feat: chat stop button, image paste, vision AI, search fixes

- Add stop button to all chat interfaces (floating, contextual, full-page)
- Add conversation sliding window (50 messages) to prevent context overflow
- Add chat timeout warning (30s toast)
- Force response language in chat system prompt (mandatory per-locale)
- Add image paste from clipboard in all note editors (card, list, input)
- Fix upload API to infer extension from MIME type for clipboard images
- Add image description support in note AI chat (base64 vision)
- Fix search regex crash on special characters (escape user input)
- Fix search case-insensitivity on PostgreSQL (mode: 'insensitive')
- Add try/catch around semantic search in chat route (prevent blocking)
- Add new chat button to floating AI assistant
- Fix empty thinking bubbles for reasoning models (filter non-text parts)
- Remove duplicate AI assistant toggle from note editor header
- Improve link metadata scraping (timeout, content-type check, relative URLs)

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

memento-note/app/api/chat/route.ts

@@ -7,6 +7,8 @@ import { auth } from '@/auth'
 import { loadTranslations, getTranslationValue, SupportedLanguage } from '@/lib/i18n'
 import { toolRegistry } from '@/lib/ai/tools'
 import { stepCountIs } from 'ai'
+import { readFile } from 'fs/promises'
+import path from 'path'
 
 export const maxDuration = 60
 
@@ -47,13 +49,14 @@ export async function POST(req: Request) {
 
   // 2. Parse request body — messages arrive as UIMessage[] from DefaultChatTransport
   const body = await req.json()
+  console.log('[Chat] body keys:', Object.keys(body), 'noteContext?', !!body.noteContext, 'images?', body.noteContext?.images?.length)
   const { messages: rawMessages, conversationId, notebookId, language, webSearch, noteContext } = body as {
     messages: UIMessage[]
     conversationId?: string
     notebookId?: string
     language?: string
     webSearch?: boolean
-    noteContext?: { title: string; content: string; tone: string }
+    noteContext?: { title: string; content: string; tone: string; images?: string[] }
   }
 
   // Convert UIMessages to CoreMessages for streamText
@@ -115,12 +118,17 @@ export async function POST(req: Request) {
   }
 
   // Also run semantic search for the specific query
-  const searchResults = await semanticSearchService.search(currentMessage, {
-    notebookId,
-    limit: notebookId ? 10 : 5,
-    threshold: notebookId ? 0.3 : 0.5,
-    defaultTitle: untitledText,
-  })
+  let searchResults: any[] = []
+  try {
+    searchResults = await semanticSearchService.search(currentMessage, {
+      notebookId,
+      limit: notebookId ? 10 : 5,
+      threshold: notebookId ? 0.3 : 0.5,
+      defaultTitle: untitledText,
+    })
+  } catch {
+    // Search failure should not block chat
+  }
 
   const searchNotes = searchResults
     .map((r) => `NOTE [${r.title || untitledText}]: ${r.content}`)
@@ -320,6 +328,27 @@ Tu as accès à ces outils pour des recherches approfondies :
     ? prompts.contextWithNotes
     : prompts.contextNoNotes
 
+  // Load note images as base64 for vision-capable models
+  let imageContextParts: Array<{ type: 'image'; image: string }> = []
+  if (noteContext?.images && noteContext.images.length > 0) {
+    console.log('[Chat] noteContext.images:', noteContext.images)
+    for (const imgPath of noteContext.images.slice(0, 4)) {
+      try {
+        const fullPath = path.join(process.cwd(), 'public', imgPath)
+        console.log('[Chat] reading image:', fullPath)
+        const buffer = await readFile(fullPath)
+        const ext = path.extname(imgPath).toLowerCase()
+        const mime = ext === '.png' ? 'image/png' : ext === '.gif' ? 'image/gif' : ext === '.webp' ? 'image/webp' : 'image/jpeg'
+        const base64 = `data:${mime};base64,${buffer.toString('base64')}`
+        imageContextParts.push({ type: 'image', image: base64 })
+        console.log('[Chat] image loaded, size:', buffer.length, 'bytes')
+      } catch (err) {
+        console.error('[Chat] failed to read image:', imgPath, err)
+      }
+    }
+    console.log('[Chat] total image parts:', imageContextParts.length)
+  }
+
   let copilotContext = ''
   if (noteContext) {
      copilotContext = `\n\n## Current Note Context
@@ -328,8 +357,9 @@ Title: ${noteContext.title || 'Untitled'}
 
 Content:
 ${noteContext.content || '(empty)'}
+${imageContextParts.length > 0 ? `\nImages: ${imageContextParts.length} image(s) attached. When the user asks about images, describe what you see in them.` : ''}
 
-The user wants you to write in a **${noteContext.tone || 'professional'}** tone. 
+The user wants you to write in a **${noteContext.tone || 'professional'}** tone.
 Keep your suggestions tailored to this note and tone. You can suggest rewrites, answer questions about the note, or draft new sections.`
   }
 
@@ -338,7 +368,9 @@ ${copilotContext}
 
 ${contextBlock}
 
-${lang === 'en' ? 'Respond in the user\'s language.' : lang === 'fr' ? 'Réponds dans la langue de l\'utilisateur.' : 'Respond in the user\'s language.'}`
+## LANGUAGE RULE (MANDATORY)
+You MUST respond in ${lang === 'en' ? 'English' : lang === 'fr' ? 'French' : lang === 'fa' ? 'Persian (Farsi)' : lang === 'es' ? 'Spanish' : lang === 'de' ? 'German' : lang === 'it' ? 'Italian' : 'English'}.
+Never switch to another language. Even if the user writes in a different language, respond in the configured language.`
 
   // 6. Build message history from DB + current messages
   const dbHistory = conversation.messages.map((m: { role: string; content: string }) => ({
@@ -355,10 +387,25 @@ ${lang === 'en' ? 'Respond in the user\'s language.' : lang === 'fr' ? 'Réponds
       currentDbMessage.role !== 'user' ||
       currentDbMessage.content !== lastIncoming.content)
 
-  const allMessages: Array<{ role: 'user' | 'assistant' | 'system'; content: string }> = isNewMessage
+  let allMessages: Array<{ role: 'user' | 'assistant' | 'system'; content: string | Array<any> }> = isNewMessage
     ? [...dbHistory, { role: lastIncoming.role, content: lastIncoming.content }]
     : dbHistory
 
+  // Inject note images as a context message for vision models
+  if (imageContextParts.length > 0) {
+    allMessages = [
+      { role: 'user', content: [{ type: 'text' as const, text: '[Attached note images — use these when the user asks about images]' }, ...imageContextParts] },
+      { role: 'assistant', content: 'Understood. I can see the attached images and will describe or analyze them when asked.' },
+      ...allMessages,
+    ]
+  }
+
+  // Sliding window: keep first 2 messages (context) + last 48 to avoid context overflow
+  const WINDOW = 50
+  if (allMessages.length > WINDOW) {
+    allMessages = [...allMessages.slice(0, 2), ...allMessages.slice(-(WINDOW - 2))]
+  }
+
   // 7. Get chat provider model
   const config = await getSystemConfig()
   const provider = getChatProvider(config)
@@ -389,7 +436,7 @@ ${lang === 'en' ? 'Respond in the user\'s language.' : lang === 'fr' ? 'Réponds
   const result = streamText({
     model,
     system: systemPrompt,
-    messages: allMessages,
+    messages: allMessages as any,
     tools: chatTools,
     stopWhen: stepCountIs(5),
     async onFinish({ text }) {

memento-note/app/api/upload/route.ts

@@ -33,9 +33,16 @@ export async function POST(request: NextRequest) {
     }
 
     const buffer = Buffer.from(await file.arrayBuffer())
-    const ext = path.extname(file.name).toLowerCase()
+    // Resolve extension from file name, falling back to MIME type (e.g. clipboard pastes)
+    let ext = path.extname(file.name).toLowerCase()
     if (!['.jpg', '.jpeg', '.png', '.gif', '.webp'].includes(ext)) {
-      return NextResponse.json({ error: 'Invalid file extension' }, { status: 400 })
+      const mimeToExt: Record<string, string> = {
+        'image/jpeg': '.jpg',
+        'image/png': '.png',
+        'image/gif': '.gif',
+        'image/webp': '.webp',
+      }
+      ext = mimeToExt[file.type] || '.png'
     }
     const filename = `${randomUUID()}${ext}`