sync: ingestion improvements from private repo

- Fixed dimension auto-assignment (locked dimensions now properly assigned)
- Added description field as grounding context for AI
- Updated embedding format: Title → Description → Content → Dimensions
- Description-weighted search (5x boost)
- Bug fixes: extraction tool dimension display, DimensionSearchModal closing
- Removed quickLink tool (use Quick Link workflow instead)
- Added regenerate-description API endpoint

src/services/agents/quickAdd.ts

@@ -13,6 +13,7 @@ export type QuickAddInputType = 'youtube' | 'website' | 'pdf' | 'note' | 'chat';
 export interface QuickAddInput {
   rawInput: string;
   mode?: QuickAddMode;
+  description?: string;
 }
 
 function isLikelyChatTranscript(raw: string): boolean {
@@ -197,24 +198,31 @@ async function handleExtractionQuickAdd(type: ExtractionQuickAddType, url: strin
   });
 }
 
-async function handleNoteQuickAdd(rawInput: string, task: string): Promise<string> {
+async function handleNoteQuickAdd(rawInput: string, task: string, userDescription?: string): Promise<string> {
   const content = rawInput.trim();
   if (!content) {
     throw new Error('Input is required to create a note');
   }
 
   const title = deriveNoteTitle(content);
+  const nodePayload: Record<string, unknown> = {
+    title,
+    content,
+    metadata: {
+      source: 'quick-add-note',
+      refined_at: new Date().toISOString(),
+    },
+  };
+
+  // If user provided a description, use it instead of auto-generating
+  if (userDescription && userDescription.trim()) {
+    nodePayload.description = userDescription.trim();
+  }
+
   const response = await fetch(`${process.env.NEXT_PUBLIC_BASE_URL || 'http://localhost:3000'}/api/nodes`, {
     method: 'POST',
     headers: { 'Content-Type': 'application/json' },
-    body: JSON.stringify({
-      title,
-      content,
-      metadata: {
-        source: 'quick-add-note',
-        refined_at: new Date().toISOString(),
-      },
-    }),
+    body: JSON.stringify(nodePayload),
   });
 
   const rawResult = await response.json();
@@ -332,7 +340,7 @@ async function handleChatTranscriptQuickAdd(rawInput: string, task: string): Pro
   });
 }
 
-export async function enqueueQuickAdd({ rawInput, mode }: QuickAddInput) {
+export async function enqueueQuickAdd({ rawInput, mode, description }: QuickAddInput) {
   const inputType = detectInputType(rawInput, mode);
   const context: string[] = (inputType === 'note' || inputType === 'chat') ? [] : [rawInput];
   const task = buildTaskPrompt(inputType, rawInput);
@@ -350,7 +358,7 @@ export async function enqueueQuickAdd({ rawInput, mode }: QuickAddInput) {
 
       let summary: string;
       if (inputType === 'note') {
-        summary = await handleNoteQuickAdd(rawInput, task);
+        summary = await handleNoteQuickAdd(rawInput, task, description);
       } else if (inputType === 'chat') {
         summary = await handleChatTranscriptQuickAdd(rawInput, task);
       } else {

src/services/database/descriptionService.ts

@@ -0,0 +1,88 @@
+import { openai as openaiProvider } from '@ai-sdk/openai';
+import { generateText } from 'ai';
+
+export interface DescriptionInput {
+  title: string;
+  content?: string;
+  metadata?: {
+    source?: string;
+    channel_name?: string;
+    author?: string;
+    site_name?: string;
+  };
+  type?: string;
+}
+
+/**
+ * Generate a 280-character description for a knowledge node.
+ * The description starts with "This is a..." and identifies the content type.
+ */
+export async function generateDescription(input: DescriptionInput): Promise<string> {
+  try {
+    const prompt = buildDescriptionPrompt(input);
+
+    console.log(`[DescriptionService] Generating description for: "${input.title}"`);
+
+    const response = await generateText({
+      model: openaiProvider('gpt-4o-mini'),
+      prompt,
+      maxOutputTokens: 100,
+      temperature: 0.3,
+    });
+
+    const description = response.text.trim();
+
+    // Ensure it starts with "This is a" and is within limit
+    const finalDescription = description.slice(0, 280);
+
+    console.log(`[DescriptionService] Generated: "${finalDescription}"`);
+
+    return finalDescription;
+  } catch (error) {
+    console.error('[DescriptionService] Error generating description:', error);
+    // Return a fallback description
+    return `This is a ${input.type || 'knowledge item'} titled "${input.title.slice(0, 200)}".`;
+  }
+}
+
+function buildDescriptionPrompt(input: DescriptionInput): string {
+  const metadataLines: string[] = [];
+
+  if (input.metadata?.source) {
+    metadataLines.push(`Source: ${input.metadata.source}`);
+  }
+  if (input.metadata?.channel_name) {
+    metadataLines.push(`Channel: ${input.metadata.channel_name}`);
+  }
+  if (input.metadata?.author) {
+    metadataLines.push(`Author: ${input.metadata.author}`);
+  }
+  if (input.metadata?.site_name) {
+    metadataLines.push(`Site: ${input.metadata.site_name}`);
+  }
+  if (input.type) {
+    metadataLines.push(`Type: ${input.type}`);
+  }
+
+  const contentPreview = input.content?.slice(0, 500) || '';
+
+  return `Generate a concise description (max 280 characters) for this knowledge item.
+
+CRITICAL REQUIREMENTS:
+- Start with "This is a..."
+- Identify the content type (article, video, paper, podcast episode, tweet, book, tutorial, etc.)
+- Be specific about what the content covers
+- Maximum 280 characters total
+
+=== KNOWLEDGE ITEM ===
+Title: ${input.title}
+${metadataLines.length > 0 ? metadataLines.join('\n') : ''}
+${contentPreview ? `\nContent preview: ${contentPreview}${input.content && input.content.length > 500 ? '...' : ''}` : ''}
+
+=== YOUR RESPONSE ===
+Write ONLY the description, nothing else. Start with "This is a..."`;
+}
+
+export const descriptionService = {
+  generateDescription
+};

src/services/database/dimensionService.ts

@@ -53,6 +53,7 @@ export class DimensionService {
     title: string;
     content?: string;
     link?: string;
+    description?: string;
   }): Promise<{ locked: string[]; keywords: string[] }> {
     try {
       const lockedDimensions = await this.getLockedDimensions();
@@ -69,7 +70,7 @@ export class DimensionService {
       const response = await generateText({
         model: openaiProvider('gpt-4o-mini'),
         prompt,
-        maxOutputTokens: 150, // Increased for two-part response
+        maxOutputTokens: 300, // Increased to accommodate more dimensions
         temperature: 0.1,
       });
 
@@ -145,16 +146,29 @@ export class DimensionService {
    * Build AI prompt for dimension assignment (locked + keyword dimensions)
    */
   private static buildAssignmentPrompt(
-    nodeData: { title: string; content?: string; link?: string },
+    nodeData: { title: string; content?: string; link?: string; description?: string },
     lockedDimensions: LockedDimension[]
   ): string {
-    const contentPreview = nodeData.content?.slice(0, 1000) || '';
+    // Use description as primary context, content as fallback
+    let nodeContextSection: string;
+    if (nodeData.description) {
+      const contentPreview = nodeData.content?.slice(0, 500) || '';
+      nodeContextSection = `DESCRIPTION: ${nodeData.description}
 
-    // Only include dimensions that have descriptions
-    const dimensionsWithDescriptions = lockedDimensions.filter(d => d.description && d.description.trim().length > 0);
+CONTENT PREVIEW: ${contentPreview}${nodeData.content && nodeData.content.length > 500 ? '...' : ''}`;
+    } else {
+      const contentPreview = nodeData.content?.slice(0, 2000) || '';
+      nodeContextSection = `CONTENT: ${contentPreview}${nodeData.content && nodeData.content.length > 2000 ? '...' : ''}`;
+    }
 
-    const dimensionsList = dimensionsWithDescriptions
-      .map(d => `DIMENSION: "${d.name}"\nDESCRIPTION: ${d.description}`)
+    // Include ALL locked dimensions, using fallback text for those without descriptions
+    const dimensionsList = lockedDimensions
+      .map(d => {
+        const description = d.description && d.description.trim().length > 0
+          ? d.description
+          : '(none - infer from name)';
+        return `DIMENSION: "${d.name}"\nDESCRIPTION: ${description}`;
+      })
       .join('\n---\n');
 
     return `You are categorizing a knowledge node. You will:
@@ -163,7 +177,7 @@ export class DimensionService {
 
 === NODE TO CATEGORIZE ===
 Title: ${nodeData.title}
-Content: ${contentPreview}${nodeData.content && nodeData.content.length > 1000 ? '...' : ''}
+${nodeContextSection}
 URL: ${nodeData.link || 'none'}
 
 === PART 1: LOCKED DIMENSIONS ===
@@ -171,7 +185,6 @@ CRITICAL: Read each dimension's DESCRIPTION carefully.
 The description defines what belongs in that dimension.
 Only assign if the content CLEARLY matches the description.
 If unsure, skip it — better to miss than assign incorrectly.
-Maximum 3 locked dimensions.
 
 AVAILABLE LOCKED DIMENSIONS:
 ${dimensionsList}
@@ -224,11 +237,6 @@ KEYWORDS:
 
         if (matchedDimension && !lockedDimensions.includes(matchedDimension.name)) {
           lockedDimensions.push(matchedDimension.name);
-
-          // Limit to 3 locked dimensions
-          if (lockedDimensions.length >= 3) {
-            break;
-          }
         }
       }
     }

src/services/database/nodes.ts

@@ -36,25 +36,27 @@ export class NodeService {
       params.push(...dimensions);
     }
 
-    // Text search in title and content (SQLite LIKE with COLLATE NOCASE)
+    // Text search in title, description, and content (SQLite LIKE with COLLATE NOCASE)
     if (search) {
-      query += ` AND (n.title LIKE ? COLLATE NOCASE OR n.content LIKE ? COLLATE NOCASE)`;
-      params.push(`%${search}%`, `%${search}%`);
+      query += ` AND (n.title LIKE ? COLLATE NOCASE OR n.description LIKE ? COLLATE NOCASE OR n.content LIKE ? COLLATE NOCASE)`;
+      params.push(`%${search}%`, `%${search}%`, `%${search}%`);
     }
 
     // Sorting logic
     if (search) {
-      // For search queries, prioritize by relevance: exact title → starts with → contains in title → content
-      query += ` ORDER BY 
-        CASE WHEN LOWER(n.title) = LOWER(?) THEN 1 ELSE 5 END,
-        CASE WHEN LOWER(n.title) LIKE LOWER(?) THEN 2 ELSE 5 END,
-        CASE WHEN n.title LIKE ? COLLATE NOCASE THEN 3 ELSE 5 END,
-        CASE WHEN n.content LIKE ? COLLATE NOCASE THEN 4 ELSE 5 END,
+      // For search queries, prioritize by relevance: exact title → starts with → contains in title → description → content
+      query += ` ORDER BY
+        CASE WHEN LOWER(n.title) = LOWER(?) THEN 1 ELSE 6 END,
+        CASE WHEN LOWER(n.title) LIKE LOWER(?) THEN 2 ELSE 6 END,
+        CASE WHEN n.title LIKE ? COLLATE NOCASE THEN 3 ELSE 6 END,
+        CASE WHEN n.description LIKE ? COLLATE NOCASE THEN 4 ELSE 6 END,
+        CASE WHEN n.content LIKE ? COLLATE NOCASE THEN 5 ELSE 6 END,
         n.updated_at DESC`;
       params.push(
         search,           // Exact match (case-insensitive)
         `${search}%`,     // Starts with search term
         `%${search}%`,    // Contains in title
+        `%${search}%`,    // Contains in description
         `%${search}%`     // Contains in content
       );
     } else if (sortBy === 'edges') {

src/services/typescript/embed-nodes.ts

@@ -17,6 +17,7 @@ interface NodeRecord {
   id: number;
   title: string;
   content: string | null;
+  description: string | null;
   dimensions_json: string;
   embedding?: Buffer | null;
   embedding_updated_at?: string | null;
@@ -106,7 +107,8 @@ Focus on the main concepts, key relationships, and practical implications.`;
     let embeddingText = formatEmbeddingText(
       node.title,
       node.content || '',
-      dimensions
+      dimensions,
+      node.description
     );
 
     // Add AI analysis if content exists
@@ -175,7 +177,7 @@ Focus on the main concepts, key relationships, and practical implications.`;
     if (nodeId) {
       // Single node
       query = `
-        SELECT n.id, n.title, n.content,
+        SELECT n.id, n.title, n.content, n.description,
                COALESCE((SELECT JSON_GROUP_ARRAY(d.dimension)
                         FROM node_dimensions d WHERE d.node_id = n.id), '[]') as dimensions_json,
                n.embedding, n.embedding_updated_at
@@ -186,7 +188,7 @@ Focus on the main concepts, key relationships, and practical implications.`;
     } else if (forceReEmbed) {
       // All nodes
       query = `
-        SELECT n.id, n.title, n.content,
+        SELECT n.id, n.title, n.content, n.description,
                COALESCE((SELECT JSON_GROUP_ARRAY(d.dimension)
                         FROM node_dimensions d WHERE d.node_id = n.id), '[]') as dimensions_json,
                n.embedding, n.embedding_updated_at
@@ -196,7 +198,7 @@ Focus on the main concepts, key relationships, and practical implications.`;
     } else {
       // Only nodes without embeddings
       query = `
-        SELECT n.id, n.title, n.content,
+        SELECT n.id, n.title, n.content, n.description,
                COALESCE((SELECT JSON_GROUP_ARRAY(d.dimension)
                         FROM node_dimensions d WHERE d.node_id = n.id), '[]') as dimensions_json,
                n.embedding, n.embedding_updated_at

src/services/typescript/sqlite-vec.ts

@@ -83,10 +83,12 @@ export function createDatabaseConnection(): Database.Database {
 export function formatEmbeddingText(
   title: string,
   content: string,
-  dimensions: string[]
+  dimensions: string[],
+  description?: string | null
 ): string {
+  const descriptionText = description && description.trim() ? description.trim() : 'none';
   const dimensionsText = dimensions.length > 0 ? dimensions.join(', ') : 'none';
-  return `Title: ${title}\n\nContent: ${content}\n\nDimensions: ${dimensionsText}`;
+  return `Title: ${title}\n\nDescription: ${descriptionText}\n\nContent: ${content}\n\nDimensions: ${dimensionsText}`;
 }
 
 /**