Revert "fix: switch embedding dimension from 1536 to 2560 for qwen-embedding-4b"

This reverts commit e09ea3a145.

mcp-server/prisma/schema.prisma

@@ -131,7 +131,7 @@ model Note {
 model NoteEmbedding {
   id        String                   @id @default(cuid())
   noteId    String                   @unique
-  embedding Unsupported("vector(2560)")
+  embedding Unsupported("vector(1536)")
   createdAt DateTime                 @default(now())
   updatedAt DateTime                 @default(now()) @updatedAt
   note      Note                     @relation(fields: [noteId], references: [id], onDelete: Cascade)

memento-note/app/api/admin/embeddings/validate/route.ts

@@ -35,7 +35,7 @@ export async function GET() {
     const invalidResult: Array<{ count: bigint }> = await prisma.$queryRawUnsafe(
       `SELECT COUNT(*)::bigint as count FROM "NoteEmbedding" e
        WHERE e."embedding" IS NULL
-          OR array_length(string_to_array(replace(replace(e."embedding"::text, '[', ''), ']', ''), ','), 1) != 2560`
+          OR array_length(string_to_array(replace(replace(e."embedding"::text, '[', ''), ']', ''), ','), 1) != 1536`
     )
     const invalidCount = Number(invalidResult[0]?.count ?? 0)
 

memento-note/lib/ai/providers/custom-openai.ts

@@ -1,13 +1,13 @@
 import { createOpenAI } from '@ai-sdk/openai';
-import { generateObject, generateText as aiGenerateText, stepCountIs } from 'ai';
+import { generateObject, generateText as aiGenerateText, embed, stepCountIs } from 'ai';
 import { z } from 'zod';
 import { AIProvider, TagSuggestion, TitleSuggestion, ToolUseOptions, ToolCallResult } from '../types';
 
 export class CustomOpenAIProvider implements AIProvider {
   private model: any;
+  private embeddingModel: any;
   private apiKey: string;
   private baseUrl: string;
-  private embeddingModelName: string;
 
   constructor(
     apiKey: string,
@@ -17,7 +17,6 @@ export class CustomOpenAIProvider implements AIProvider {
   ) {
     this.apiKey = apiKey;
     this.baseUrl = baseUrl.endsWith('/') ? baseUrl.slice(0, -1) : baseUrl;
-    this.embeddingModelName = embeddingModelName;
     // Create OpenAI-compatible client with custom base URL
     // Use .chat() to force /chat/completions endpoint (avoids Responses API)
     const customClient = createOpenAI({
@@ -45,16 +44,7 @@ export class CustomOpenAIProvider implements AIProvider {
     });
 
     this.model = customClient.chat(modelName);
-  }
+    this.embeddingModel = customClient.embedding(embeddingModelName);
-
-  private async fetchWithTimeout(url: string, options: RequestInit, timeoutMs: number = 60_000): Promise<Response> {
-    const controller = new AbortController()
-    const timer = setTimeout(() => controller.abort(), timeoutMs)
-    try {
-      return await fetch(url, { ...options, signal: controller.signal })
-    } finally {
-      clearTimeout(timer)
-    }
   }
 
   async generateTags(content: string): Promise<TagSuggestion[]> {
@@ -80,40 +70,13 @@ export class CustomOpenAIProvider implements AIProvider {
 
   async getEmbeddings(text: string): Promise<number[]> {
     try {
-      const response = await this.fetchWithTimeout(`${this.baseUrl}/embeddings`, {
+      const { embedding } = await embed({
-        method: 'POST',
+        model: this.embeddingModel,
-        headers: {
+        value: text,
-          'Content-Type': 'application/json',
-          'Authorization': `Bearer ${this.apiKey}`,
-          'HTTP-Referer': 'https://localhost:3000',
-          'X-Title': 'Memento AI',
-        },
-        body: JSON.stringify({
-          model: this.embeddingModelName,
-          input: text,
-        }),
       });
-
+      return embedding;
-      if (!response.ok) {
-        const errText = await response.text();
-        throw new Error(`${this.baseUrl}/embeddings error ${response.status}: ${errText}`);
-      }
-
-      const data = await response.json();
-
-      // Standard OpenAI-compatible response: { data: [{ embedding: number[] }] }
-      if (data.data && Array.isArray(data.data) && data.data[0]?.embedding) {
-        return data.data[0].embedding;
-      }
-
-      // Fallback: some providers return { embedding: number[] }
-      if (data.embedding && Array.isArray(data.embedding)) {
-        return data.embedding;
-      }
-
-      throw new Error(`Unexpected embeddings response shape: ${JSON.stringify(data)}`);
     } catch (e) {
-      console.error('Error generating embeddings (CustomOpenAI):', e);
+      console.error('Error generating embeddings (Custom OpenAI):', e);
       throw e;
     }
   }

memento-note/lib/ai/providers/openrouter.ts

@@ -1,35 +1,21 @@
 import { createOpenAI } from '@ai-sdk/openai';
-import { generateObject, generateText as aiGenerateText, stepCountIs } from 'ai';
+import { generateObject, generateText as aiGenerateText, embed, stepCountIs } from 'ai';
 import { z } from 'zod';
 import { AIProvider, TagSuggestion, TitleSuggestion, ToolUseOptions, ToolCallResult } from '../types';
 
 export class OpenRouterProvider implements AIProvider {
   private model: any;
-  private apiKey: string;
+  private embeddingModel: any;
-  private baseUrl: string;
-  private embeddingModelName: string;
 
   constructor(apiKey: string, modelName: string = 'anthropic/claude-3-haiku', embeddingModelName: string = 'openai/text-embedding-3-small') {
-    this.apiKey = apiKey;
-    this.baseUrl = 'https://openrouter.ai/api/v1';
-    this.embeddingModelName = embeddingModelName;
     // Create OpenAI-compatible client for OpenRouter
     const openrouter = createOpenAI({
-      baseURL: this.baseUrl,
+      baseURL: 'https://openrouter.ai/api/v1',
       apiKey: apiKey,
     });
 
     this.model = openrouter.chat(modelName);
-  }
+    this.embeddingModel = openrouter.embedding(embeddingModelName);
-
-  private async fetchWithTimeout(url: string, options: RequestInit, timeoutMs: number = 60_000): Promise<Response> {
-    const controller = new AbortController()
-    const timer = setTimeout(() => controller.abort(), timeoutMs)
-    try {
-      return await fetch(url, { ...options, signal: controller.signal })
-    } finally {
-      clearTimeout(timer)
-    }
   }
 
   async generateTags(content: string): Promise<TagSuggestion[]> {
@@ -55,38 +41,11 @@ export class OpenRouterProvider implements AIProvider {
 
   async getEmbeddings(text: string): Promise<number[]> {
     try {
-      const response = await this.fetchWithTimeout(`${this.baseUrl}/embeddings`, {
+      const { embedding } = await embed({
-        method: 'POST',
+        model: this.embeddingModel,
-        headers: {
+        value: text,
-          'Content-Type': 'application/json',
-          'Authorization': `Bearer ${this.apiKey}`,
-          'HTTP-Referer': 'https://localhost:3000',
-          'X-Title': 'Memento AI',
-        },
-        body: JSON.stringify({
-          model: this.embeddingModelName,
-          input: text,
-        }),
       });
-
+      return embedding;
-      if (!response.ok) {
-        const errText = await response.text();
-        throw new Error(`OpenRouter embeddings error ${response.status}: ${errText}`);
-      }
-
-      const data = await response.json();
-
-      // OpenRouter returns { data: [{ embedding: number[] }] }
-      if (data.data && Array.isArray(data.data) && data.data[0]?.embedding) {
-        return data.data[0].embedding;
-      }
-
-      // Fallback: some OpenAI-compatible providers return { embedding: number[] }
-      if (data.embedding && Array.isArray(data.embedding)) {
-        return data.embedding;
-      }
-
-      throw new Error(`Unexpected OpenRouter embeddings response shape: ${JSON.stringify(data)}`);
     } catch (e) {
       console.error('Error generating embeddings (OpenRouter):', e);
       throw e;

memento-note/lib/ai/services/embedding.service.ts

@@ -1,7 +1,7 @@
 /**
  * Embedding Service
  * Generates vector embeddings for semantic search and similarity analysis.
- * Stores embeddings as native pgvector(2560) in PostgreSQL.
+ * Stores embeddings as native pgvector(1536) in PostgreSQL.
  */
 
 import { getAIProvider } from '../factory'
@@ -14,7 +14,7 @@ export interface EmbeddingResult {
 }
 
 export class EmbeddingService {
-  private readonly EMBEDDING_DIMENSION = 2560
+  private readonly EMBEDDING_DIMENSION = 1536
 
   async generateEmbedding(text: string): Promise<EmbeddingResult> {
     if (!text || text.trim().length === 0) {

memento-note/prisma/migrations/20260512120000_pgvector_and_fts_search/migration.sql

@@ -1,7 +1,7 @@
 -- Phase 1: Enable pgvector extension
 CREATE EXTENSION IF NOT EXISTS vector;
 
--- Phase 2: Convert embedding column from text to vector(2560)
+-- Phase 2: Convert embedding column from text to vector(1536)
 -- Idempotent: detects current column type and only converts when needed.
 -- Handles all partial states from previous failed migration attempts:
 --   A) embedding is text → direct ALTER COLUMN TYPE conversion
@@ -25,8 +25,8 @@ BEGIN
     IF _emb_type IS NOT NULL THEN
         ALTER TABLE "NoteEmbedding" DROP COLUMN IF EXISTS "_vec_tmp";
         ALTER TABLE "NoteEmbedding"
-            ALTER COLUMN "embedding" TYPE vector(2560)
+            ALTER COLUMN "embedding" TYPE vector(1536)
-            USING "embedding"::vector(2560);
+            USING "embedding"::vector(1536);
         RETURN;
     END IF;
 

memento-note/prisma/schema.prisma

@@ -300,7 +300,7 @@ model UserAISettings {
 model NoteEmbedding {
   id        String   @id @default(cuid())
   noteId    String   @unique
-  embedding Unsupported("vector(2560)")
+  embedding Unsupported("vector(1536)")
   createdAt DateTime @default(now())
   updatedAt DateTime @default(now()) @updatedAt
   note      Note     @relation(fields: [noteId], references: [id], onDelete: Cascade)

memento-note/scripts/migrate-embeddings.ts

@@ -41,7 +41,7 @@ async function main() {
       // Embedding will be generated by the indexNote method which handles pgvector format
       await prisma.$executeRawUnsafe(
         `INSERT INTO "NoteEmbedding" ("id", "noteId", "embedding", "createdAt", "updatedAt")
-         VALUES (gen_random_uuid(), $1, '[0]'::vector(2560), now(), now())
+         VALUES (gen_random_uuid(), $1, '[0]'::vector(1536), now(), now())
          ON CONFLICT ("noteId") DO NOTHING`,
         note.id
       )

memento-note/tests/ai-provider.spec.ts

@@ -47,7 +47,7 @@ test.describe('AI Provider Configuration Tests', () => {
 
     // Verify embeddings provider is working
     expect(result.embeddingsProvider).toBe('openai');
-    expect(result.embeddingLength).toBe(2560); // OpenAI text-embedding-3-small
+    expect(result.embeddingLength).toBe(1536); // OpenAI text-embedding-3-small
     expect(result.details?.provider).toBe('OpenAI');
   });
 

memento-note/tests/migration/integrity.test.ts

@@ -232,7 +232,7 @@ describe('Data Integrity Tests', () => {
       const vecStr = '[0.1,0.2,0.3,0.4,0.5]'
       await prisma.$executeRawUnsafe(
         `INSERT INTO "NoteEmbedding" ("id", "noteId", "embedding", "createdAt", "updatedAt")
-         VALUES (gen_random_uuid(), $1, $2::vector(2560), now(), now())`,
+         VALUES (gen_random_uuid(), $1, $2::vector(1536), now(), now())`,
         note.id,
         vecStr
       )