Synaplan Integration - Releases

Knowledge-file lifecycle + per-file chat, built on Synaplan's file-lifecycle API (CORE-4).

Added

• Chat about this file — a new Files action opens a focused AI chat about a single file (text / JSON / XML), wiring up the previously unused per-file chat modal.
• Update / Remove in Knowledge — the "Add to AI Knowledge" dialog now detects when a file is already indexed and switches to Update in Knowledge (re-index in place, no duplicates) and offers Remove from Knowledge.
• "Changed since indexed" indicator — the dialog shows a warning when the Nextcloud file was modified after it was last indexed, with one-click update.

Changed

• Knowledge uploads now send the Nextcloud file id (source_id) and etag (source_etag) and use overwrite-by-source, so re-adding a file (or adding a changed one) updates its knowledge entry in place instead of creating a duplicate.

Requires

• For the update/stale/remove features: a Synaplan backend with the knowledge-file lifecycle API (source_id / source_etag / overwrite, check-stale, mark-stale). Against older backends the dialog degrades to plain "Add to Knowledge".

Added

• Backend environment switch (Live / Local). Admin settings now keep two separate Synaplan profiles — Live (production) and Local (development) — each with its own URL and API key, plus an Active environment toggle that flips the whole app (chat, media, knowledge, proxy) between them in one click. Makes it easy for any developer to point the app at a local Synaplan dev backend (e.g. http://localhost:8000) and switch back to live without re-entering credentials.

Major upgrade focused on language handling, personalisation, real streaming, and a clearer, better-formatted chat experience.

Added

• App version in the chat tagline — the Research/launcher header now shows the running version (e.g. "Ask anything v1.2.1 — powered by Synaplan"), read dynamically from the installed app version.
• Configurable answer language — a new "Language & AI behaviour" admin section lets you pick a default answer language and toggle "Use each user's Nextcloud interface language when supported". A German user then gets German answers automatically, with the default as the fallback. This replaces the previous behaviour where chat and document tools were effectively always English. The resolved language is injected as a system instruction into the streaming chat and passed through to summaries/translations, and the Summarize/Translate dialogs pre-select it.
• Personal memories in chat — when Synaplan's memory service (Qdrant) is reachable and the admin has enabled it ("Allow personal memories in chat"), the Research/launcher chat shows a "Use my memories" switch (on by default). Relevant memories are searched per question and injected as context so answers are personalised. The control is hidden entirely when the service is unavailable.
• Live chat status row — at-a-glance chips show the active model, the answer language, whether image (/pic) and video (/vid) generation are available, the selected knowledge base, and the memory state. Each answer is captioned with the model that actually produced it.
• Staged progress while waiting — instead of a static "Thinking…", the chat now reports what the backend is doing: Sorting your request → Searching the knowledge base → Recalling your memories → Generating the answer, stopping the moment streaming begins.
• Markdown rendering for document results — Summarize and Translate results now render Markdown (headings, bold, lists, code, links) via NcRichText, matching the chat, instead of showing raw **/- source.
• Client-config endpoint (api#clientConfig) exposing the resolved language and memory availability to the frontend.
• Staged upload progress — the "Add to Knowledge" dialog shows a progress bar and a phase checklist (Uploading → Extracting text → Creating chunks → Generating embeddings) with elapsed time.

Changed

• Real token-by-token streaming — chat answers now stream live. The SSE relay was rewritten to a direct cURL pass-through (SseProxyResponse) that writes each chunk straight to the browser and defeats PHP/Nextcloud output buffering (disables zlib/output buffering, Accept-Encoding: identity, X-Accel-Buffering: no, connection priming). Previously the answer arrived all at once after a long pause.
• Higher-contrast status chips — active capabilities use a solid brand fill with guaranteed-contrast text; disabled ones are clearly muted but still legible in both light and dark mode.
• Single-ring loading spinner — Summarize/Translate dialogs use one clean spinner instead of the twin-circle icon.
• Tighter Markdown spacing — list items in rendered answers no longer have oversized gaps.
• Per-chunk CSS injection (relativeCSSInjection) so shared component styles (NcSelect, NcCheckboxRadioSwitch) are injected by every entry that uses them — fixes unstyled dropdowns/switches on the Research page, the in-page launcher, and the file-action dialogs.

Fixed

• Answers are no longer "always English" — the chat/streaming path previously sent a hard-coded en output language; it now respects the admin default and/or the user's interface language.
• Admin toggles are clickable — the language/memory switches (and the chat "Use my memories" switch) used the wrong prop/event binding for @nextcloud/vue v9 and could not be toggled; they now use the correct v-model.
• Knowledge upload no longer shows a false success — when the server stores 0 chunks (empty/image-only file, or the embedding service is unavailable), the dialog reports a clear error instead of a green "File added" screen with "Chunks created: 0".
• Removed leftover "this is a demo" text from the knowledge-upload progress hint.

Security / dependencies

• Updated vite to 7.3.5, fixing a moderate dev-server advisory affecting 7.0.0–7.3.3.
• Refreshed dependency lockfiles: friendsofphp/php-cs-fixer 3.95.2, prettier 3.8.4, @nextcloud/vue 9.8.2, @nextcloud/axios 2.6.0, @nextcloud/dialogs 7.4.0, vue 3.5.38, and @nextcloud/initial-state 3.0.0.

Major upgrade focused on language handling, personalisation, real streaming, and a clearer, better-formatted chat experience.

Added

• Configurable answer language — a new "Language & AI behaviour" admin section lets you pick a default answer language and toggle "Use each user's Nextcloud interface language when supported". A German user then gets German answers automatically, with the default as the fallback. This replaces the previous behaviour where chat and document tools were effectively always English. The resolved language is injected as a system instruction into the streaming chat and passed through to summaries/translations, and the Summarize/Translate dialogs pre-select it.
• Personal memories in chat — when Synaplan's memory service (Qdrant) is reachable and the admin has enabled it ("Allow personal memories in chat"), the Research/launcher chat shows a "Use my memories" switch (on by default). Relevant memories are searched per question and injected as context so answers are personalised. The control is hidden entirely when the service is unavailable.
• Live chat status row — at-a-glance chips show the active model, the answer language, whether image (/pic) and video (/vid) generation are available, the selected knowledge base, and the memory state. Each answer is captioned with the model that actually produced it.
• Staged progress while waiting — instead of a static "Thinking…", the chat now reports what the backend is doing: Sorting your request → Searching the knowledge base → Recalling your memories → Generating the answer, stopping the moment streaming begins.
• Markdown rendering for document results — Summarize and Translate results now render Markdown (headings, bold, lists, code, links) via NcRichText, matching the chat, instead of showing raw **/- source.
• Client-config endpoint (api#clientConfig) exposing the resolved language and memory availability to the frontend.
• Staged upload progress — the "Add to Knowledge" dialog shows a progress bar and a phase checklist (Uploading → Extracting text → Creating chunks → Generating embeddings) with elapsed time.

Changed

• Real token-by-token streaming — chat answers now stream live. The SSE relay was rewritten to a direct cURL pass-through (SseProxyResponse) that writes each chunk straight to the browser and defeats PHP/Nextcloud output buffering (disables zlib/output buffering, Accept-Encoding: identity, X-Accel-Buffering: no, connection priming). Previously the answer arrived all at once after a long pause.
• Higher-contrast status chips — active capabilities use a solid brand fill with guaranteed-contrast text; disabled ones are clearly muted but still legible in both light and dark mode.
• Single-ring loading spinner — Summarize/Translate dialogs use one clean spinner instead of the twin-circle icon.
• Tighter Markdown spacing — list items in rendered answers no longer have oversized gaps.

Fixed

• Answers are no longer "always English" — the chat/streaming path previously sent a hard-coded en output language; it now respects the admin default and/or the user's interface language.
• Admin toggles are clickable — the language/memory switches (and the chat "Use my memories" switch) used the wrong prop/event binding for @nextcloud/vue v9 and could not be toggled; they now use the correct v-model.
• Knowledge upload no longer shows a false success — when the server stores 0 chunks (empty/image-only file, or the embedding service is unavailable), the dialog reports a clear error instead of a green "File added" screen with "Chunks created: 0".
• Removed leftover "this is a demo" text from the knowledge-upload progress hint.

Security / dependencies

• Updated vite to 7.3.5, fixing a moderate dev-server advisory affecting 7.0.0–7.3.3.
• Refreshed dependency lockfiles: friendsofphp/php-cs-fixer 3.95.2, prettier 3.8.4, @nextcloud/vue 9.8.2, @nextcloud/axios 2.6.0, @nextcloud/dialogs 7.4.0, vue 3.5.38, and @nextcloud/initial-state 3.0.0.

Added

• Summarize documents via file context menu (bullet points, paragraph, abstractive; short/medium/long; 12 output languages)
• Translate documents via file context menu (12 target languages)
• Add to AI Knowledge — upload files to the Synaplan vector knowledge base with group management
• Research Chat — full-page AI assistant accessible from top navigation with knowledge group and LLM model selection
• Support for binary document formats (PDF, DOCX, ODT, XLSX, PPTX) via Synaplan's Tika text extraction
• Admin settings page with Synaplan URL and API key configuration
• Connection test functionality in admin settings
• Top-level "Synaplan" navigation entry with custom bird logo
• Upload progress spinner with elapsed time counter for long-running knowledge uploads
• Improved success state with high-contrast SVG checkmark for dark theme compatibility
• Unit tests for all controllers (42 tests, 114 assertions)
• CI pipeline with PHP lint (PSR-12), PHPUnit, ESLint, Prettier, and Vite build checks

Knowledge-file lifecycle + per-file chat, built on Synaplan's file-lifecycle API (CORE-4).

Added

• Chat about this file — a new Files action opens a focused AI chat about a single file (text / JSON / XML), wiring up the previously unused per-file chat modal.
• Update / Remove in Knowledge — the "Add to AI Knowledge" dialog now detects when a file is already indexed and switches to Update in Knowledge (re-index in place, no duplicates) and offers Remove from Knowledge.
• "Changed since indexed" indicator — the dialog shows a warning when the Nextcloud file was modified after it was last indexed, with one-click update.

Changed

• Knowledge uploads now send the Nextcloud file id (source_id) and etag (source_etag) and use overwrite-by-source, so re-adding a file (or adding a changed one) updates its knowledge entry in place instead of creating a duplicate.

Requires

• For the update/stale/remove features: a Synaplan backend with the knowledge-file lifecycle API (source_id / source_etag / overwrite, check-stale, mark-stale). Against older backends the dialog degrades to plain "Add to Knowledge".

Added

• Backend environment switch (Live / Local). Admin settings now keep two separate Synaplan profiles — Live (production) and Local (development) — each with its own URL and API key, plus an Active environment toggle that flips the whole app (chat, media, knowledge, proxy) between them in one click. Makes it easy for any developer to point the app at a local Synaplan dev backend (e.g. http://localhost:8000) and switch back to live without re-entering credentials.

Major upgrade focused on language handling, personalisation, real streaming, and a clearer, better-formatted chat experience.

Added

• App version in the chat tagline — the Research/launcher header now shows the running version (e.g. "Ask anything v1.2.1 — powered by Synaplan"), read dynamically from the installed app version.
• Configurable answer language — a new "Language & AI behaviour" admin section lets you pick a default answer language and toggle "Use each user's Nextcloud interface language when supported". A German user then gets German answers automatically, with the default as the fallback. This replaces the previous behaviour where chat and document tools were effectively always English. The resolved language is injected as a system instruction into the streaming chat and passed through to summaries/translations, and the Summarize/Translate dialogs pre-select it.
• Personal memories in chat — when Synaplan's memory service (Qdrant) is reachable and the admin has enabled it ("Allow personal memories in chat"), the Research/launcher chat shows a "Use my memories" switch (on by default). Relevant memories are searched per question and injected as context so answers are personalised. The control is hidden entirely when the service is unavailable.
• Live chat status row — at-a-glance chips show the active model, the answer language, whether image (/pic) and video (/vid) generation are available, the selected knowledge base, and the memory state. Each answer is captioned with the model that actually produced it.
• Staged progress while waiting — instead of a static "Thinking…", the chat now reports what the backend is doing: Sorting your request → Searching the knowledge base → Recalling your memories → Generating the answer, stopping the moment streaming begins.
• Markdown rendering for document results — Summarize and Translate results now render Markdown (headings, bold, lists, code, links) via NcRichText, matching the chat, instead of showing raw **/- source.
• Client-config endpoint (api#clientConfig) exposing the resolved language and memory availability to the frontend.
• Staged upload progress — the "Add to Knowledge" dialog shows a progress bar and a phase checklist (Uploading → Extracting text → Creating chunks → Generating embeddings) with elapsed time.

Changed

• Real token-by-token streaming — chat answers now stream live. The SSE relay was rewritten to a direct cURL pass-through (SseProxyResponse) that writes each chunk straight to the browser and defeats PHP/Nextcloud output buffering (disables zlib/output buffering, Accept-Encoding: identity, X-Accel-Buffering: no, connection priming). Previously the answer arrived all at once after a long pause.
• Higher-contrast status chips — active capabilities use a solid brand fill with guaranteed-contrast text; disabled ones are clearly muted but still legible in both light and dark mode.
• Single-ring loading spinner — Summarize/Translate dialogs use one clean spinner instead of the twin-circle icon.
• Tighter Markdown spacing — list items in rendered answers no longer have oversized gaps.
• Per-chunk CSS injection (relativeCSSInjection) so shared component styles (NcSelect, NcCheckboxRadioSwitch) are injected by every entry that uses them — fixes unstyled dropdowns/switches on the Research page, the in-page launcher, and the file-action dialogs.

Fixed

• Answers are no longer "always English" — the chat/streaming path previously sent a hard-coded en output language; it now respects the admin default and/or the user's interface language.
• Admin toggles are clickable — the language/memory switches (and the chat "Use my memories" switch) used the wrong prop/event binding for @nextcloud/vue v9 and could not be toggled; they now use the correct v-model.
• Knowledge upload no longer shows a false success — when the server stores 0 chunks (empty/image-only file, or the embedding service is unavailable), the dialog reports a clear error instead of a green "File added" screen with "Chunks created: 0".
• Removed leftover "this is a demo" text from the knowledge-upload progress hint.

Security / dependencies

• Updated vite to 7.3.5, fixing a moderate dev-server advisory affecting 7.0.0–7.3.3.
• Refreshed dependency lockfiles: friendsofphp/php-cs-fixer 3.95.2, prettier 3.8.4, @nextcloud/vue 9.8.2, @nextcloud/axios 2.6.0, @nextcloud/dialogs 7.4.0, vue 3.5.38, and @nextcloud/initial-state 3.0.0.

Major upgrade focused on language handling, personalisation, real streaming, and a clearer, better-formatted chat experience.

Added

• Configurable answer language — a new "Language & AI behaviour" admin section lets you pick a default answer language and toggle "Use each user's Nextcloud interface language when supported". A German user then gets German answers automatically, with the default as the fallback. This replaces the previous behaviour where chat and document tools were effectively always English. The resolved language is injected as a system instruction into the streaming chat and passed through to summaries/translations, and the Summarize/Translate dialogs pre-select it.
• Personal memories in chat — when Synaplan's memory service (Qdrant) is reachable and the admin has enabled it ("Allow personal memories in chat"), the Research/launcher chat shows a "Use my memories" switch (on by default). Relevant memories are searched per question and injected as context so answers are personalised. The control is hidden entirely when the service is unavailable.
• Live chat status row — at-a-glance chips show the active model, the answer language, whether image (/pic) and video (/vid) generation are available, the selected knowledge base, and the memory state. Each answer is captioned with the model that actually produced it.
• Staged progress while waiting — instead of a static "Thinking…", the chat now reports what the backend is doing: Sorting your request → Searching the knowledge base → Recalling your memories → Generating the answer, stopping the moment streaming begins.
• Markdown rendering for document results — Summarize and Translate results now render Markdown (headings, bold, lists, code, links) via NcRichText, matching the chat, instead of showing raw **/- source.
• Client-config endpoint (api#clientConfig) exposing the resolved language and memory availability to the frontend.
• Staged upload progress — the "Add to Knowledge" dialog shows a progress bar and a phase checklist (Uploading → Extracting text → Creating chunks → Generating embeddings) with elapsed time.

Changed

• Real token-by-token streaming — chat answers now stream live. The SSE relay was rewritten to a direct cURL pass-through (SseProxyResponse) that writes each chunk straight to the browser and defeats PHP/Nextcloud output buffering (disables zlib/output buffering, Accept-Encoding: identity, X-Accel-Buffering: no, connection priming). Previously the answer arrived all at once after a long pause.
• Higher-contrast status chips — active capabilities use a solid brand fill with guaranteed-contrast text; disabled ones are clearly muted but still legible in both light and dark mode.
• Single-ring loading spinner — Summarize/Translate dialogs use one clean spinner instead of the twin-circle icon.
• Tighter Markdown spacing — list items in rendered answers no longer have oversized gaps.

Fixed

• Answers are no longer "always English" — the chat/streaming path previously sent a hard-coded en output language; it now respects the admin default and/or the user's interface language.
• Admin toggles are clickable — the language/memory switches (and the chat "Use my memories" switch) used the wrong prop/event binding for @nextcloud/vue v9 and could not be toggled; they now use the correct v-model.
• Knowledge upload no longer shows a false success — when the server stores 0 chunks (empty/image-only file, or the embedding service is unavailable), the dialog reports a clear error instead of a green "File added" screen with "Chunks created: 0".
• Removed leftover "this is a demo" text from the knowledge-upload progress hint.

Security / dependencies

• Updated vite to 7.3.5, fixing a moderate dev-server advisory affecting 7.0.0–7.3.3.
• Refreshed dependency lockfiles: friendsofphp/php-cs-fixer 3.95.2, prettier 3.8.4, @nextcloud/vue 9.8.2, @nextcloud/axios 2.6.0, @nextcloud/dialogs 7.4.0, vue 3.5.38, and @nextcloud/initial-state 3.0.0.

Added

• Summarize documents via file context menu (bullet points, paragraph, abstractive; short/medium/long; 12 output languages)
• Translate documents via file context menu (12 target languages)
• Add to AI Knowledge — upload files to the Synaplan vector knowledge base with group management
• Research Chat — full-page AI assistant accessible from top navigation with knowledge group and LLM model selection
• Support for binary document formats (PDF, DOCX, ODT, XLSX, PPTX) via Synaplan's Tika text extraction
• Admin settings page with Synaplan URL and API key configuration
• Connection test functionality in admin settings
• Top-level "Synaplan" navigation entry with custom bird logo
• Upload progress spinner with elapsed time counter for long-running knowledge uploads
• Improved success state with high-contrast SVG checkmark for dark theme compatibility
• Unit tests for all controllers (42 tests, 114 assertions)
• CI pipeline with PHP lint (PSR-12), PHPUnit, ESLint, Prettier, and Vite build checks

Added

• Backend environment switch (Live / Local). Admin settings now keep two separate Synaplan profiles — Live (production) and Local (development) — each with its own URL and API key, plus an Active environment toggle that flips the whole app (chat, media, knowledge, proxy) between them in one click. Makes it easy for any developer to point the app at a local Synaplan dev backend (e.g. http://localhost:8000) and switch back to live without re-entering credentials.

Major upgrade focused on language handling, personalisation, real streaming, and a clearer, better-formatted chat experience.

Added

• App version in the chat tagline — the Research/launcher header now shows the running version (e.g. "Ask anything v1.2.1 — powered by Synaplan"), read dynamically from the installed app version.
• Configurable answer language — a new "Language & AI behaviour" admin section lets you pick a default answer language and toggle "Use each user's Nextcloud interface language when supported". A German user then gets German answers automatically, with the default as the fallback. This replaces the previous behaviour where chat and document tools were effectively always English. The resolved language is injected as a system instruction into the streaming chat and passed through to summaries/translations, and the Summarize/Translate dialogs pre-select it.
• Personal memories in chat — when Synaplan's memory service (Qdrant) is reachable and the admin has enabled it ("Allow personal memories in chat"), the Research/launcher chat shows a "Use my memories" switch (on by default). Relevant memories are searched per question and injected as context so answers are personalised. The control is hidden entirely when the service is unavailable.
• Live chat status row — at-a-glance chips show the active model, the answer language, whether image (/pic) and video (/vid) generation are available, the selected knowledge base, and the memory state. Each answer is captioned with the model that actually produced it.
• Staged progress while waiting — instead of a static "Thinking…", the chat now reports what the backend is doing: Sorting your request → Searching the knowledge base → Recalling your memories → Generating the answer, stopping the moment streaming begins.
• Markdown rendering for document results — Summarize and Translate results now render Markdown (headings, bold, lists, code, links) via NcRichText, matching the chat, instead of showing raw **/- source.
• Client-config endpoint (api#clientConfig) exposing the resolved language and memory availability to the frontend.
• Staged upload progress — the "Add to Knowledge" dialog shows a progress bar and a phase checklist (Uploading → Extracting text → Creating chunks → Generating embeddings) with elapsed time.

Changed

• Real token-by-token streaming — chat answers now stream live. The SSE relay was rewritten to a direct cURL pass-through (SseProxyResponse) that writes each chunk straight to the browser and defeats PHP/Nextcloud output buffering (disables zlib/output buffering, Accept-Encoding: identity, X-Accel-Buffering: no, connection priming). Previously the answer arrived all at once after a long pause.
• Higher-contrast status chips — active capabilities use a solid brand fill with guaranteed-contrast text; disabled ones are clearly muted but still legible in both light and dark mode.
• Single-ring loading spinner — Summarize/Translate dialogs use one clean spinner instead of the twin-circle icon.
• Tighter Markdown spacing — list items in rendered answers no longer have oversized gaps.
• Per-chunk CSS injection (relativeCSSInjection) so shared component styles (NcSelect, NcCheckboxRadioSwitch) are injected by every entry that uses them — fixes unstyled dropdowns/switches on the Research page, the in-page launcher, and the file-action dialogs.

Fixed

• Answers are no longer "always English" — the chat/streaming path previously sent a hard-coded en output language; it now respects the admin default and/or the user's interface language.
• Admin toggles are clickable — the language/memory switches (and the chat "Use my memories" switch) used the wrong prop/event binding for @nextcloud/vue v9 and could not be toggled; they now use the correct v-model.
• Knowledge upload no longer shows a false success — when the server stores 0 chunks (empty/image-only file, or the embedding service is unavailable), the dialog reports a clear error instead of a green "File added" screen with "Chunks created: 0".
• Removed leftover "this is a demo" text from the knowledge-upload progress hint.

Security / dependencies

• Updated vite to 7.3.5, fixing a moderate dev-server advisory affecting 7.0.0–7.3.3.
• Refreshed dependency lockfiles: friendsofphp/php-cs-fixer 3.95.2, prettier 3.8.4, @nextcloud/vue 9.8.2, @nextcloud/axios 2.6.0, @nextcloud/dialogs 7.4.0, vue 3.5.38, and @nextcloud/initial-state 3.0.0.

Major upgrade focused on language handling, personalisation, real streaming, and a clearer, better-formatted chat experience.

Added

• Configurable answer language — a new "Language & AI behaviour" admin section lets you pick a default answer language and toggle "Use each user's Nextcloud interface language when supported". A German user then gets German answers automatically, with the default as the fallback. This replaces the previous behaviour where chat and document tools were effectively always English. The resolved language is injected as a system instruction into the streaming chat and passed through to summaries/translations, and the Summarize/Translate dialogs pre-select it.
• Personal memories in chat — when Synaplan's memory service (Qdrant) is reachable and the admin has enabled it ("Allow personal memories in chat"), the Research/launcher chat shows a "Use my memories" switch (on by default). Relevant memories are searched per question and injected as context so answers are personalised. The control is hidden entirely when the service is unavailable.
• Live chat status row — at-a-glance chips show the active model, the answer language, whether image (/pic) and video (/vid) generation are available, the selected knowledge base, and the memory state. Each answer is captioned with the model that actually produced it.
• Staged progress while waiting — instead of a static "Thinking…", the chat now reports what the backend is doing: Sorting your request → Searching the knowledge base → Recalling your memories → Generating the answer, stopping the moment streaming begins.
• Markdown rendering for document results — Summarize and Translate results now render Markdown (headings, bold, lists, code, links) via NcRichText, matching the chat, instead of showing raw **/- source.
• Client-config endpoint (api#clientConfig) exposing the resolved language and memory availability to the frontend.
• Staged upload progress — the "Add to Knowledge" dialog shows a progress bar and a phase checklist (Uploading → Extracting text → Creating chunks → Generating embeddings) with elapsed time.

Changed

• Real token-by-token streaming — chat answers now stream live. The SSE relay was rewritten to a direct cURL pass-through (SseProxyResponse) that writes each chunk straight to the browser and defeats PHP/Nextcloud output buffering (disables zlib/output buffering, Accept-Encoding: identity, X-Accel-Buffering: no, connection priming). Previously the answer arrived all at once after a long pause.
• Higher-contrast status chips — active capabilities use a solid brand fill with guaranteed-contrast text; disabled ones are clearly muted but still legible in both light and dark mode.
• Single-ring loading spinner — Summarize/Translate dialogs use one clean spinner instead of the twin-circle icon.
• Tighter Markdown spacing — list items in rendered answers no longer have oversized gaps.

Fixed

• Answers are no longer "always English" — the chat/streaming path previously sent a hard-coded en output language; it now respects the admin default and/or the user's interface language.
• Admin toggles are clickable — the language/memory switches (and the chat "Use my memories" switch) used the wrong prop/event binding for @nextcloud/vue v9 and could not be toggled; they now use the correct v-model.
• Knowledge upload no longer shows a false success — when the server stores 0 chunks (empty/image-only file, or the embedding service is unavailable), the dialog reports a clear error instead of a green "File added" screen with "Chunks created: 0".
• Removed leftover "this is a demo" text from the knowledge-upload progress hint.

Security / dependencies

• Updated vite to 7.3.5, fixing a moderate dev-server advisory affecting 7.0.0–7.3.3.
• Refreshed dependency lockfiles: friendsofphp/php-cs-fixer 3.95.2, prettier 3.8.4, @nextcloud/vue 9.8.2, @nextcloud/axios 2.6.0, @nextcloud/dialogs 7.4.0, vue 3.5.38, and @nextcloud/initial-state 3.0.0.

Added

• Summarize documents via file context menu (bullet points, paragraph, abstractive; short/medium/long; 12 output languages)
• Translate documents via file context menu (12 target languages)
• Add to AI Knowledge — upload files to the Synaplan vector knowledge base with group management
• Research Chat — full-page AI assistant accessible from top navigation with knowledge group and LLM model selection
• Support for binary document formats (PDF, DOCX, ODT, XLSX, PPTX) via Synaplan's Tika text extraction
• Admin settings page with Synaplan URL and API key configuration
• Connection test functionality in admin settings
• Top-level "Synaplan" navigation entry with custom bird logo
• Upload progress spinner with elapsed time counter for long-running knowledge uploads
• Improved success state with high-contrast SVG checkmark for dark theme compatibility
• Unit tests for all controllers (42 tests, 114 assertions)
• CI pipeline with PHP lint (PSR-12), PHPUnit, ESLint, Prettier, and Vite build checks

Added

• Backend environment switch (Live / Local). Admin settings now keep two separate Synaplan profiles — Live (production) and Local (development) — each with its own URL and API key, plus an Active environment toggle that flips the whole app (chat, media, knowledge, proxy) between them in one click. Makes it easy for any developer to point the app at a local Synaplan dev backend (e.g. http://localhost:8000) and switch back to live without re-entering credentials.

Major upgrade focused on language handling, personalisation, real streaming, and a clearer, better-formatted chat experience.

Added

• App version in the chat tagline — the Research/launcher header now shows the running version (e.g. "Ask anything v1.2.1 — powered by Synaplan"), read dynamically from the installed app version.
• Configurable answer language — a new "Language & AI behaviour" admin section lets you pick a default answer language and toggle "Use each user's Nextcloud interface language when supported". A German user then gets German answers automatically, with the default as the fallback. This replaces the previous behaviour where chat and document tools were effectively always English. The resolved language is injected as a system instruction into the streaming chat and passed through to summaries/translations, and the Summarize/Translate dialogs pre-select it.
• Personal memories in chat — when Synaplan's memory service (Qdrant) is reachable and the admin has enabled it ("Allow personal memories in chat"), the Research/launcher chat shows a "Use my memories" switch (on by default). Relevant memories are searched per question and injected as context so answers are personalised. The control is hidden entirely when the service is unavailable.
• Live chat status row — at-a-glance chips show the active model, the answer language, whether image (/pic) and video (/vid) generation are available, the selected knowledge base, and the memory state. Each answer is captioned with the model that actually produced it.
• Staged progress while waiting — instead of a static "Thinking…", the chat now reports what the backend is doing: Sorting your request → Searching the knowledge base → Recalling your memories → Generating the answer, stopping the moment streaming begins.
• Markdown rendering for document results — Summarize and Translate results now render Markdown (headings, bold, lists, code, links) via NcRichText, matching the chat, instead of showing raw **/- source.
• Client-config endpoint (api#clientConfig) exposing the resolved language and memory availability to the frontend.
• Staged upload progress — the "Add to Knowledge" dialog shows a progress bar and a phase checklist (Uploading → Extracting text → Creating chunks → Generating embeddings) with elapsed time.

Changed

• Real token-by-token streaming — chat answers now stream live. The SSE relay was rewritten to a direct cURL pass-through (SseProxyResponse) that writes each chunk straight to the browser and defeats PHP/Nextcloud output buffering (disables zlib/output buffering, Accept-Encoding: identity, X-Accel-Buffering: no, connection priming). Previously the answer arrived all at once after a long pause.
• Higher-contrast status chips — active capabilities use a solid brand fill with guaranteed-contrast text; disabled ones are clearly muted but still legible in both light and dark mode.
• Single-ring loading spinner — Summarize/Translate dialogs use one clean spinner instead of the twin-circle icon.
• Tighter Markdown spacing — list items in rendered answers no longer have oversized gaps.
• Per-chunk CSS injection (relativeCSSInjection) so shared component styles (NcSelect, NcCheckboxRadioSwitch) are injected by every entry that uses them — fixes unstyled dropdowns/switches on the Research page, the in-page launcher, and the file-action dialogs.

Fixed

• Answers are no longer "always English" — the chat/streaming path previously sent a hard-coded en output language; it now respects the admin default and/or the user's interface language.
• Admin toggles are clickable — the language/memory switches (and the chat "Use my memories" switch) used the wrong prop/event binding for @nextcloud/vue v9 and could not be toggled; they now use the correct v-model.
• Knowledge upload no longer shows a false success — when the server stores 0 chunks (empty/image-only file, or the embedding service is unavailable), the dialog reports a clear error instead of a green "File added" screen with "Chunks created: 0".
• Removed leftover "this is a demo" text from the knowledge-upload progress hint.

Security / dependencies

• Updated vite to 7.3.5, fixing a moderate dev-server advisory affecting 7.0.0–7.3.3.
• Refreshed dependency lockfiles: friendsofphp/php-cs-fixer 3.95.2, prettier 3.8.4, @nextcloud/vue 9.8.2, @nextcloud/axios 2.6.0, @nextcloud/dialogs 7.4.0, vue 3.5.38, and @nextcloud/initial-state 3.0.0.

Major upgrade focused on language handling, personalisation, real streaming, and a clearer, better-formatted chat experience.

Added

• Configurable answer language — a new "Language & AI behaviour" admin section lets you pick a default answer language and toggle "Use each user's Nextcloud interface language when supported". A German user then gets German answers automatically, with the default as the fallback. This replaces the previous behaviour where chat and document tools were effectively always English. The resolved language is injected as a system instruction into the streaming chat and passed through to summaries/translations, and the Summarize/Translate dialogs pre-select it.
• Personal memories in chat — when Synaplan's memory service (Qdrant) is reachable and the admin has enabled it ("Allow personal memories in chat"), the Research/launcher chat shows a "Use my memories" switch (on by default). Relevant memories are searched per question and injected as context so answers are personalised. The control is hidden entirely when the service is unavailable.
• Live chat status row — at-a-glance chips show the active model, the answer language, whether image (/pic) and video (/vid) generation are available, the selected knowledge base, and the memory state. Each answer is captioned with the model that actually produced it.
• Staged progress while waiting — instead of a static "Thinking…", the chat now reports what the backend is doing: Sorting your request → Searching the knowledge base → Recalling your memories → Generating the answer, stopping the moment streaming begins.
• Markdown rendering for document results — Summarize and Translate results now render Markdown (headings, bold, lists, code, links) via NcRichText, matching the chat, instead of showing raw **/- source.
• Client-config endpoint (api#clientConfig) exposing the resolved language and memory availability to the frontend.
• Staged upload progress — the "Add to Knowledge" dialog shows a progress bar and a phase checklist (Uploading → Extracting text → Creating chunks → Generating embeddings) with elapsed time.

Changed

• Real token-by-token streaming — chat answers now stream live. The SSE relay was rewritten to a direct cURL pass-through (SseProxyResponse) that writes each chunk straight to the browser and defeats PHP/Nextcloud output buffering (disables zlib/output buffering, Accept-Encoding: identity, X-Accel-Buffering: no, connection priming). Previously the answer arrived all at once after a long pause.
• Higher-contrast status chips — active capabilities use a solid brand fill with guaranteed-contrast text; disabled ones are clearly muted but still legible in both light and dark mode.
• Single-ring loading spinner — Summarize/Translate dialogs use one clean spinner instead of the twin-circle icon.
• Tighter Markdown spacing — list items in rendered answers no longer have oversized gaps.

Fixed

• Answers are no longer "always English" — the chat/streaming path previously sent a hard-coded en output language; it now respects the admin default and/or the user's interface language.
• Admin toggles are clickable — the language/memory switches (and the chat "Use my memories" switch) used the wrong prop/event binding for @nextcloud/vue v9 and could not be toggled; they now use the correct v-model.
• Knowledge upload no longer shows a false success — when the server stores 0 chunks (empty/image-only file, or the embedding service is unavailable), the dialog reports a clear error instead of a green "File added" screen with "Chunks created: 0".
• Removed leftover "this is a demo" text from the knowledge-upload progress hint.

Security / dependencies

• Updated vite to 7.3.5, fixing a moderate dev-server advisory affecting 7.0.0–7.3.3.
• Refreshed dependency lockfiles: friendsofphp/php-cs-fixer 3.95.2, prettier 3.8.4, @nextcloud/vue 9.8.2, @nextcloud/axios 2.6.0, @nextcloud/dialogs 7.4.0, vue 3.5.38, and @nextcloud/initial-state 3.0.0.

Added

• Summarize documents via file context menu (bullet points, paragraph, abstractive; short/medium/long; 12 output languages)
• Translate documents via file context menu (12 target languages)
• Add to AI Knowledge — upload files to the Synaplan vector knowledge base with group management
• Research Chat — full-page AI assistant accessible from top navigation with knowledge group and LLM model selection
• Support for binary document formats (PDF, DOCX, ODT, XLSX, PPTX) via Synaplan's Tika text extraction
• Admin settings page with Synaplan URL and API key configuration
• Connection test functionality in admin settings
• Top-level "Synaplan" navigation entry with custom bird logo
• Upload progress spinner with elapsed time counter for long-running knowledge uploads
• Improved success state with high-contrast SVG checkmark for dark theme compatibility
• Unit tests for all controllers (42 tests, 114 assertions)
• CI pipeline with PHP lint (PSR-12), PHPUnit, ESLint, Prettier, and Vite build checks

Added

• Backend environment switch (Live / Local). Admin settings now keep two separate Synaplan profiles — Live (production) and Local (development) — each with its own URL and API key, plus an Active environment toggle that flips the whole app (chat, media, knowledge, proxy) between them in one click. Makes it easy for any developer to point the app at a local Synaplan dev backend (e.g. http://localhost:8000) and switch back to live without re-entering credentials.

Major upgrade focused on language handling, personalisation, real streaming, and a clearer, better-formatted chat experience.

Added

• App version in the chat tagline — the Research/launcher header now shows the running version (e.g. "Ask anything v1.2.1 — powered by Synaplan"), read dynamically from the installed app version.
• Configurable answer language — a new "Language & AI behaviour" admin section lets you pick a default answer language and toggle "Use each user's Nextcloud interface language when supported". A German user then gets German answers automatically, with the default as the fallback. This replaces the previous behaviour where chat and document tools were effectively always English. The resolved language is injected as a system instruction into the streaming chat and passed through to summaries/translations, and the Summarize/Translate dialogs pre-select it.
• Personal memories in chat — when Synaplan's memory service (Qdrant) is reachable and the admin has enabled it ("Allow personal memories in chat"), the Research/launcher chat shows a "Use my memories" switch (on by default). Relevant memories are searched per question and injected as context so answers are personalised. The control is hidden entirely when the service is unavailable.
• Live chat status row — at-a-glance chips show the active model, the answer language, whether image (/pic) and video (/vid) generation are available, the selected knowledge base, and the memory state. Each answer is captioned with the model that actually produced it.
• Staged progress while waiting — instead of a static "Thinking…", the chat now reports what the backend is doing: Sorting your request → Searching the knowledge base → Recalling your memories → Generating the answer, stopping the moment streaming begins.
• Markdown rendering for document results — Summarize and Translate results now render Markdown (headings, bold, lists, code, links) via NcRichText, matching the chat, instead of showing raw **/- source.
• Client-config endpoint (api#clientConfig) exposing the resolved language and memory availability to the frontend.
• Staged upload progress — the "Add to Knowledge" dialog shows a progress bar and a phase checklist (Uploading → Extracting text → Creating chunks → Generating embeddings) with elapsed time.

Changed

• Real token-by-token streaming — chat answers now stream live. The SSE relay was rewritten to a direct cURL pass-through (SseProxyResponse) that writes each chunk straight to the browser and defeats PHP/Nextcloud output buffering (disables zlib/output buffering, Accept-Encoding: identity, X-Accel-Buffering: no, connection priming). Previously the answer arrived all at once after a long pause.
• Higher-contrast status chips — active capabilities use a solid brand fill with guaranteed-contrast text; disabled ones are clearly muted but still legible in both light and dark mode.
• Single-ring loading spinner — Summarize/Translate dialogs use one clean spinner instead of the twin-circle icon.
• Tighter Markdown spacing — list items in rendered answers no longer have oversized gaps.
• Per-chunk CSS injection (relativeCSSInjection) so shared component styles (NcSelect, NcCheckboxRadioSwitch) are injected by every entry that uses them — fixes unstyled dropdowns/switches on the Research page, the in-page launcher, and the file-action dialogs.

Fixed

• Answers are no longer "always English" — the chat/streaming path previously sent a hard-coded en output language; it now respects the admin default and/or the user's interface language.
• Admin toggles are clickable — the language/memory switches (and the chat "Use my memories" switch) used the wrong prop/event binding for @nextcloud/vue v9 and could not be toggled; they now use the correct v-model.
• Knowledge upload no longer shows a false success — when the server stores 0 chunks (empty/image-only file, or the embedding service is unavailable), the dialog reports a clear error instead of a green "File added" screen with "Chunks created: 0".
• Removed leftover "this is a demo" text from the knowledge-upload progress hint.

Security / dependencies

• Updated vite to 7.3.5, fixing a moderate dev-server advisory affecting 7.0.0–7.3.3.
• Refreshed dependency lockfiles: friendsofphp/php-cs-fixer 3.95.2, prettier 3.8.4, @nextcloud/vue 9.8.2, @nextcloud/axios 2.6.0, @nextcloud/dialogs 7.4.0, vue 3.5.38, and @nextcloud/initial-state 3.0.0.

Major upgrade focused on language handling, personalisation, real streaming, and a clearer, better-formatted chat experience.

Added

• Configurable answer language — a new "Language & AI behaviour" admin section lets you pick a default answer language and toggle "Use each user's Nextcloud interface language when supported". A German user then gets German answers automatically, with the default as the fallback. This replaces the previous behaviour where chat and document tools were effectively always English. The resolved language is injected as a system instruction into the streaming chat and passed through to summaries/translations, and the Summarize/Translate dialogs pre-select it.
• Personal memories in chat — when Synaplan's memory service (Qdrant) is reachable and the admin has enabled it ("Allow personal memories in chat"), the Research/launcher chat shows a "Use my memories" switch (on by default). Relevant memories are searched per question and injected as context so answers are personalised. The control is hidden entirely when the service is unavailable.
• Live chat status row — at-a-glance chips show the active model, the answer language, whether image (/pic) and video (/vid) generation are available, the selected knowledge base, and the memory state. Each answer is captioned with the model that actually produced it.
• Staged progress while waiting — instead of a static "Thinking…", the chat now reports what the backend is doing: Sorting your request → Searching the knowledge base → Recalling your memories → Generating the answer, stopping the moment streaming begins.
• Markdown rendering for document results — Summarize and Translate results now render Markdown (headings, bold, lists, code, links) via NcRichText, matching the chat, instead of showing raw **/- source.
• Client-config endpoint (api#clientConfig) exposing the resolved language and memory availability to the frontend.
• Staged upload progress — the "Add to Knowledge" dialog shows a progress bar and a phase checklist (Uploading → Extracting text → Creating chunks → Generating embeddings) with elapsed time.

Changed

• Real token-by-token streaming — chat answers now stream live. The SSE relay was rewritten to a direct cURL pass-through (SseProxyResponse) that writes each chunk straight to the browser and defeats PHP/Nextcloud output buffering (disables zlib/output buffering, Accept-Encoding: identity, X-Accel-Buffering: no, connection priming). Previously the answer arrived all at once after a long pause.
• Higher-contrast status chips — active capabilities use a solid brand fill with guaranteed-contrast text; disabled ones are clearly muted but still legible in both light and dark mode.
• Single-ring loading spinner — Summarize/Translate dialogs use one clean spinner instead of the twin-circle icon.
• Tighter Markdown spacing — list items in rendered answers no longer have oversized gaps.

Fixed

• Answers are no longer "always English" — the chat/streaming path previously sent a hard-coded en output language; it now respects the admin default and/or the user's interface language.
• Admin toggles are clickable — the language/memory switches (and the chat "Use my memories" switch) used the wrong prop/event binding for @nextcloud/vue v9 and could not be toggled; they now use the correct v-model.
• Knowledge upload no longer shows a false success — when the server stores 0 chunks (empty/image-only file, or the embedding service is unavailable), the dialog reports a clear error instead of a green "File added" screen with "Chunks created: 0".
• Removed leftover "this is a demo" text from the knowledge-upload progress hint.

Security / dependencies

• Updated vite to 7.3.5, fixing a moderate dev-server advisory affecting 7.0.0–7.3.3.
• Refreshed dependency lockfiles: friendsofphp/php-cs-fixer 3.95.2, prettier 3.8.4, @nextcloud/vue 9.8.2, @nextcloud/axios 2.6.0, @nextcloud/dialogs 7.4.0, vue 3.5.38, and @nextcloud/initial-state 3.0.0.

Added

• Summarize documents via file context menu (bullet points, paragraph, abstractive; short/medium/long; 12 output languages)
• Translate documents via file context menu (12 target languages)
• Add to AI Knowledge — upload files to the Synaplan vector knowledge base with group management
• Research Chat — full-page AI assistant accessible from top navigation with knowledge group and LLM model selection
• Support for binary document formats (PDF, DOCX, ODT, XLSX, PPTX) via Synaplan's Tika text extraction
• Admin settings page with Synaplan URL and API key configuration
• Connection test functionality in admin settings
• Top-level "Synaplan" navigation entry with custom bird logo
• Upload progress spinner with elapsed time counter for long-running knowledge uploads
• Improved success state with high-contrast SVG checkmark for dark theme compatibility
• Unit tests for all controllers (42 tests, 114 assertions)
• CI pipeline with PHP lint (PSR-12), PHPUnit, ESLint, Prettier, and Vite build checks