The front-end can also upload files!

File Upload

• New attachment button in chat input box, supporting drag-and-drop, paste, and click uploads
• Supports common formats including images, videos, audio, PDF, and Office documents
• Single file max 50MB, up to 5 files simultaneously, automatic deduplication
• Uploads via HTTP multipart, with real progress bars, no longer limited by WebSocket message size
• Videos over 30 seconds are automatically trimmed to first, middle, and last 10 seconds concatenated, speeding up subsequent analysis
• Drag overlay uses frosted glass effect, matching native HA theme style
• Compact and elegant floating preview bar, silent removal on upload failure
• Disabled by default, enable in Settings → Interface Display → File Upload

Media Analysis Routing

• When images/videos are attached, AI is forced to prioritize calling MediaAnalyze to analyze content before responding
• Fixes the issue where first-round intent recognition skipped media analysis and replied "unable to understand"

Token Statistics Optimization

• After each round, real token usage is fetched from the backend, replacing frontend rough estimates
• Attachment labels, thinking content, and tool call results are all included in statistics
• Context window size uses the backend actual value, no longer hardcoded

Event Loop Blocking Fix

• Synchronous calls such as list_installed_skills, list_homeassistant_guide_docs are all wrapped with async_add_executor_job
• Involves ListInstalledSkillsTool, ReflectAndProposeTool, async_step_skill_editor, _ensure_prompt_store_fresh