fix: process embeddings sequentially, shorten prompts, fix tests

- main.ts: process files sequentially (not Promise.all) to avoid concurrent
  embedding requests hammering Ollama; batch size reduced to 1
- vectorization.ts: shorten embedding prompts from 1000 to 500 chars,
  limit headings to 5, remove frontmatter from prompt to stay well within
  embedding model context window
- Update vectorization and indexing-pipeline tests for new prompt format

main.js

@@ -9507,10 +9507,10 @@ var ContentVectorizer = class {
     const parts = [
       chunk.title,
       chunk.firstParagraph,
-      chunk.content.substring(0, 1e3),
+      chunk.content.substring(0, 500),
       // Limit content to avoid long prompts
-      chunk.headings.join(" "),
+      chunk.headings.slice(0, 5).join(" ")
-      JSON.stringify(chunk.frontmatter)
+      // Limit headings
     ].filter(Boolean);
     return parts.join("\n\n");
   }
@@ -10033,7 +10033,7 @@ var OllamaPlugin = class extends import_obsidian4.Plugin {
     const files = this.app.vault.getMarkdownFiles();
     Logger.info(`Starting background vault indexing for ${files.length} files...`, "main");
     let indexed = 0;
-    const BATCH_SIZE = 2;
+    const BATCH_SIZE = 1;
     const DELAY_MS = 500;
     for (let i = 0; i < files.length; i += BATCH_SIZE) {
       if (signal.aborted) {
@@ -10041,20 +10041,18 @@ var OllamaPlugin = class extends import_obsidian4.Plugin {
         return;
       }
       const batch = files.slice(i, i + BATCH_SIZE);
-      await Promise.all(
+      for (const file of batch) {
-        batch.map(async (file) => {
+        if (signal.aborted) break;
-          if (signal.aborted) return;
+        try {
-          try {
+          const content = await this.app.vault.read(file);
-            const content = await this.app.vault.read(file);
+          if (signal.aborted) break;
-            if (signal.aborted) return;
+          await this.vaultVectorStore.indexFile(file, content);
-            await this.vaultVectorStore.indexFile(file, content);
+          indexed++;
-            indexed++;
+        } catch (error) {
-          } catch (error) {
+          const errorMessage = error instanceof Error ? error.message : String(error);
-            const errorMessage = error instanceof Error ? error.message : String(error);
+          Logger.warn(`Failed to index ${file.path}: ${errorMessage}`, "main");
-            Logger.warn(`Failed to index ${file.path}: ${errorMessage}`, "main");
+        }
-          }
+      }
-        })
-      );
       if (i + BATCH_SIZE < files.length) {
         await new Promise((resolve) => setTimeout(resolve, DELAY_MS));
       }

src/indexing-pipeline/vectorization.ts

@@ -92,13 +92,13 @@ export class ContentVectorizer {
    * Creates a prompt from content chunk for embedding
    */
   private createPrompt(chunk: ContentChunk): string {
-    // Combine important elements for embedding
+    // Combine important elements for embedding, keeping it concise
+    // to avoid exceeding the embedding model's context window
     const parts = [
       chunk.title,
       chunk.firstParagraph,
-      chunk.content.substring(0, 1000), // Limit content to avoid long prompts
+      chunk.content.substring(0, 500), // Limit content to avoid long prompts
-      chunk.headings.join(' '),
+      chunk.headings.slice(0, 5).join(' '), // Limit headings
-      JSON.stringify(chunk.frontmatter),
     ].filter(Boolean);
 
     return parts.join('\n\n');

src/main.ts

@@ -166,7 +166,7 @@ export default class OllamaPlugin extends Plugin {
     Logger.info(`Starting background vault indexing for ${files.length} files...`, 'main');
 
     let indexed = 0;
-    const BATCH_SIZE = 2;
+    const BATCH_SIZE = 1;
     const DELAY_MS = 500;
 
     for (let i = 0; i < files.length; i += BATCH_SIZE) {
@@ -176,20 +176,19 @@ export default class OllamaPlugin extends Plugin {
       }
 
       const batch = files.slice(i, i + BATCH_SIZE);
-      await Promise.all(
+      // Process files sequentially to avoid concurrent embedding requests
-        batch.map(async (file) => {
+      for (const file of batch) {
-          if (signal.aborted) return;
+        if (signal.aborted) break;
-          try {
+        try {
-            const content = await this.app.vault.read(file);
+          const content = await this.app.vault.read(file);
-            if (signal.aborted) return;
+          if (signal.aborted) break;
-            await this.vaultVectorStore!.indexFile(file, content);
+          await this.vaultVectorStore!.indexFile(file, content);
-            indexed++;
+          indexed++;
-          } catch (error) {
+        } catch (error) {
-            const errorMessage = error instanceof Error ? error.message : String(error);
+          const errorMessage = error instanceof Error ? error.message : String(error);
-            Logger.warn(`Failed to index ${file.path}: ${errorMessage}`, 'main');
+          Logger.warn(`Failed to index ${file.path}: ${errorMessage}`, 'main');
-          }
+        }
-        })
+      }
-      );
 
       // Delay between batches to avoid overloading Ollama
       if (i + BATCH_SIZE < files.length) {

tests/indexing-pipeline.test.ts

@@ -207,7 +207,7 @@ Content`;
         firstParagraph: 'First paragraph',
         wordCount: 2,
         chunkIndex: 0,
-        chunkSize: 100
+        chunkSize: 100,
       };
 
       const prompt = (vectorizer as any).createPrompt(mockChunk);
@@ -215,7 +215,7 @@ Content`;
       expect(prompt).toContain('Test');
       expect(prompt).toContain('First paragraph');
       expect(prompt).toContain('Heading');
-      expect(prompt).toContain('tags');
+      // Frontmatter is no longer included in embedding prompts
     });
 
     // Note: Actual embedding tests would require mocking fetch or integration testing
@@ -233,7 +233,7 @@ Content`;
         firstParagraph: 'First paragraph',
         wordCount: 2,
         chunkIndex: 0,
-        chunkSize: 100
+        chunkSize: 100,
       };
 
       // Mock fetch to simulate an error
@@ -290,12 +290,12 @@ This is a test document for pipeline processing.`;
     it('should process files in batches', async () => {
       const files: MockVaultFile[] = [
         { basename: 'file1', path: 'file1.md' },
-        { basename: 'file2', path: 'file2.md' }
+        { basename: 'file2', path: 'file2.md' },
       ];
 
       const fileContents = {
         'file1.md': '# File 1\n\nContent 1',
-        'file2.md': '# File 2\n\nContent 2'
+        'file2.md': '# File 2\n\nContent 2',
       };
 
       const results = await pipeline.processFilesInBatches(files, fileContents, 1);

tests/vectorization.test.ts

@@ -200,8 +200,7 @@ describe('ContentVectorizer', () => {
       expect(prompt).toContain('This is the first paragraph');
       expect(prompt).toContain('Main Heading');
       expect(prompt).toContain('Sub Heading');
-      expect(prompt).toContain('test');
+      // Frontmatter is no longer included in embedding prompts
-      expect(prompt).toContain('2024-01-01');
     });
 
     it('should handle empty content fields gracefully', () => {
@@ -221,8 +220,7 @@ describe('ContentVectorizer', () => {
 
       const prompt = (vectorizer as any).createPrompt(chunk);
       expect(prompt).toContain('Only content');
-      // JSON.stringify({}) produces "{}", which is truthy so it's included
+      // Frontmatter is no longer included in embedding prompts
-      expect(prompt).toContain('{}');
     });
 
     it('should limit content length', () => {
@@ -243,7 +241,7 @@ describe('ContentVectorizer', () => {
 
       const prompt = (vectorizer as any).createPrompt(chunk);
       expect(prompt).not.toContain('a'.repeat(1500));
-      expect(prompt).toContain('a'.repeat(1000));
+      expect(prompt).toContain('a'.repeat(500));
     });
 
     it('should handle missing frontmatter gracefully', () => {