* fix: preserve access_control when cloning prompts
Cloned prompts now inherit the original prompt's access_control settings
instead of defaulting to null/public. This ensures that permission and
sharing restrictions are carried over with clones as expected.
If the original prompt doesn't have access_control set, defaults to {}
(private) which matches the behavior of newly created prompts.
Fixes#19360
* fix: clone access control
Use json.dumps with ensure_ascii=False to avoid converting Chinese
descriptions into Unicode escape sequences.
This makes logs easier to read and significantly improves LLM tool
selection accuracy (from ~48% to ~67%) by preserving the original
natural language context.
* i18n(de): Optimize and standardize German translations for consistency and tone
- Unified user addressing to the formal "Sie" (polite form) throughout the UI, retaining "Du" only for specific persona contexts
- Fixed context-specific translation errors (e.g., "Call" from "Anrufen" to "Aufrufen", "Generation" from "Generation" to "Generierung", "Dokomentenintelligenz" typo)
- Corrected hyphenation for compound nouns (e.g., "JSON-Datei", "API-Schlüssel", "Prompt-Variable") and general grammar
- Improved tone and clarity of error messages and prompts for better readability and a more professional UI tone
- Standardized technical concepts across all translation files
* Update translations based on the feedback provided
---------
Co-authored-by: Tim Baek <tim@openwebui.com>
Fixes#19742 where file attachment containers did not inherit the chat direction setting, causing layout issues when typing in RTL languages like Persian.
Added dir={$settings?.chatDirection ?? 'auto'} to file container divs in:
- MessageInput.svelte (chat input file attachments)
- UserMessage.svelte (user message file display)
- ResponseMessage.svelte (response message file display)
- channel/Messages/Message.svelte (channel message file display)
The fix ensures file containers automatically detect and respect RTL text direction, maintaining proper layout alignment for RTL languages while preserving LTR behavior for Latin text.
G30