Out/Link In can route it to one or more `ttsultimate` nodes. Output positions 1–4 remain unchanged. Existing flows that relied on the saved `ttsUltimateNodeId` selection must physically connect output 5 after upgrading; the old selection is no longer used. Help, wiki pages and importable Cerebrum Ultimate examples were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
+- **Cerebrum Ultimate — two-way Telegram voice chat**: the `windkh/node-red-contrib-telegrambot` preset now accepts native OGG/Opus voice messages, downloads them through the receiver-provided Telegram link with same-origin, timeout, 20 MB and five-minute safeguards, redacts the token-bearing URL, and transcribes them through OpenAI-compatible `/audio/transcriptions`. Successful voice requests receive native Telegram OGG/Opus replies synthesized by `/audio/speech`, including a localized AI-generated-voice disclosure, the text caption and confirmation keyboard; failed or oversized synthesis falls back to the complete text answer. Voice processing is intentionally plug-and-play only with the selected **OpenAI-compatible** chat provider: it reuses that provider's endpoint and API key with built-in `gpt-4o-mini-transcribe`, `gpt-4o-mini-tts` and `alloy` defaults. The separate Telegram voice-service section, credential and model fields were removed. Anthropic, Ollama and Bionic LM Studio remain available for text chat; a voice request with one of them returns a localized provider-selection instruction before audio download. Existing saved preset mappings are upgraded at runtime, the direct Telegram example now ships with the preset selected, and help/wiki documentation covers provider and privacy requirements in EN, IT, DE, FR, ES and zh-CN.<br/>
 
 **Version 6.3.25** - August 2026<br/>
 
@@ -58,59 +61,59 @@
 
 **Version 6.3.24** - August 2026<br/>
 
-- **KNX AI — working Telegram confirmation buttons**: changed the `windkh/node-red-contrib-telegrambot` preset from inline callbacks to a one-time Telegram reply keyboard. **Confirm** and **Cancel** now return as normal localized messages through the already connected `telegram receiver`, eliminating the silent no-op caused by a missing `callback_query` event node. Existing saved preset output mappings are upgraded at runtime, the keyboard is removed after the decision, and legacy callback events remain accepted. The importable direct-chat example no longer requires extra callback wiring; help and wiki documentation were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
-- **KNX AI — native CHAT learning memory**: replaced the verbose Markdown/JSON V2 learning file with the direct line-oriented `knxai-chat-context.knxctx` V3 format. Sessions, instructions, turns and camera watches are stored as editable tab-separated `SESSION`, `INSTRUCTION`, `TURN`, `CAMERA_WATCH` and `END_SESSION` records with escaped text fields, without JSON serialization or a regenerated preview. The Web UI keeps copy, download, restore, revision-safe save and protected reinitialization. Previous `.md` Markdown/JSON V2 and Base64 V1 files are deliberately not read, imported, migrated or deleted; KNX AI starts a new native context instead. Help and wiki documentation were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
-- **KNX AI — compact LLM settings layout**: constrained the assistant connection form to a readable responsive width and removed the fixed 290 px indentation that allowed the local-context hint and status line to widen the complete Node-RED properties panel. The model autocomplete now reopens after mouse release, so the full provider list remains visible after a normal click instead of appearing only while the button is held.<br/>
-- **KNX AI — resettable CHAT learning**: added a protected **Reinitialize Memory** action to **Settings → AI Chat Learning**. After explicit confirmation it atomically replaces the shared learning file with a new empty native context, clears sessions, persistent instructions, camera watches and pending chat confirmations across every KNX AI node using the same storage, and refreshes the editor immediately. Concurrent changes remain protected by the existing content revision check.<br/>
-- **KNX AI — complete model picker**: clicking the model field now opens the complete list returned by the selected provider regardless of the text already present, while typed text can still filter the same list and custom model names remain accepted.<br/>
-- **KNX AI — native compact context archives**: replaced JSONL serialization for KNX telegram and detected-adapter history with KNX AI's direct line-oriented `.knxctx` format. Typed payloads, flags and bounded adapter metadata round-trip without repeated JSON property names, while bus and archive summaries are rendered as compact `key=value` context instead of JSON before reaching the model. Existing JSONL archives are deliberately not read or migrated; new daily files use `YYYY-MM-DD.knxctx`. The user-editable CHAT learning file remains separate because it is also a portable user backup surface. Context cards, editor help and wiki documentation were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
-- **KNX AI — user-selectable local chat context**: added a localized **Chat context amount** selector for Ollama and Bionic LM Studio with 4K, 8K and 16K choices; 16K remains the default. The selected budget proportionally bounds KNX traffic, archives, ETS catalog, Node-RED inventory, memory and adapter context, and is also sent as Ollama's `num_ctx`, while the complete agent tool contract stays available without phrase classifiers, keyword gates or linguistic intent routing. The context overview now reports the configured operational limit. Editor help and wiki documentation were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
-- **KNX AI — reliable conversational recall**: verified that session IDs and recent turns were already persisted correctly, then moved the current session memory next to the current request in the operational prompt so small local models do not lose user-supplied facts inside large KNX context. The structured memory tool now explicitly covers durable user facts such as preferred name and language as well as preferences and instructions, while remaining model-selected with no phrase classifiers or intent routing; credentials, security codes and API keys are excluded.<br/>
-- **KNX AI — editable and portable CHAT learning**: added **Settings → AI Chat Learning** to the Vue web UI for viewing the exact absolute path and editing, copying, downloading or restoring the shared learning file. The **Conversations & home** tab in the Node-RED node configuration includes a localized shortcut that opens the Web UI directly on this editor for the current node. Saves are validated, bounded to 512 KB, written atomically and applied immediately to every KNX AI node sharing the same storage, while revision checks prevent stale editors from overwriting newer learning. UI strings, editor help and wiki documentation were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
-- **KNX AI — intent-free agentic chat tools**: removed hard-coded semantic phrase classifiers from conversational routing. Every Ask/CHAT request now lets the model reason directly over the current goal, persistent user guidance, AI Education and the available structured KNX, camera, TTS Ultimate and memory tools. Routine planning no longer depends on a multilingual keyword list, TTS announcements no longer require an action alias, and durable chat instructions are remembered or forgotten through model-selected structured actions instead of regular-expression intent detection. Runtime argument validation, camera-adapter availability checks, ETS/DPT safeguards and command confirmation remain unchanged.<br/>
-- **KNX AI — learned KNX group-address roles**: a `neutral` GA is now treated as unresolved rather than permanently uncontrollable. The conversational model can learn or forget the exact `command`, `status` or `neutral` role from trusted user teaching, persistent chat guidance, AI Education or unequivocal ETS semantics through the structured `gaRoleActions` tool. Learned role, reason and evidence are persisted per node in `knxai-config-<node-id>.json`, synchronized into bounded semantic home memory and reused after restart. A command role learned in the current answer can validate a write in that same turn, while exact ETS membership, immutable DPT, payload validation and configured confirmation still apply. Help and wiki documentation were aligned in EN, IT, DE, FR, ES and zh-CN.<br/>
+- **Cerebrum Ultimate — working Telegram confirmation buttons**: changed the `windkh/node-red-contrib-telegrambot` preset from inline callbacks to a one-time Telegram reply keyboard. **Confirm** and **Cancel** now return as normal localized messages through the already connected `telegram receiver`, eliminating the silent no-op caused by a missing `callback_query` event node. Existing saved preset output mappings are upgraded at runtime, the keyboard is removed after the decision, and legacy callback events remain accepted. The importable direct-chat example no longer requires extra callback wiring; help and wiki documentation were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
+- **Cerebrum Ultimate — native CHAT learning memory**: replaced the verbose Markdown/JSON V2 learning file with the direct line-oriented `knxai-chat-context.knxctx` V3 format. Sessions, instructions, turns and camera watches are stored as editable tab-separated `SESSION`, `INSTRUCTION`, `TURN`, `CAMERA_WATCH` and `END_SESSION` records with escaped text fields, without JSON serialization or a regenerated preview. The Web UI keeps copy, download, restore, revision-safe save and protected reinitialization. Previous `.md` Markdown/JSON V2 and Base64 V1 files are deliberately not read, imported, migrated or deleted; Cerebrum Ultimate starts a new native context instead. Help and wiki documentation were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
+- **Cerebrum Ultimate — compact LLM settings layout**: constrained the assistant connection form to a readable responsive width and removed the fixed 290 px indentation that allowed the local-context hint and status line to widen the complete Node-RED properties panel. The model autocomplete now reopens after mouse release, so the full provider list remains visible after a normal click instead of appearing only while the button is held.<br/>
+- **Cerebrum Ultimate — resettable CHAT learning**: added a protected **Reinitialize Memory** action to **Settings → AI Chat Learning**. After explicit confirmation it atomically replaces the shared learning file with a new empty native context, clears sessions, persistent instructions, camera watches and pending chat confirmations across every Cerebrum Ultimate node using the same storage, and refreshes the editor immediately. Concurrent changes remain protected by the existing content revision check.<br/>
+- **Cerebrum Ultimate — complete model picker**: clicking the model field now opens the complete list returned by the selected provider regardless of the text already present, while typed text can still filter the same list and custom model names remain accepted.<br/>
+- **Cerebrum Ultimate — native compact context archives**: replaced JSONL serialization for KNX telegram and detected-adapter history with Cerebrum Ultimate's direct line-oriented `.knxctx` format. Typed payloads, flags and bounded adapter metadata round-trip without repeated JSON property names, while bus and archive summaries are rendered as compact `key=value` context instead of JSON before reaching the model. Existing JSONL archives are deliberately not read or migrated; new daily files use `YYYY-MM-DD.knxctx`. The user-editable CHAT learning file remains separate because it is also a portable user backup surface. Context cards, editor help and wiki documentation were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
+- **Cerebrum Ultimate — user-selectable local chat context**: added a localized **Chat context amount** selector for Ollama and Bionic LM Studio with 4K, 8K and 16K choices; 16K remains the default. The selected budget proportionally bounds KNX traffic, archives, ETS catalog, Node-RED inventory, memory and adapter context, and is also sent as Ollama's `num_ctx`, while the complete agent tool contract stays available without phrase classifiers, keyword gates or linguistic intent routing. The context overview now reports the configured operational limit. Editor help and wiki documentation were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
+- **Cerebrum Ultimate — reliable conversational recall**: verified that session IDs and recent turns were already persisted correctly, then moved the current session memory next to the current request in the operational prompt so small local models do not lose user-supplied facts inside large KNX context. The structured memory tool now explicitly covers durable user facts such as preferred name and language as well as preferences and instructions, while remaining model-selected with no phrase classifiers or intent routing; credentials, security codes and API keys are excluded.<br/>
+- **Cerebrum Ultimate — editable and portable CHAT learning**: added **Settings → AI Chat Learning** to the Vue web UI for viewing the exact absolute path and editing, copying, downloading or restoring the shared learning file. The **Conversations & home** tab in the Node-RED node configuration includes a localized shortcut that opens the Web UI directly on this editor for the current node. Saves are validated, bounded to 512 KB, written atomically and applied immediately to every Cerebrum Ultimate node sharing the same storage, while revision checks prevent stale editors from overwriting newer learning. UI strings, editor help and wiki documentation were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
+- **Cerebrum Ultimate — intent-free agentic chat tools**: removed hard-coded semantic phrase classifiers from conversational routing. Every Ask/CHAT request now lets the model reason directly over the current goal, persistent user guidance, AI Education and the available structured KNX, camera, TTS Ultimate and memory tools. Routine planning no longer depends on a multilingual keyword list, TTS announcements no longer require an action alias, and durable chat instructions are remembered or forgotten through model-selected structured actions instead of regular-expression intent detection. Runtime argument validation, camera-adapter availability checks, ETS/DPT safeguards and command confirmation remain unchanged.<br/>
+- **Cerebrum Ultimate — learned KNX group-address roles**: a `neutral` GA is now treated as unresolved rather than permanently uncontrollable. The conversational model can learn or forget the exact `command`, `status` or `neutral` role from trusted user teaching, persistent chat guidance, AI Education or unequivocal ETS semantics through the structured `gaRoleActions` tool. Learned role, reason and evidence are persisted per node in `knxai-config-<node-id>.json`, synchronized into bounded semantic home memory and reused after restart. A command role learned in the current answer can validate a write in that same turn, while exact ETS membership, immutable DPT, payload validation and configured confirmation still apply. Help and wiki documentation were aligned in EN, IT, DE, FR, ES and zh-CN.<br/>
 
 **Version 6.3.22** - August 2026<br/>
 
-- **KNX AI — visible context size and real prompt weight**: the “Chat context overview” card now shows the current operational context limit and the actual UTF-8 size of the last operational chat prompt. Ollama, Bionic, OpenAI-compatible and Claude token-usage fields are captured when returned by the provider; otherwise the UI clearly labels the token count as an estimate. Local limits reflect the active context without exceeding KNX AI's 16K cap.<br/>
-- **KNX AI — documentation-free operational chat**: Telegram, RedBot and custom CHAT adapters no longer inject packaged help, README, changelog, wiki or example snippets into model prompts. They retain live and archived KNX data, ETS semantics, Node-RED inventory, conversation and home memory, AI Education, cameras and TTS; the web Assistant keeps documentation access for package-support questions.<br/>
-- **KNX AI — tolerant safe-read normalization**: state queries from small local models no longer become rejected writes when the model returns exact ETS destinations but omits the operation discriminator and payload. Payload-free operations are safely inferred as `GroupValue_Read`; payload-bearing operations remain writes and retain every ETS-role, DPT, payload and confirmation safeguard.<br/>
-- **KNX AI — fixed 16K Ollama runtime and prompt budget**: Ollama's advertised model maximum is now informational instead of being sent back as the requested `num_ctx`. KNX AI caps `num_ctx` and its relevance-selected semantic prompt at 16K, preventing 131K KV-cache allocation and full-context prefill. An audit confirmed that OpenAI and Claude receive only output limits, not a client-selected maximum context window.<br/>
-- **KNX AI — fixed 16K Bionic prompt budget**: Bionic's reported active or maximum context can no longer promote conversations to the full 131K data view. KNX AI always builds a relevance-selected 16K semantic prompt for Bionic while preserving the provider's loaded instance, open-ended reasoning, routines, KNX control, cameras and TTS capabilities.<br/>
-- **KNX AI — preserve Bionic LM Studio context**: model discovery now keeps the context length of an already loaded Bionic instance instead of confusing the model's maximum capability with the desired runtime setting. KNX AI no longer loads inactive models through the management API: the first chat request leaves JIT loading and saved per-model defaults to Bionic, while KNX AI uses a conservative 16K prompt budget until the active configuration can be inspected.<br/>
-- **KNX AI — neutral structured-response example**: the conversational contract now demonstrates empty action arrays instead of a fabricated group address, DPT, camera and payload. This prevents small local models from copying placeholder operations into unrelated replies such as greetings while preserving open-ended reasoning and every existing KNX, camera, TTS and routine capability.<br/>
+- **Cerebrum Ultimate — visible context size and real prompt weight**: the “Chat context overview” card now shows the current operational context limit and the actual UTF-8 size of the last operational chat prompt. Ollama, Bionic, OpenAI-compatible and Claude token-usage fields are captured when returned by the provider; otherwise the UI clearly labels the token count as an estimate. Local limits reflect the active context without exceeding Cerebrum Ultimate's 16K cap.<br/>
+- **Cerebrum Ultimate — documentation-free operational chat**: Telegram, RedBot and custom CHAT adapters no longer inject packaged help, README, changelog, wiki or example snippets into model prompts. They retain live and archived KNX data, ETS semantics, Node-RED inventory, conversation and home memory, AI Education, cameras and TTS; the web Assistant keeps documentation access for package-support questions.<br/>
+- **Cerebrum Ultimate — tolerant safe-read normalization**: state queries from small local models no longer become rejected writes when the model returns exact ETS destinations but omits the operation discriminator and payload. Payload-free operations are safely inferred as `GroupValue_Read`; payload-bearing operations remain writes and retain every ETS-role, DPT, payload and confirmation safeguard.<br/>
+- **Cerebrum Ultimate — fixed 16K Ollama runtime and prompt budget**: Ollama's advertised model maximum is now informational instead of being sent back as the requested `num_ctx`. Cerebrum Ultimate caps `num_ctx` and its relevance-selected semantic prompt at 16K, preventing 131K KV-cache allocation and full-context prefill. An audit confirmed that OpenAI and Claude receive only output limits, not a client-selected maximum context window.<br/>
+- **Cerebrum Ultimate — fixed 16K Bionic prompt budget**: Bionic's reported active or maximum context can no longer promote conversations to the full 131K data view. Cerebrum Ultimate always builds a relevance-selected 16K semantic prompt for Bionic while preserving the provider's loaded instance, open-ended reasoning, routines, KNX control, cameras and TTS capabilities.<br/>
+- **Cerebrum Ultimate — preserve Bionic LM Studio context**: model discovery now keeps the context length of an already loaded Bionic instance instead of confusing the model's maximum capability with the desired runtime setting. Cerebrum Ultimate no longer loads inactive models through the management API: the first chat request leaves JIT loading and saved per-model defaults to Bionic, while Cerebrum Ultimate uses a conservative 16K prompt budget until the active configuration can be inspected.<br/>
+- **Cerebrum Ultimate — neutral structured-response example**: the conversational contract now demonstrates empty action arrays instead of a fabricated group address, DPT, camera and payload. This prevents small local models from copying placeholder operations into unrelated replies such as greetings while preserving open-ended reasoning and every existing KNX, camera, TTS and routine capability.<br/>
 
 **Version 6.3.21** - August 2026<br/>
 
-- **KNX AI — persistent adapter-event history**: every event published by an automatically detected adapter is now normalized into vendor-neutral metadata and appended to a node-specific daily JSONL archive. The archive follows the 10-day KNX retention, guaranteeing at least 24 hours of camera and future adapter events across Node-RED restarts without storing snapshot image data. The editor context card exposes the actual adapter-history directories.<br/>
-- **KNX AI — authoritative historical queries**: web Assistant, Telegram, RedBot and custom CHAT channels now query both the adapter-event archive and the existing KNX daily telegram files. Historical prompts include totals calculated across every stored row in the requested interval plus relevance-selected detail samples, preventing sample size from being reported as the total. Natural-language ranges now include multilingual “last N hours” requests and are bounded by available retention.<br/>
+- **Cerebrum Ultimate — persistent adapter-event history**: every event published by an automatically detected adapter is now normalized into vendor-neutral metadata and appended to a node-specific daily JSONL archive. The archive follows the 10-day KNX retention, guaranteeing at least 24 hours of camera and future adapter events across Node-RED restarts without storing snapshot image data. The editor context card exposes the actual adapter-history directories.<br/>
+- **Cerebrum Ultimate — authoritative historical queries**: web Assistant, Telegram, RedBot and custom CHAT channels now query both the adapter-event archive and the existing KNX daily telegram files. Historical prompts include totals calculated across every stored row in the requested interval plus relevance-selected detail samples, preventing sample size from being reported as the total. Natural-language ranges now include multilingual “last N hours” requests and are bounded by available retention.<br/>
 
 **Version 6.3.19** - August 2026<br/>
 
-- **KNX AI — conversational multi-step routines**: added coordinated routines such as leaving home, bedtime and cinema mode. KNX AI can now perform a first pass with up to 20 fresh KNX state reads, use the authoritative bus results to prepare an ordered plan of up to 12 validated writes, request one confirmation and then execute the complete routine. After confirmation it waits up to four seconds for immediate matching bus feedback, reports verified and unverified operations without treating missing immediate feedback as a device failure, and dispatches any explicitly requested TTS Ultimate announcement only after execution. Routine details, preliminary readings and execution results are exposed as structured chat metadata. The importable confirmation example, editor help and wiki documentation were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
-- **KNX AI — localized output labels**: corrected the four output-pin labels to use the node-scoped Node-RED translation catalog, so Summary, Anomalies, Assistant and KNX Commands follow the active editor language.<br/>
+- **Cerebrum Ultimate — conversational multi-step routines**: added coordinated routines such as leaving home, bedtime and cinema mode. Cerebrum Ultimate can now perform a first pass with up to 20 fresh KNX state reads, use the authoritative bus results to prepare an ordered plan of up to 12 validated writes, request one confirmation and then execute the complete routine. After confirmation it waits up to four seconds for immediate matching bus feedback, reports verified and unverified operations without treating missing immediate feedback as a device failure, and dispatches any explicitly requested TTS Ultimate announcement only after execution. Routine details, preliminary readings and execution results are exposed as structured chat metadata. The importable confirmation example, editor help and wiki documentation were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
+- **Cerebrum Ultimate — localized output labels**: corrected the four output-pin labels to use the node-scoped Node-RED translation catalog, so Summary, Anomalies, Assistant and KNX Commands follow the active editor language.<br/>
 
 **Version 6.3.18** - August 2026<br/>
 
-- **KNX AI — focused configuration with fixed defaults**: aligned the hidden traffic-analysis defaults to a 600-second history, disk archival enabled for 10 days, automatic summaries disabled, a 120-second analysis window, 5,000 in-memory events and a Top-12 list. The Node-RED project inventory is now always included, while traffic analysis, chat input/output mappings and their obsolete editor controls remain hidden. Documentation language selection was removed because the AI now chooses the appropriate language. Proactive-home switches and advanced proactive settings were removed without legacy-value migration: user-managed AI Education is now the sole configuration surface for this behaviour. Editor help, examples and wiki documentation were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
-- **KNX AI — horizontal accordion and visible chat context**: rebuilt the editor accordion with the same horizontal-tab layout and visual behaviour as KNXUltimate. Automatically detected adapters now appear at the end, followed by a new context card that summarizes what chat can access, including generated context filenames and the absolute KNX telegram archive directories.<br/>
-- **KNX AI — TTS Ultimate announcements**: added automatic discovery of `node-red-contrib-tts-ultimate` and a selector for the TTS Ultimate node to use across all flows. Chat responses can now dispatch spoken announcements through the selected TTS/Sonos path while preserving the existing adapter registry design.<br/>
-- **KNX AI — local model providers and responsiveness**: promoted Ollama out of beta and added Bionic LM Studio. Model selection now discovers each local model's maximum context length and configures it when loaded; smaller context windows automatically receive reduced, relevance-selected prompts and bounded output budgets, with progressive context compaction on provider overflow errors. Local requests use a ten-minute minimum timeout, preserve detailed provider errors and retry supported OpenAI-compatible parameter variants. Chat users receive localized thinking feedback while inference is running, and the node status now reports only incoming requests and thinking state instead of KNX telegram traffic.<br/>
+- **Cerebrum Ultimate — focused configuration with fixed defaults**: aligned the hidden traffic-analysis defaults to a 600-second history, disk archival enabled for 10 days, automatic summaries disabled, a 120-second analysis window, 5,000 in-memory events and a Top-12 list. The Node-RED project inventory is now always included, while traffic analysis, chat input/output mappings and their obsolete editor controls remain hidden. Documentation language selection was removed because the AI now chooses the appropriate language. Proactive-home switches and advanced proactive settings were removed without legacy-value migration: user-managed AI Education is now the sole configuration surface for this behaviour. Editor help, examples and wiki documentation were updated in EN, IT, DE, FR, ES and zh-CN.<br/>
+- **Cerebrum Ultimate — horizontal accordion and visible chat context**: rebuilt the editor accordion with the same horizontal-tab layout and visual behaviour as KNXUltimate. Automatically detected adapters now appear at the end, followed by a new context card that summarizes what chat can access, including generated context filenames and the absolute KNX telegram archive directories.<br/>
+- **Cerebrum Ultimate — TTS Ultimate announcements**: added automatic discovery of `node-red-contrib-tts-ultimate` and a selector for the TTS Ultimate node to use across all flows. Chat responses can now dispatch spoken announcements through the selected TTS/Sonos path while preserving the existing adapter registry design.<br/>
+- **Cerebrum Ultimate — local model providers and responsiveness**: promoted Ollama out of beta and added Bionic LM Studio. Model selection now discovers each local model's maximum context length and configures it when loaded; smaller context windows automatically receive reduced, relevance-selected prompts and bounded output budgets, with progressive context compaction on provider overflow errors. Local requests use a ten-minute minimum timeout, preserve detailed provider errors and retry supported OpenAI-compatible parameter variants. Chat users receive localized thinking feedback while inference is running, and the node status now reports only incoming requests and thinking state instead of KNX telegram traffic.<br/>
 
 **Version 6.3.17** - August 2026<br/>
 
-- **KNX AI — unified Assistant context for chat channels**: Telegram, RedBot and custom chat adapters now receive the same complete KNX analysis context as the web Assistant, including the full relevant traffic selection, summaries, areas, learned home memory, Node-RED flow context, Function-node source and documentation snippets enabled by the node configuration. The chat-specific ETS catalog, camera capabilities, persistent conversation context, command validation and confirmation safeguards remain layered on top.<br/>
+- **Cerebrum Ultimate — unified Assistant context for chat channels**: Telegram, RedBot and custom chat adapters now receive the same complete KNX analysis context as the web Assistant, including the full relevant traffic selection, summaries, areas, learned home memory, Node-RED flow context, Function-node source and documentation snippets enabled by the node configuration. The chat-specific ETS catalog, camera capabilities, persistent conversation context, command validation and confirmation safeguards remain layered on top.<br/>
 
 **Version 6.3.16** - August 2026<br/>
 
-- **KNX AI — automatically detected camera adapters**: installed camera suites can now register vendor-neutral camera catalogs, connectivity state, snapshots and smart events directly in the KNX AI chat runtime, without selectors or camera-node wiring. The “Conversations & home” editor section shows every detected adapter together with its registered controllers and cameras. `node-red-contrib-unifi-ultimate` is the first provider: users can request or visually analyze a current Protect snapshot and create persistent Telegram/RedBot notifications for motion, smart line crossings and intrusion/loiter zones. Generic smart-detection rules now correctly distinguish classified people, animals, vehicles, faces, license plates and packages from unclassified motion, and match classifications across Protect smart zones, lines and loiter zones. Chat snapshots no longer force full-HD; the provider chooses a supported quality and retries only transient failures at standard quality. An offline camera is identified explicitly before Chat AI proposes an unavailable snapshot. Camera rules share `knxai-chat-context.md` and survive Node-RED restarts; Telegram adapters emit snapshots as native photos. Existing `node-red-contrib-telegrambot` presets saved before image support are upgraded at runtime for snapshot replies, including explicit JPEG filename and MIME metadata. The same registry contract is ready for packages such as `hikvision-ultimate`.<br/>
+- **Cerebrum Ultimate — automatically detected camera adapters**: installed camera suites can now register vendor-neutral camera catalogs, connectivity state, snapshots and smart events directly in the Cerebrum Ultimate chat runtime, without selectors or camera-node wiring. The “Conversations & home” editor section shows every detected adapter together with its registered controllers and cameras. `node-red-contrib-unifi-ultimate` is the first provider: users can request or visually analyze a current Protect snapshot and create persistent Telegram/RedBot notifications for motion, smart line crossings and intrusion/loiter zones. Generic smart-detection rules now correctly distinguish classified people, animals, vehicles, faces, license plates and packages from unclassified motion, and match classifications across Protect smart zones, lines and loiter zones. Chat snapshots no longer force full-HD; the provider chooses a supported quality and retries only transient failures at standard quality. An offline camera is identified explicitly before Chat AI proposes an unavailable snapshot. Camera rules share `knxai-chat-context.md` and survive Node-RED restarts; Telegram adapters emit snapshots as native photos. Existing `node-red-contrib-telegrambot` presets saved before image support are upgraded at runtime for snapshot replies, including explicit JPEG filename and MIME metadata. The same registry contract is ready for packages such as `hikvision-ultimate`.<br/>
 
 **Version 6.3.15** - August 2026<br/>
 
-- **KNX AI — persistent shared CHAT context**: every Ask/chat session now reloads its bounded recent conversation and explicit long-term user instructions after Node-RED restarts. Requests such as “remember not to use the term unknown” are retained per session in the atomically written shared file `knxultimatestorage/knxai/memory/knxai-chat-context.md`, independently from home intelligence memory and even when KNX actuator control is disabled. All KNX AI nodes using the same storage share one live context; the file keeps up to 8 recent turns and 20 durable instructions per session, 50 sessions total, and a 512 KB hard limit. `clear_chat` removes only the selected shared session and `reset` clears the shared context.<br/>
-- **KNX AI — unified memory files**: all KNX AI instances using the same storage now share `knxai-home-memory.md` and `knxai-chat-context.md`, without `node.id` in either filename. Per-node memory files from earlier builds are deliberately not migrated or read.<br/>
-- **KNX AI — fixed 5 MB home memory**: the shared home-memory file now has a fixed 5 MB limit. The `homeMemoryMaxKb` editor option has been removed to simplify the UI, and saved values no longer affect runtime behavior.<br/>
-- **KNX AI — RedBot Telegram chat adapter**: added a selectable `RedBot / node-red-contrib-chatbot (Telegram)` preset for direct `chatbot-telegram-receive → KNX AI → chatbot-telegram-send` wiring. It preserves RedBot conversation tracking, maps normalized inbound text and postbacks, and emits native `inline-buttons` confirmation actions.<br/>
+- **Cerebrum Ultimate — persistent shared CHAT context**: every Ask/chat session now reloads its bounded recent conversation and explicit long-term user instructions after Node-RED restarts. Requests such as “remember not to use the term unknown” are retained per session in the atomically written shared file `knxultimatestorage/knxai/memory/knxai-chat-context.md`, independently from home intelligence memory and even when KNX actuator control is disabled. All Cerebrum Ultimate nodes using the same storage share one live context; the file keeps up to 8 recent turns and 20 durable instructions per session, 50 sessions total, and a 512 KB hard limit. `clear_chat` removes only the selected shared session and `reset` clears the shared context.<br/>
+- **Cerebrum Ultimate — unified memory files**: all Cerebrum Ultimate instances using the same storage now share `knxai-home-memory.md` and `knxai-chat-context.md`, without `node.id` in either filename. Per-node memory files from earlier builds are deliberately not migrated or read.<br/>
+- **Cerebrum Ultimate — fixed 5 MB home memory**: the shared home-memory file now has a fixed 5 MB limit. The `homeMemoryMaxKb` editor option has been removed to simplify the UI, and saved values no longer affect runtime behavior.<br/>
+- **Cerebrum Ultimate — RedBot Telegram chat adapter**: added a selectable `RedBot / node-red-contrib-chatbot (Telegram)` preset for direct `chatbot-telegram-receive → Cerebrum Ultimate → chatbot-telegram-send` wiring. It preserves RedBot conversation tracking, maps normalized inbound text and postbacks, and emits native `inline-buttons` confirmation actions.<br/>
 
 **Version 6.3.14** - August 2026<br/>
 
@@ -119,7 +122,7 @@
 **Version 6.3.13** - August 2026<br/>
 
 - **Matter**: promoted Matter Controller, Matter Bridge and Expose KNX to Matter from BETA to production-ready status; removed BETA labels and experimental warnings from the palette, editor help, examples and multilingual documentation while retaining pairing, backup and endpoint-structure safety guidance.<br/>
-- **Matter visual identity**: changed the Matter node background to the official Matter Day color (`#F3FFFF`) and the controller/bridge icon artwork to Matter Night (`#131926`) for readable, standards-aligned palette and canvas presentation.<br/>
+- **Matter visual identity**: changed the Matter node background to the official Matter Day color (`#8FB1FA`) and the controller/bridge icon artwork to Matter Night (`#131926`) for readable, standards-aligned palette and canvas presentation.<br/>
 
 **Version 6.3.12** - August 2026<br/>
 
@@ -195,25 +198,25 @@
 **Version 6.2.2** - July 2026<br/>
 
 - Bump KNX Engine to 6.0.2<br/>
-- **KNX AI — streamlined configuration and JavaScript chat editors**: removed the raw-hex, custom system-prompt, documentation-toggle, telegram-type and anomaly/pattern controls from the node editor. KNX AI now always listens to GroupValue Read/Response/Write, initializes pattern/anomaly analysis with its established defaults, excludes raw hex from prompts and always includes relevant documentation context. The chat input/output mappings now use the same color-coded JavaScript editors as KNXUltimate (green input and yellow output), including safe save/cancel lifecycle handling; **No adapter** is the default and keeps both editors hidden. All editor `form-tips` were removed, and model refresh now uses only a rotating button icon while loading instead of reporting the number of models found. Missing numeric settings now receive their established defaults instead of invalidating older nodes; this includes 120 minutes for the open threshold, 360 minutes for repeat cooldown and 256 KB for home memory. The KNX gateway is the only editor field that remains mandatory.<br/>
+- **Cerebrum Ultimate — streamlined configuration and JavaScript chat editors**: removed the raw-hex, custom system-prompt, documentation-toggle, telegram-type and anomaly/pattern controls from the node editor. Cerebrum Ultimate now always listens to GroupValue Read/Response/Write, initializes pattern/anomaly analysis with its established defaults, excludes raw hex from prompts and always includes relevant documentation context. The chat input/output mappings now use the same color-coded JavaScript editors as KNXUltimate (green input and yellow output), including safe save/cancel lifecycle handling; **No adapter** is the default and keeps both editors hidden. All editor `form-tips` were removed, and model refresh now uses only a rotating button icon while loading instead of reporting the number of models found. Missing numeric settings now receive their established defaults instead of invalidating older nodes; this includes 120 minutes for the open threshold, 360 minutes for repeat cooldown and 256 KB for home memory. The KNX gateway is the only editor field that remains mandatory.<br/>
 
 **Version 6.2.1** - July 2026<br/>
 
-- **KNX AI — bounded proactive home intelligence**: added an opt-in **Proactive home & memory** section that derives a multilingual semantic model from ETS hierarchy, names, roles and DPTs and can proactively report reliably recognized covers, windows or doors left open beyond a configurable threshold. Notifications use output 3 with `msg.knxAi.type = "proactive_notification"`, retain the Telegram recipient through `msg.inputMessage`, obey quiet hours, a per-object cooldown and a maximum of three messages per hour, and never emit autonomous KNX commands on output 4. The LLM treats the user-only **AI Education** text as authoritative when deciding whether a candidate notification is useful; if Education cannot be evaluated, the notification is suppressed. A Markdown home reference is loaded at Node-RED startup, updated atomically and periodically, cleans stale temporary files, and is limited to 64–1,024 KB (256 KB by default), 120 significant observations, 80 aggregate habits, 80 notification records and 300 semantic ETS objects; older low-priority entries are pruned before the hard file limit can be exceeded. The direct Telegram example demonstrates the bounded memory and proactive settings.<br/>
-- **KNX AI — clearer accordion editor and practical guidance**: replaced the crowded vertical tabs with three main accordion sections: **AI assistant**, **Conversations & home**, and **KNX traffic analysis**. Opening one section shows all related options together, with no nested tabs or hidden advanced details and no changes to persisted field IDs or values. Help and wiki documentation in every supported language now include realistic values for recipient, threshold, quiet hours, cooldown and memory, a copyable **AI Education** example, expected proactive/suppressed outcomes, and the KNX safety boundaries.<br/>
+- **Cerebrum Ultimate — bounded proactive home intelligence**: added an opt-in **Proactive home & memory** section that derives a multilingual semantic model from ETS hierarchy, names, roles and DPTs and can proactively report reliably recognized covers, windows or doors left open beyond a configurable threshold. Notifications use output 3 with `msg.knxAi.type = "proactive_notification"`, retain the Telegram recipient through `msg.inputMessage`, obey quiet hours, a per-object cooldown and a maximum of three messages per hour, and never emit autonomous KNX commands on output 4. The LLM treats the user-only **AI Education** text as authoritative when deciding whether a candidate notification is useful; if Education cannot be evaluated, the notification is suppressed. A Markdown home reference is loaded at Node-RED startup, updated atomically and periodically, cleans stale temporary files, and is limited to 64–1,024 KB (256 KB by default), 120 significant observations, 80 aggregate habits, 80 notification records and 300 semantic ETS objects; older low-priority entries are pruned before the hard file limit can be exceeded. The direct Telegram example demonstrates the bounded memory and proactive settings.<br/>
+- **Cerebrum Ultimate — clearer accordion editor and practical guidance**: replaced the crowded vertical tabs with three main accordion sections: **AI assistant**, **Conversations & home**, and **KNX traffic analysis**. Opening one section shows all related options together, with no nested tabs or hidden advanced details and no changes to persisted field IDs or values. Help and wiki documentation in every supported language now include realistic values for recipient, threshold, quiet hours, cooldown and memory, a copyable **AI Education** example, expected proactive/suppressed outcomes, and the KNX safety boundaries.<br/>
 
 **Version 6.2.0** - July 2026<br/>
 
-- **KNX AI — DPT 1 command payload normalization**: fixed actuator writes being rejected when an otherwise valid AI response represented a DPT 1.xxx value as numeric `1`/`0` (including quoted values) instead of a JSON boolean. Exact safe equivalents `true`/`false`, `1`/`0`, `on`/`off` and ETS value labels are now normalized to real booleans before local validation, confirmation and Universal Mode output; other numeric values remain rejected.<br/>
+- **Cerebrum Ultimate — DPT 1 command payload normalization**: fixed actuator writes being rejected when an otherwise valid AI response represented a DPT 1.xxx value as numeric `1`/`0` (including quoted values) instead of a JSON boolean. Exact safe equivalents `true`/`false`, `1`/`0`, `on`/`off` and ETS value labels are now normalized to real booleans before local validation, confirmation and Universal Mode output; other numeric values remain rejected.<br/>
 - **Sidebar package cleanup**: removed the unused **KNX Debug** and **KNX Monitor** sidebar tabs, their dedicated admin endpoints and the Debug logger's global 5,000-entry memory buffer. The independent Flow Bubbles plugin and standard Node-RED logging remain unchanged.<br/>
-- **KNX AI — conversational actuator control**: `ask` inputs can now produce a text reply on the existing AI Assistant output and separately emit locally validated `GroupValue_Write` messages on a new KNX commands output for a KNX Ultimate node in Universal Mode. Actuator control is opt-in and off by default; commands are limited to exact imported ETS group addresses classified as commands, must use the ETS DPT, pass payload/range validation and are capped per request. A second option, enabled by default, previews the exact GA/DPT/payload plan and waits up to five minutes for confirmation from the same chat session before emitting anything; cancellation, stale-plan replacement and validation immediately before output prevent an old or changed plan from reaching KNX. Recent chat turns are kept in bounded per-session RAM history, including detected Telegram chat IDs, and `clear_chat` clears both conversation and pending commands for one session without affecting traffic history. A ready-to-import conversational control example is available under `examples/`, including simulated chat input, confirmation/cancellation and the Universal Mode connection.<br/>
-- **KNX AI — fresh KNX state reads**: natural-language requests for a current value can now emit locally validated `GroupValue_Read` operations for exact objects in the imported ETS catalog, including status and read-only objects. Output 4 supplies the Universal Mode contract (`msg.destination`, exact ETS `msg.dpt`, `msg.event = "GroupValue_Read"` and `msg.readstatus = true`), waits up to six seconds for a fresh response/write telegram, and returns localized decoded values plus `msg.knxAi.readResults` on output 3. Reads are immediate and never require confirmation; the existing command-role and confirmation protections remain mandatory for writes. The conversational example now includes a live room-temperature read.<br/>
-- **KNX AI — request-language confirmation instructions**: whenever validated commands are waiting for confirmation, the assistant now always appends deterministic confirmation/cancellation instructions in the current request language. Explicit message language remains authoritative, followed by the LLM language field, local request-text detection and Telegram/configuration fallback.<br/>
-- **KNX AI — button-ready confirmation metadata**: pending confirmation replies now expose `msg.knxAi.confirmationRequest` with session, expiry, command count and localized `confirm`/`cancel` actions. Each action includes a stable `callbackData` value and a ready-to-send Node-RED message, allowing Telegram and other chat nodes to present buttons instead of requiring typed confirmation.<br/>
-- **KNX AI — input message passthrough**: every AI Assistant and KNX command output now includes an independent copy of the triggering input message in `msg.inputMessage`, preserving Telegram/chat metadata and custom flow properties for downstream routing. Message-cloning and output-dispatch failures are contained and reported so malformed custom properties or downstream runtime errors cannot produce an uncaught exception or rejection.<br/>
-- **KNX AI — streamlined vertical editor tabs**: the editor now opens on a compact **Quick setup** tab containing only the common AI choices (enable, provider, credential, model, KNX commands and confirmation). Capture, storage, detection, AI context and advanced connection controls use the same left-hand vertical tab layout and shared CSS as the Matter editors, with technical fields grouped by topic and no changes to persisted field IDs or configuration values.<br/>
-- **KNX AI — selectable chat adapters and direct Telegrambot wiring**: a new **Chat adapters** tab loads preset input/output mappings from `resources/KNXAIChatAdapterMappings.js` and inserts their editable synchronous JavaScript into two persisted, full-width text boxes. The included `windkh/node-red-contrib-telegrambot` preset accepts receiver text and callback-query events directly, preserves chat/session/language data, converts output 3 directly to the Telegram sender contract and creates localized inline confirmation buttons from `msg.knxAi.confirmationRequest`. Adapter syntax/runtime failures are contained and reported, while unrelated outputs remain protected. A ready-to-import direct receiver/event → KNX AI → sender example is included under `examples/`; the Telegram package remains an optional external dependency.<br/>
-- **KNX AI — chat model compatibility**: the refreshed OpenAI-compatible model list now excludes known legacy completion-only models such as `gpt-3.5-turbo-instruct`, `davinci-002` and `babbage-002`. If a manually entered or third-party model rejects `/v1/chat/completions`, the assistant now reports the selected model and endpoint and suggests a compatible chat model instead of forwarding the provider's opaque error unchanged. Models that accept only the default temperature are retried without the custom `temperature` field, and token-limit fallbacks can be combined in the same request without discarding the prompt or conversation.<br/>
+- **Cerebrum Ultimate — conversational actuator control**: `ask` inputs can now produce a text reply on the existing AI Assistant output and separately emit locally validated `GroupValue_Write` messages on a new KNX commands output for a KNX Ultimate node in Universal Mode. Actuator control is opt-in and off by default; commands are limited to exact imported ETS group addresses classified as commands, must use the ETS DPT, pass payload/range validation and are capped per request. A second option, enabled by default, previews the exact GA/DPT/payload plan and waits up to five minutes for confirmation from the same chat session before emitting anything; cancellation, stale-plan replacement and validation immediately before output prevent an old or changed plan from reaching KNX. Recent chat turns are kept in bounded per-session RAM history, including detected Telegram chat IDs, and `clear_chat` clears both conversation and pending commands for one session without affecting traffic history. A ready-to-import conversational control example is available under `examples/`, including simulated chat input, confirmation/cancellation and the Universal Mode connection.<br/>
+- **Cerebrum Ultimate — fresh KNX state reads**: natural-language requests for a current value can now emit locally validated `GroupValue_Read` operations for exact objects in the imported ETS catalog, including status and read-only objects. Output 4 supplies the Universal Mode contract (`msg.destination`, exact ETS `msg.dpt`, `msg.event = "GroupValue_Read"` and `msg.readstatus = true`), waits up to six seconds for a fresh response/write telegram, and returns localized decoded values plus `msg.knxAi.readResults` on output 3. Reads are immediate and never require confirmation; the existing command-role and confirmation protections remain mandatory for writes. The conversational example now includes a live room-temperature read.<br/>
+- **Cerebrum Ultimate — request-language confirmation instructions**: whenever validated commands are waiting for confirmation, the assistant now always appends deterministic confirmation/cancellation instructions in the current request language. Explicit message language remains authoritative, followed by the LLM language field, local request-text detection and Telegram/configuration fallback.<br/>
+- **Cerebrum Ultimate — button-ready confirmation metadata**: pending confirmation replies now expose `msg.knxAi.confirmationRequest` with session, expiry, command count and localized `confirm`/`cancel` actions. Each action includes a stable `callbackData` value and a ready-to-send Node-RED message, allowing Telegram and other chat nodes to present buttons instead of requiring typed confirmation.<br/>
+- **Cerebrum Ultimate — input message passthrough**: every AI Assistant and KNX command output now includes an independent copy of the triggering input message in `msg.inputMessage`, preserving Telegram/chat metadata and custom flow properties for downstream routing. Message-cloning and output-dispatch failures are contained and reported so malformed custom properties or downstream runtime errors cannot produce an uncaught exception or rejection.<br/>
+- **Cerebrum Ultimate — streamlined vertical editor tabs**: the editor now opens on a compact **Quick setup** tab containing only the common AI choices (enable, provider, credential, model, KNX commands and confirmation). Capture, storage, detection, AI context and advanced connection controls use the same left-hand vertical tab layout and shared CSS as the Matter editors, with technical fields grouped by topic and no changes to persisted field IDs or configuration values.<br/>
+- **Cerebrum Ultimate — selectable chat adapters and direct Telegrambot wiring**: a new **Chat adapters** tab loads preset input/output mappings from `resources/KNXAIChatAdapterMappings.js` and inserts their editable synchronous JavaScript into two persisted, full-width text boxes. The included `windkh/node-red-contrib-telegrambot` preset accepts receiver text and callback-query events directly, preserves chat/session/language data, converts output 3 directly to the Telegram sender contract and creates localized inline confirmation buttons from `msg.knxAi.confirmationRequest`. Adapter syntax/runtime failures are contained and reported, while unrelated outputs remain protected. A ready-to-import direct receiver/event → Cerebrum Ultimate → sender example is included under `examples/`; the Telegram package remains an optional external dependency.<br/>
+- **Cerebrum Ultimate — chat model compatibility**: the refreshed OpenAI-compatible model list now excludes known legacy completion-only models such as `gpt-3.5-turbo-instruct`, `davinci-002` and `babbage-002`. If a manually entered or third-party model rejects `/v1/chat/completions`, the assistant now reports the selected model and endpoint and suggests a compatible chat model instead of forwarding the provider's opaque error unchanged. Models that accept only the default temperature are retried without the custom `temperature` field, and token-limit fallbacks can be combined in the same request without discarding the prompt or conversation.<br/>
 - **Control Matter from KNX — contextual flow PIN help**: **Node Input/Output PINs** now reveals the capability-driven **Flow input** section directly below the selector instead of adding a separate editor tab. The section remains hidden while the pins are disabled. Light endpoints now receive contextual, copyable examples for their supported top-level `msg.on`, `msg.dimming`, `msg.color_temperature` and `msg.color` contracts, while non-light endpoints retain their advertised simple and advanced Matter examples.<br/>
 - **Matter Bridge — contextual flow PIN help**: **Node Input/Output PINs** now sits outside the editor tabs and, when enabled, reveals a contextual **Flow input/output** section directly below it. Copyable Flow → Matter and Matter → Flow examples are filtered to the selected device type, covering lights, covers, thermostats, fans, sensors, alarms and the flow-only robot vacuum while documenting only the message contracts already accepted or emitted by the runtime.<br/>
 
@@ -367,9 +370,9 @@
 
 **Version 4.3.20** - June 2026<br/>
 
-- KNX AI: new **Flow Builder** (BETA) tab in the AI web page — describe an automation in plain language and the AI generates a ready-to-import **Node-RED flow** (JSON) using KNX Ultimate nodes, the Philips Hue nodes and native Function/logic nodes, wired to your imported group addresses. Copy the JSON and paste it via Node-RED _Menu > Import_. Generated node ids, wiring and config-node references are validated and rewired server-side before output.<br/>
-- KNX AI: added **Anthropic (Claude)** as an LLM provider alongside OpenAI-compatible and Ollama. Uses Claude's native Messages API; model and API key are configurable from the node (default `claude-opus-4-8`), with model auto-discovery.<br/>
-- KNX AI: Flow Builder UI is fully localized (EN/IT/DE/FR/ES) and laid out side-by-side (prompt and generated JSON); minor sidebar fix so long menu labels are no longer truncated.<br/>
+- Cerebrum Ultimate: new **Flow Builder** (BETA) tab in the AI web page — describe an automation in plain language and the AI generates a ready-to-import **Node-RED flow** (JSON) using KNX Ultimate nodes, the Philips Hue nodes and native Function/logic nodes, wired to your imported group addresses. Copy the JSON and paste it via Node-RED _Menu > Import_. Generated node ids, wiring and config-node references are validated and rewired server-side before output.<br/>
+- Cerebrum Ultimate: added **Anthropic (Claude)** as an LLM provider alongside OpenAI-compatible and Ollama. Uses Claude's native Messages API; model and API key are configurable from the node (default `claude-opus-4-8`), with model auto-discovery.<br/>
+- Cerebrum Ultimate: Flow Builder UI is fully localized (EN/IT/DE/FR/ES) and laid out side-by-side (prompt and generated JSON); minor sidebar fix so long menu labels are no longer truncated.<br/>
 
 **Version 4.3.19** - June 2026<br/>
 
@@ -425,14 +428,14 @@
 
 **Version 4.3.8** - April 2026<br/>
 
-- FIX: **KNX AI Web Assistant** `Ask` tab now truly uses the full available content width on desktop too; removed the leftover `50vw` cap that still kept the panel at half width.<br/>
+- FIX: **Cerebrum Ultimate Web Assistant** `Ask` tab now truly uses the full available content width on desktop too; removed the leftover `50vw` cap that still kept the panel at half width.<br/>
 - FIX: **KNX DEVICE** in universal mode (`listenAllGA`) now falls back to `raw` instead of raising a datapoint-detection error when no ETS DPT is available and automatic inference fails; raw bytes remain available in `msg.knx.rawValue`.<br/>
-- NEW: **KNX AI** can now optionally archive captured telegrams to disk under `knxultimatestorage`, with configurable retention in days and automatic reuse of the archive for `Ask` queries.<br/>
-- CHANGE: **KNX AI** `Ask` now uses the disk archive by default when enabled, honoring explicit time references and otherwise searching the last 24 hours plus current RAM events.<br/>
-- CHANGE: **KNX AI** flow context for the AI prompt now includes the whole **Node-RED project inventory**, not only KNX nodes or the current flow, so `function`/`change`/`inject`/`template` nodes with KNX-related logic are visible to the assistant.<br/>
-- FIX: **KNX AI Ask** now includes the real JavaScript source of Node-RED `function` nodes for code-review questions, instead of exposing only short flattened snippets that could make the assistant say it could not inspect the full code.<br/>
-- UI: **KNX AI** editor options were reorganized into clearer sections, with technical tuning moved under **Advanced**, several low-value LLM tuning fields removed from the editor, and localized tab labels/help/docs updated in all supported languages.<br/>
-- UI: **KNX AI Web Assistant** `Ask` panel now uses the full available page width instead of being capped to half-page width on desktop.<br/>
+- NEW: **Cerebrum Ultimate** can now optionally archive captured telegrams to disk under `knxultimatestorage`, with configurable retention in days and automatic reuse of the archive for `Ask` queries.<br/>
+- CHANGE: **Cerebrum Ultimate** `Ask` now uses the disk archive by default when enabled, honoring explicit time references and otherwise searching the last 24 hours plus current RAM events.<br/>
+- CHANGE: **Cerebrum Ultimate** flow context for the AI prompt now includes the whole **Node-RED project inventory**, not only KNX nodes or the current flow, so `function`/`change`/`inject`/`template` nodes with KNX-related logic are visible to the assistant.<br/>
+- FIX: **Cerebrum Ultimate Ask** now includes the real JavaScript source of Node-RED `function` nodes for code-review questions, instead of exposing only short flattened snippets that could make the assistant say it could not inspect the full code.<br/>
+- UI: **Cerebrum Ultimate** editor options were reorganized into clearer sections, with technical tuning moved under **Advanced**, several low-value LLM tuning fields removed from the editor, and localized tab labels/help/docs updated in all supported languages.<br/>
+- UI: **Cerebrum Ultimate Web Assistant** `Ask` panel now uses the full available page width instead of being capped to half-page width on desktop.<br/>
 
 **Version 4.3.6** - April 2026<br/>
 
@@ -458,15 +461,15 @@
 
 **Version 4.3.2** - April 2026<br/>
 
-- Docs/help/wiki: updated **KNX AI** help HTML and wiki pages in all supported languages (EN/IT/DE/FR/ES/zh-CN) to reflect the latest LLM/Ollama UX changes.<br/>
+- Docs/help/wiki: updated **Cerebrum Ultimate** help HTML and wiki pages in all supported languages (EN/IT/DE/FR/ES/zh-CN) to reflect the latest LLM/Ollama UX changes.<br/>
 - Docs/help/wiki: documented the new **Ollama quick setup** flow (**Download model** -> **Install it**).<br/>
-- Docs/help/wiki: clarified that the **LLM Assistant** tab is shown first in the KNX AI editor for faster setup.<br/>
+- Docs/help/wiki: clarified that the **LLM Assistant** tab is shown first in the Cerebrum Ultimate editor for faster setup.<br/>
 
 **Version 4.3.1** - April 2026<br/>
 
 - Bumped KNXEngine to 5.5.2<br/>
-- UI: **KNX AI Web** now always opens on the first menu item (**Overview**).<br/>
-- FIX: **KNX AI Web Assistant** preset prompts now send the exact current button text to the AI request, so localized UI labels are used directly (no forced English fallback).<br/>
+- UI: **Cerebrum Ultimate Web** now always opens on the first menu item (**Overview**).<br/>
+- FIX: **Cerebrum Ultimate Web Assistant** preset prompts now send the exact current button text to the AI request, so localized UI labels are used directly (no forced English fallback).<br/>
 
 **Version 4.3.0** - April 2026<br/>
 
@@ -480,41 +483,41 @@
 
 **Version 4.2.13** - April 2026<br/>
 
-- FIX: **KNX AI Web** hardened authenticated requests by always attaching Bearer auth on API/audio calls (query token or Node-RED stored editor token fallback), preventing silent audio failures in authenticated sessions.<br/>
-- FIX: **KNX AI Tests** now validates audio responses (`audio/*`) and reports auth/session issues explicitly instead of failing silently when HTML/login pages are returned.<br/>
-- NEW: **KNX AI Tests** now announces test completion by voice at the end of the test run (single and repeat-stop end states).<br/>
-- UI: **KNX AI Tests** added a blocking wait popup while building a Test Plan from the template builder, aligned with the existing AI Areas wait overlay behavior.<br/>
+- FIX: **Cerebrum Ultimate Web** hardened authenticated requests by always attaching Bearer auth on API/audio calls (query token or Node-RED stored editor token fallback), preventing silent audio failures in authenticated sessions.<br/>
+- FIX: **Cerebrum Ultimate Tests** now validates audio responses (`audio/*`) and reports auth/session issues explicitly instead of failing silently when HTML/login pages are returned.<br/>
+- NEW: **Cerebrum Ultimate Tests** now announces test completion by voice at the end of the test run (single and repeat-stop end states).<br/>
+- UI: **Cerebrum Ultimate Tests** added a blocking wait popup while building a Test Plan from the template builder, aligned with the existing AI Areas wait overlay behavior.<br/>
 - i18n: normalized **Duplicate / Duplicate Plan** action labels to imperative form across supported UI languages for consistency in the Tests page.<br/>
 
 **Version 4.2.12** - April 2026<br/>
 
-- FIX: **KNX AI Web** and **KNX Viewer Web** admin routes now normalize `access_token` auth safely by mapping query token to `Authorization: Bearer ...` only when missing, then removing `access_token` from `req.query` before permission checks.<br/>
+- FIX: **Cerebrum Ultimate Web** and **KNX Viewer Web** admin routes now normalize `access_token` auth safely by mapping query token to `Authorization: Bearer ...` only when missing, then removing `access_token` from `req.query` before permission checks.<br/>
 - FIX: resolved intermittent `400 Bad Request` on authenticated web pages caused by duplicate bearer token sources (header + query string).<br/>
 - CHANGE: extracted shared auth normalization helper to `nodes/utils/httpAdminAccessToken.js` to keep AI/Viewer behavior aligned and easier to maintain.<br/>
 
 **Version 4.2.11** - April 2026<br/>
 
-- UI: **KNX AI Web** improved assistant workflow with chat-style layout (prompt input under messages), clearer prompt focus, and streamlined Ask page text.<br/>
-- UI: **KNX AI Web** added clearer loading feedback for AI operations (area regeneration/deletion and planner generation), including a centered blocking wait overlay during long-running area generation.<br/>
-- NEW: **KNX AI Web** added bulk action **Delete AI Areas** and localized labels for **Regenerate AI Areas** / **Delete AI Areas** across supported UI languages.<br/>
-- FIX: **KNX AI backend** hardened JSON extraction/parsing for LLM outputs and improved timeout/token-limit diagnostics in error messages.<br/>
-- CHANGE: **KNX AI backend** raised default/forced `llmMaxTokens` handling for structured responses and aligned editor defaults for high-token completions.<br/>
-- Docs/wiki: improved **KNX AI Dashboard** pages for end users, added localized guidance updates, and made support CTA links visible in docs navigation.<br/>
+- UI: **Cerebrum Ultimate Web** improved assistant workflow with chat-style layout (prompt input under messages), clearer prompt focus, and streamlined Ask page text.<br/>
+- UI: **Cerebrum Ultimate Web** added clearer loading feedback for AI operations (area regeneration/deletion and planner generation), including a centered blocking wait overlay during long-running area generation.<br/>
+- NEW: **Cerebrum Ultimate Web** added bulk action **Delete AI Areas** and localized labels for **Regenerate AI Areas** / **Delete AI Areas** across supported UI languages.<br/>
+- FIX: **Cerebrum Ultimate backend** hardened JSON extraction/parsing for LLM outputs and improved timeout/token-limit diagnostics in error messages.<br/>
+- CHANGE: **Cerebrum Ultimate backend** raised default/forced `llmMaxTokens` handling for structured responses and aligned editor defaults for high-token completions.<br/>
+- Docs/wiki: improved **Cerebrum Ultimate Dashboard** pages for end users, added localized guidance updates, and made support CTA links visible in docs navigation.<br/>
 
 **Version 4.2.10** - April 2026<br/>
 
-- UI: **KNX AI Web** sidebar menu style aligned to Homebridge (font size/weight, spacing, icon/text alignment and sidebar widths) for closer visual consistency.<br/>
-- UI: **KNX AI Web** removed sidebar menu subtitles to keep navigation labels clean and compact.<br/>
-- UI: **KNX AI Web** **Assistant** tab: the **Ask** panel is now half-page width on desktop and full width on smaller screens.<br/>
+- UI: **Cerebrum Ultimate Web** sidebar menu style aligned to Homebridge (font size/weight, spacing, icon/text alignment and sidebar widths) for closer visual consistency.<br/>
+- UI: **Cerebrum Ultimate Web** removed sidebar menu subtitles to keep navigation labels clean and compact.<br/>
+- UI: **Cerebrum Ultimate Web** **Assistant** tab: the **Ask** panel is now half-page width on desktop and full width on smaller screens.<br/>
 
 **Version 4.2.7** - April 2026<br/>
 
-- UI: **KNX AI Web** layout and visual style aligned to the Homebridge-like interface, with unified orange accent palette and cleaned component styling.<br/>
-- CHANGE: **KNX AI Web** removed legacy theme management and related theme files, keeping a single coherent visual theme.<br/>
-- UI: **KNX AI Web** navigation/workspace refactor: dedicated pages for **Settings** and **Test Results**, plus improved menu structure and labels.<br/>
-- NEW: **KNX AI Web** **Settings** now includes split tabs (**KNX AI Node** and **Import / Export**) and moved node selection/auto-refresh controls into the dedicated node settings panel.<br/>
-- NEW: **KNX AI Web** **Test Results** can now be exported to **PDF** directly from the selected report.<br/>
-- FIX: **KNX AI Web** area editor unsaved-changes detection no longer prompts discard/stay immediately after saving a newly created area and pressing **Close**.<br/>
+- UI: **Cerebrum Ultimate Web** layout and visual style aligned to the Homebridge-like interface, with unified orange accent palette and cleaned component styling.<br/>
+- CHANGE: **Cerebrum Ultimate Web** removed legacy theme management and related theme files, keeping a single coherent visual theme.<br/>
+- UI: **Cerebrum Ultimate Web** navigation/workspace refactor: dedicated pages for **Settings** and **Test Results**, plus improved menu structure and labels.<br/>
+- NEW: **Cerebrum Ultimate Web** **Settings** now includes split tabs (**Cerebrum Ultimate Node** and **Import / Export**) and moved node selection/auto-refresh controls into the dedicated node settings panel.<br/>
+- NEW: **Cerebrum Ultimate Web** **Test Results** can now be exported to **PDF** directly from the selected report.<br/>
+- FIX: **Cerebrum Ultimate Web** area editor unsaved-changes detection no longer prompts discard/stay immediately after saving a newly created area and pressing **Close**.<br/>
 
 **Version 4.2.5** - March 2026<br/>
 
@@ -527,19 +530,19 @@
 **Version 4.2.4** - March 2026<br/>
 
 - NEW: **KNX Viewer Web** added a new Vue-based web page, opened directly from the **KNXViewer** node editor, to visualize KNX lights and dimmers in a modern dashboard.<br/>
-- UI: **KNX Viewer Web** uses a visual style aligned with **KNX AI**, including live light/dimmer cards, search, node selection and auto-refresh.<br/>
+- UI: **KNX Viewer Web** uses a visual style aligned with **Cerebrum Ultimate**, including live light/dimmer cards, search, node selection and auto-refresh.<br/>
 - IMPROVE: **KNX Viewer Web** ON lights now use a much stronger yellow highlight for clearer visual feedback.<br/>
 - CHANGE: package build/publish flow now also includes the **KNX Viewer Web** Vue bundle.<br/>
 
 **Version 4.2.3** - March 2026<br/>
 
-- IMPROVE: **KNX AI Flow Map** added a new toggle to show or hide `knxUltimate` nodes running in **Universal Mode**; they are now hidden by default to keep the topology cleaner.<br/>
+- IMPROVE: **Cerebrum Ultimate Flow Map** added a new toggle to show or hide `knxUltimate` nodes running in **Universal Mode**; they are now hidden by default to keep the topology cleaner.<br/>
 
 **Version 4.2.2** - March 2026<br/>
 
-- CHANGE: **KNX AI Web Page** is now the official Vue-based dashboard served from `/knxUltimateAI/sidebar/page`.<br/>
-- CHANGE: removed the old **KNX AI** legacy web page and its dedicated editor button, keeping the Vue preview as the only web UI entrypoint.<br/>
-- IMPROVE: **KNX AI Flow Map** in the Vue dashboard: wider layout, fullscreen toggle, animated/dashed traffic arrows, refined marker sizing and better panel width usage.<br/>
+- CHANGE: **Cerebrum Ultimate Web Page** is now the official Vue-based dashboard served from `/knxUltimateAI/sidebar/page`.<br/>
+- CHANGE: removed the old **Cerebrum Ultimate** legacy web page and its dedicated editor button, keeping the Vue preview as the only web UI entrypoint.<br/>
+- IMPROVE: **Cerebrum Ultimate Flow Map** in the Vue dashboard: wider layout, fullscreen toggle, animated/dashed traffic arrows, refined marker sizing and better panel width usage.<br/>
 
 **Version 4.2.1** - March 2026<br/>
 
@@ -550,8 +553,8 @@
 
 **Version 4.1.35** - March 2026<br/>
 
-- NEW: **KNX AI** anomalies output now always emits dedicated bus connection events when the KNX gateway connection is lost and when it is restored.<br/>
-- NEW: **KNX AI Web Dashboard** added a bus connection persistence section with a green/red time bar showing connected/disconnected periods over the history window.<br/>
+- NEW: **Cerebrum Ultimate** anomalies output now always emits dedicated bus connection events when the KNX gateway connection is lost and when it is restored.<br/>
+- NEW: **Cerebrum Ultimate Web Dashboard** added a bus connection persistence section with a green/red time bar showing connected/disconnected periods over the history window.<br/>
 
 **Version 4.1.34** - March 2026<br/>
 
@@ -577,12 +580,12 @@
 
 **Version 4.1.30** - March 2026<br/>
 
-- NEW: **KNX AI** now uses a full **Web Dashboard** as the main UI, opened from the node editor via **Open Web Page** button.<br/>
-- CHANGE: removed the legacy `knxUltimateAI-sidebar-plugin` file and aligned KNX AI admin routes to the Web Dashboard workflow.<br/>
-- IMPROVE: **KNX AI Flow Map** UX/readability: stable layout reset on page open, clearer legend, active/idle direction rendering and dashboard visual refinements.<br/>
-- UI: flattened the KNX AI Web Dashboard style (removed 3D/shadow-heavy look) for a cleaner production UI.<br/>
-- Docs/help/wiki: refreshed **KNX AI** documentation in all supported languages (EN/IT/DE/FR/ES/zh-CN), updated `KNX-AI-Sidebar` pages to Web Dashboard compatibility notes, and updated docs navigation section naming/ordering.<br/>
-- FIX: **KNX AI Ask chat**: SVG answers are now rendered correctly even with multiple SVG blocks, fenced `svg/xml/html` code blocks, or entity-escaped SVG markup.<br/>
+- NEW: **Cerebrum Ultimate** now uses a full **Web Dashboard** as the main UI, opened from the node editor via **Open Web Page** button.<br/>
+- CHANGE: removed the legacy `knxUltimateAI-sidebar-plugin` file and aligned Cerebrum Ultimate admin routes to the Web Dashboard workflow.<br/>
+- IMPROVE: **Cerebrum Ultimate Flow Map** UX/readability: stable layout reset on page open, clearer legend, active/idle direction rendering and dashboard visual refinements.<br/>
+- UI: flattened the Cerebrum Ultimate Web Dashboard style (removed 3D/shadow-heavy look) for a cleaner production UI.<br/>
+- Docs/help/wiki: refreshed **Cerebrum Ultimate** documentation in all supported languages (EN/IT/DE/FR/ES/zh-CN), updated `Cerebrum Ultimate-Sidebar` pages to Web Dashboard compatibility notes, and updated docs navigation section naming/ordering.<br/>
+- FIX: **Cerebrum Ultimate Ask chat**: SVG answers are now rendered correctly even with multiple SVG blocks, fenced `svg/xml/html` code blocks, or entity-escaped SVG markup.<br/>
 - FIX: **KNX Hue Light**: KNX read response for brightness now returns `0` when the light is OFF and brightness status is configured to follow HUE on/off.<br/>
 - FIX: **KNX Hue Light**: ignore dimming-only cached brightness updates from Hue while light is OFF, preventing wrong KNX brightness status jumps.<br/>
 - IMPROVE: **KNX Hue Light** grouped light handling: when an ON event arrives without dimming payload, brightness is fetched from bridge snapshot before publishing KNX status.<br/>
@@ -607,7 +610,7 @@
 
 **Version 4.1.26** - February 2026<br/>
 
-- i18n: **KNX AI** sidebar tab: localized the Summary and UI strings in all supported languages (including output pin labels).<br/>
+- i18n: **Cerebrum Ultimate** sidebar tab: localized the Summary and UI strings in all supported languages (including output pin labels).<br/>
 
 **Version 4.1.25** - February 2026<br/>
 
@@ -641,17 +644,17 @@
 
 - NEW: **KNX Multi Routing**: added **Server KNX/IP** mode (standalone KNXnet/IP tunneling server) that outputs/accepts RAW telegrams on the node ports (no gateway required).<br/>
 
-- NEW: KNX AI sidebar: added another TAB in the Node-Red's toolbar buttons dedicated to AI analysis.<br>
-- NEW: added **KNX AI** node (traffic analyzer + optional LLM assistant) and **KNX AI** sidebar tab.<br/>
+- NEW: Cerebrum Ultimate sidebar: added another TAB in the Node-Red's toolbar buttons dedicated to AI analysis.<br>
+- NEW: added **Cerebrum Ultimate** node (traffic analyzer + optional LLM assistant) and **Cerebrum Ultimate** sidebar tab.<br/>
 - NEW: added **KNX Multi Routing** node to bridge multiple `knxUltimate-config` gateways via Node-RED (RAW telegram forwarding + source/destination metadata).<br/>
 - NEW: added **KNX Router Filter** node to filter routed RAW telegrams (event type + GA/source patterns + loop prevention).<br/>
 - CHANGE: `knxUltimateAI` LLM: enabled extra context by default (Flow inventory + documentation snippets) to improve answer quality; both can be disabled in the node settings.<br/>
 - NEW: `knxUltimateAI` LLM: added documentation snippets retrieval from built-in help/README/examples (and `docs/wiki` when available in the local install) to help the assistant explain node behaviour and configuration.<br/>
-- IMPROVE: KNX AI sidebar chat: improved Markdown rendering (including tables), for better readability of AI answers.<br/>
-- Docs: completed the help + wiki pages for **KNX AI**, **KNX Multi Routing** and **KNX Router Filter** in all supported languages, and added them to the docs homepages.<br/>
+- IMPROVE: Cerebrum Ultimate sidebar chat: improved Markdown rendering (including tables), for better readability of AI answers.<br/>
+- Docs: completed the help + wiki pages for **Cerebrum Ultimate**, **KNX Multi Routing** and **KNX Router Filter** in all supported languages, and added them to the docs homepages.<br/>
 - Docs: added the new nodes to the wiki navbar in all languages (so they appear in the left navigation menu).<br/>
 - Docs: added new wiki **Samples** pages (with diagrams) for the 3 new nodes.<br/>
-- Examples: added importable Node-RED flow JSON examples for **KNX AI**, **KNX Multi Routing** and **KNX Router Filter** (including KNX Multi Routing **Server KNX/IP** sample).<br/>
+- Examples: added importable Node-RED flow JSON examples for **Cerebrum Ultimate**, **KNX Multi Routing** and **KNX Router Filter** (including KNX Multi Routing **Server KNX/IP** sample).<br/>
 
 **Version 4.1.15** - January 2026<br/>