feat: early vector search testing

src/plugins/built-in/globalSearch/SearchBar.svelte

@@ -6,7 +6,7 @@
   import { type StaticCommandItem } from './commands';
   import type { CombinedResult } from './types';
   import { createSearchIndexes, performSearch as doSearch } from './searchUtils';
-  import { highlightMatch, highlightSnippet } from './highlightUtils';
+  import { highlightMatch, highlightSnippet, stripHtmlButKeepHighlights } from './highlightUtils';
   import Fuse from 'fuse.js';
   import Calculator from './Calculator.svelte';
   import { actionMap } from './indexing/actions';
@@ -105,14 +105,14 @@
     };
   });
 
-  const performSearch = () => {
+  const performSearch = async () => {
     isLoading = true;
     selectedIndex = 0;
 
     const term = searchTerm.trim().toLowerCase();
     
     if (commandsFuse && dynamicContentFuse) {
-      combinedResults = doSearch(
+      combinedResults = await doSearch(
         term, 
         commandsFuse, 
         dynamicContentFuse, 
@@ -288,8 +288,9 @@
                       onclick={() => executeItemAction(dynamicItem)}
                     >
                       <div class="flex items-center w-full">
+                        <div class="flex-none w-8 h-8 text-xl font-IconFamily flex items-center justify-center {isSelected ? 'text-zinc-900 dark:text-white' : 'text-zinc-600 dark:text-zinc-400'}">{dynamicItem.metadata?.icon || '\ue924'}</div>
                         <span class="ml-4 text-lg truncate">
-                          {@html highlightMatch(dynamicItem.text, searchTerm, result.matches)}
+                          {@html stripHtmlButKeepHighlights(highlightMatch(dynamicItem.text, searchTerm, result.matches))}
                         </span>
                         <span class="flex-none ml-auto text-xs text-zinc-500 dark:text-zinc-400">
                           {dynamicItem.category} 
@@ -297,7 +298,7 @@
                       </div>
                       {#if dynamicItem.content}
                         <div class="mt-1 ml-12 text-sm text-zinc-600 dark:text-zinc-400 line-clamp-2 text-start">
-                          {@html highlightSnippet(dynamicItem.content, searchTerm, result.matches)}
+                          {@html stripHtmlButKeepHighlights(highlightSnippet(dynamicItem.content, searchTerm, result.matches))}
                         </div>
                       {/if}
                     </button>

src/plugins/built-in/globalSearch/client-vector-search-docs.md

@@ -0,0 +1,597 @@
+# client-vector-search
+
+A client side vector search library that can embed, search, and cache. Works on the browser and server side.
+
+It outperforms OpenAI's text-embedding-ada-002 and is way faster than Pinecone and other VectorDBs.
+
+I'm the founder of [searchbase.app](https://searchbase.app) and we needed this for our product and customers. We'll be using this library in production. You can be sure it'll be maintained and improved.
+
+- Embed documents using transformers by default: gte-small (~30mb).
+- Calculate cosine similarity between embeddings.
+- Create an index and search on the client side
+- Cache vectors with browser caching support.
+
+Lots of improvements are coming!
+
+## Roadmap
+
+Our goal is to build a super simple, fast vector search that works with couple hundred to thousands vectors. ~1k vectors per user covers 99% of the use cases.
+
+We'll initially keep things super simple and sub 100ms
+
+### TODOs
+- [ ] add HNSW index that works on node and browser env, don't rely on hnsw binder libs
+- [ ] add a proper testing suite and ci/cd for the lib
+  - [ ] simple health tests
+    - [ ] mock the @xenova/transformers for jest, it's not happy with it
+  - [ ] performance tests, recall, memory usage, cpu usage etc.
+
+
+## Installation
+
+```bash
+npm i client-vector-search
+```
+
+
+## Quickstart
+
+This library provides a plug-and-play solution for embedding and vector search. It's designed to be easy to use, efficient, and versatile. Here's a quick start guide:
+
+
+```ts
+  import { getEmbedding, EmbeddingIndex } from 'client-vector-search';
+
+  // getEmbedding is an async function, so you need to use 'await' or '.then()' to get the result
+  const embedding = await getEmbedding("Apple"); // Returns embedding as number[]
+
+  // Each object should have an 'embedding' property of type number[]
+  const initialObjects = [
+  { id: 1, name: "Apple", embedding: embedding },
+  { id: 2, name: "Banana", embedding: await getEmbedding("Banana") },
+  { id: 3, name: "Cheddar", embedding: await getEmbedding("Cheddar")},
+  { id: 4, name: "Space", embedding: await getEmbedding("Space")},
+  { id: 5, name: "database", embedding: await getEmbedding("database")},
+  ];
+  const index = new EmbeddingIndex(initialObjects); // Creates an index
+
+  // The query should be an embedding of type number[]
+  const queryEmbedding = await getEmbedding('Fruit'); // Query embedding
+  const results = await index.search(queryEmbedding, { topK: 5 }); // Returns top similar objects
+
+  // specify the storage type
+  await index.saveIndex('indexedDB');
+  const results = await index.search([1, 2, 3], {
+    topK: 5,
+    useStorage: 'indexedDB',
+    // storageOptions: { // use only if you overrode the defaults
+    //   indexedDBName: 'clientVectorDB',
+    //   indexedDBObjectStoreName: 'ClientEmbeddingStore',
+    // },
+  });
+
+  console.log(results);
+
+  await index.deleteIndexedDB(); // if you overrode default, specify db name
+```
+
+## Trouble-shooting
+
+### NextJS
+To use it inside NextJS projects you'll need to update the `next.config.js` file to include the following:
+
+```js
+module.exports = {
+  // Override the default webpack configuration
+  webpack: (config) => {
+    // See https://webpack.js.org/configuration/resolve/#resolvealias
+    config.resolve.alias = {
+      ...config.resolve.alias,
+      sharp$: false,
+      "onnxruntime-node$": false,
+    };
+    return config;
+  },
+};
+```
+
+#### Model load after page is loaded
+
+You can initialize the model before using it to generate embeddings. This will ensure that the model is loaded before you use it and provide a better UX.
+
+```js
+import { initializeModel } from "client-vector-search"
+...
+  useEffect(() => {
+    try {
+      initializeModel();
+    } catch (e) {
+      console.log(e);
+    }
+  }, []);
+```
+
+## Usage Guide
+
+This guide provides a step-by-step walkthrough of the library's main features. It covers everything from generating embeddings for a string to performing operations on the index such as adding, updating, and removing objects. It also includes instructions on how to save the index to a database and perform search operations within it.
+
+Until we have a reference documentation, you can find all the methods and their usage in this guide. Each step is accompanied by a code snippet to illustrate the usage of the method in question. Make sure to follow along and try out the examples in your own environment to get a better understanding of how everything works.
+
+Let's get started!
+
+### Step 1: Generate Embeddings for String
+Generate embeddings for a given string using the `getEmbedding` method.
+
+```ts
+const embedding = await getEmbedding("Apple"); // Returns embedding as number[]
+```
+> **Note**: `getEmbedding` is asynchronous; make sure to use `await`.
+
+---
+
+### Step 2: Calculate Cosine Similarity
+Calculate the cosine similarity between two embeddings.
+
+```ts
+const similarity = cosineSimilarity(embedding1, embedding2, 6);
+```
+> **Note**: Both embeddings should be of the same length.
+
+---
+
+### Step 3: Create an Index
+Create an index with an initial array of objects. Each object must have an 'embedding' property.
+
+```ts
+const initialObjects = [...];
+const index = new EmbeddingIndex(initialObjects);
+```
+
+---
+
+### Step 4: Add to Index
+Add an object to the index.
+
+```ts
+const objectToAdd = { id: 6, name: 'Cat', embedding: await getEmbedding('Cat') };
+index.add(objectToAdd);
+```
+
+---
+
+### Step 5: Update Index
+Update an existing object in the index.
+
+```ts
+const vectorToUpdate = { id: 6, name: 'Dog', embedding: await getEmbedding('Dog') };
+index.update({ id: 6 }, vectorToUpdate);
+```
+
+---
+
+### Step 6: Remove from Index
+Remove an object from the index.
+
+```ts
+index.remove({ id: 6 });
+```
+
+---
+
+### Step 7: Retrieve from Index
+Retrieve an object from the index.
+
+```ts
+const vector = index.get({ id: 1 });
+```
+
+---
+
+### Step 8: Search the Index
+Search the index with a query embedding.
+
+```ts
+const queryEmbedding = await getEmbedding('Fruit');
+const results = await index.search(queryEmbedding, { topK: 5 });
+```
+
+---
+
+### Step 9: Print the Index
+Print the entire index to the console.
+
+```ts
+index.printIndex();
+```
+
+---
+
+### Step 10: Save Index to IndexedDB (for browser)
+Save the index to a persistent IndexedDB database. Note
+
+```ts
+await index.saveIndex("indexedDB", { DBName: "clientVectorDB", objectStoreName:"ClientEmbeddingStore"})
+```
+
+---
+
+### Important: Search in indexedDB
+Perform a search operation in the IndexedDB.
+
+```ts
+const results = await index.search(queryEmbedding, {
+  topK: 5,
+  useStorage: "indexedDB",
+  storageOptions: { // only if you want to override the default options, defaults are below
+    indexedDBName: 'clientVectorDB',
+    indexedDBObjectStoreName: 'ClientEmbeddingStore'
+  }
+});
+
+---
+
+### Delete Database
+To delete an entire database.
+
+```ts
+await IndexedDbManager.deleteIndexedDB("clientVectorDB");
+```
+
+---
+
+### Delete Object Store
+To delete an object store from a database.
+
+```ts
+await IndexedDbManager.deleteIndexedDBObjectStore("clientVectorDB", "ClientEmbeddingStore");
+```
+
+---
+
+### Retrieve All Objects
+To retrieve all objects from a specific object store.
+
+```ts
+const allObjects = await IndexedDbManager.getAllObjectsFromIndexedDB("clientVectorDB", "ClientEmbeddingStore");
+```
+
+
+
+
+# THE MAIN INDEX.TS FILE THAT YOU ARE IMPORTING FROM
+```index.ts
+const DEFAULT_TOP_K = 3;
+
+interface Filter {
+  [key: string]: any;
+}
+
+import Cache from './cache';
+import { IndexedDbManager } from './indexedDB';
+import { cosineSimilarity } from './utils';
+export { ExperimentalHNSWIndex } from './hnsw';
+
+// uncomment if you want to test indexedDB implementation in node env for faster dev cycle
+// import { IDBFactory } from 'fake-indexeddb';
+// const indexedDB = new IDBFactory();
+
+export interface SearchResult {
+  similarity: number;
+  object: any;
+}
+
+type StorageOptions = 'indexedDB' | 'localStorage' | 'none';
+
+/**
+ * Interface for search options in the EmbeddingIndex class.
+ * topK: The number of top similar items to return.
+ * filter: An optional filter to apply to the objects before searching.
+ * useStorage: A flag to indicate whether to use storage options like indexedDB or localStorage.
+ */
+interface SearchOptions {
+  topK?: number;
+  filter?: Filter;
+  useStorage?: StorageOptions;
+  storageOptions?: { indexedDBName: string; indexedDBObjectStoreName: string }; // TODO: generalize it to localStorage as well
+}
+
+const cacheInstance = Cache.getInstance();
+
+let pipe: any;
+let currentModel: string;
+
+export const initializeModel = async (
+  model: string = 'Xenova/gte-small',
+): Promise<void> => {
+  if (model !== currentModel) {
+    const transformersModule = await import('@xenova/transformers');
+    const pipeline = transformersModule.pipeline;
+    pipe = await pipeline('feature-extraction', model);
+    currentModel = model;
+  }
+};
+
+export const getEmbedding = async (
+  text: string,
+  precision: number = 7,
+  options = { pooling: 'mean', normalize: false },
+  model = 'Xenova/gte-small',
+): Promise<number[]> => {
+  const cachedEmbedding = cacheInstance.get(text);
+  if (cachedEmbedding) {
+    return Promise.resolve(cachedEmbedding);
+  }
+
+  if (model !== currentModel) {
+    await initializeModel(model);
+  }
+
+  const output = await pipe(text, options);
+  const roundedOutput = Array.from(output.data as number[]).map(
+    (value: number) => parseFloat(value.toFixed(precision)),
+  );
+  cacheInstance.set(text, roundedOutput);
+  return Array.from(roundedOutput);
+};
+
+export class EmbeddingIndex {
+  private objects: Filter[];
+  private keys: string[];
+
+  constructor(initialObjects?: Filter[]) {
+    // TODO: add support for options while creating index such as  {... indexedDB: true, ...}
+    this.objects = [];
+    this.keys = [];
+    if (initialObjects && initialObjects.length > 0) {
+      initialObjects.forEach((obj) => this.validateAndAdd(obj));
+      if (initialObjects[0]) {
+        this.keys = Object.keys(initialObjects[0]);
+      }
+    }
+  }
+
+  private findVectorIndex(filter: Filter): number {
+    return this.objects.findIndex((object) =>
+      Object.keys(filter).every((key) => object[key] === filter[key]),
+    );
+  }
+
+  private validateAndAdd(obj: Filter) {
+    if (!Array.isArray(obj.embedding) || obj.embedding.some(isNaN)) {
+      throw new Error(
+        'Object must have an embedding property of type number[]',
+      );
+    }
+    if (this.keys.length === 0) {
+      this.keys = Object.keys(obj);
+    } else if (!this.keys.every((key) => key in obj)) {
+      throw new Error(
+        'Object must have the same properties as the initial objects',
+      );
+    }
+    this.objects.push(obj);
+  }
+
+  add(obj: Filter) {
+    this.validateAndAdd(obj);
+  }
+
+  // Method to update an existing vector in the index
+  update(filter: Filter, vector: Filter) {
+    const index = this.findVectorIndex(filter);
+    if (index === -1) {
+      throw new Error('Vector not found');
+    }
+    if (vector.hasOwnProperty('embedding')) {
+      // Validate and add the new vector
+      this.validateAndAdd(vector);
+    }
+    // Replace the old vector with the new one
+    this.objects[index] = Object.assign(this.objects[index] as Filter, vector);
+  }
+
+  // Method to remove a vector from the index
+  remove(filter: Filter) {
+    const index = this.findVectorIndex(filter);
+    if (index === -1) {
+      throw new Error('Vector not found');
+    }
+    // Remove the vector from the index
+    this.objects.splice(index, 1);
+  }
+
+  // Method to remove multiple vectors from the index
+  removeBatch(filters: Filter[]) {
+    filters.forEach((filter) => {
+      const index = this.findVectorIndex(filter);
+      if (index !== -1) {
+        // Remove the vector from the index
+        this.objects.splice(index, 1);
+      }
+    });
+  }
+
+  // Method to retrieve a vector from the index
+  get(filter: Filter) {
+    const vector = this.objects[this.findVectorIndex(filter)];
+    return vector || null;
+  }
+
+  size(): number {
+    // Returns the size of the index
+    return this.objects.length;
+  }
+
+  clear() {
+    this.objects = [];
+  }
+
+  async search(
+    queryEmbedding: number[],
+    options: SearchOptions = {
+      topK: 3,
+      useStorage: 'none',
+      storageOptions: {
+        indexedDBName: 'clientVectorDB',
+        indexedDBObjectStoreName: 'ClientEmbeddingStore',
+      },
+    },
+  ): Promise<SearchResult[]> {
+    const topK = options.topK || DEFAULT_TOP_K;
+    const filter = options.filter || {};
+    const useStorage = options.useStorage || 'none';
+
+    if (useStorage === 'indexedDB') {
+      const DBname = options.storageOptions?.indexedDBName || 'clientVectorDB';
+      const objectStoreName =
+        options.storageOptions?.indexedDBObjectStoreName ||
+        'ClientEmbeddingStore';
+
+      if (typeof indexedDB === 'undefined') {
+        console.error('IndexedDB is not supported');
+        throw new Error('IndexedDB is not supported');
+      }
+      const results = await this.loadAndSearchFromIndexedDB(
+        DBname,
+        objectStoreName,
+        queryEmbedding,
+        topK,
+        filter,
+      );
+      return results;
+    } else {
+      // Compute similarities
+      const similarities = this.objects
+        .filter((object) =>
+          Object.keys(filter).every((key) => object[key] === filter[key]),
+        )
+        .map((obj) => ({
+          similarity: cosineSimilarity(queryEmbedding, obj.embedding),
+          object: obj,
+        }));
+
+      // Sort by similarity and return topK results
+      return similarities
+        .sort((a, b) => b.similarity - a.similarity)
+        .slice(0, topK);
+    }
+  }
+
+  printIndex() {
+    console.log('Index Content:');
+    this.objects.forEach((obj, idx) => {
+      console.log(`Item ${idx + 1}:`, obj);
+    });
+  }
+
+  async saveIndex(
+    storageType: string,
+    options: { DBName: string; objectStoreName: string } = {
+      DBName: 'clientVectorDB',
+      objectStoreName: 'ClientEmbeddingStore',
+    },
+  ) {
+    if (storageType === 'indexedDB') {
+      await this.saveToIndexedDB(options.DBName, options.objectStoreName);
+    } else {
+      throw new Error(
+        `Unsupported storage type: ${storageType} \n Supported storage types: "indexedDB"`,
+      );
+    }
+  }
+
+  async saveToIndexedDB(
+    DBname: string = 'clientVectorDB',
+    objectStoreName: string = 'ClientEmbeddingStore',
+  ): Promise<void> {
+    if (typeof indexedDB === 'undefined') {
+      console.error('IndexedDB is not defined');
+      throw new Error('IndexedDB is not supported');
+    }
+
+    if (!this.objects || this.objects.length === 0) {
+      throw new Error('Index is empty. Nothing to save');
+    }
+
+    try {
+      const db = await IndexedDbManager.create(DBname, objectStoreName);
+      await db.addToIndexedDB(this.objects);
+      console.log(
+        `Index saved to database '${DBname}' object store '${objectStoreName}'`,
+      );
+    } catch (error) {
+      console.error('Error saving index to database:', error);
+      throw new Error('Error saving index to database');
+    }
+  }
+
+  async loadAndSearchFromIndexedDB(
+    DBname: string = 'clientVectorDB',
+    objectStoreName: string = 'ClientEmbeddingStore',
+    queryEmbedding: number[],
+    topK: number,
+    filter: { [key: string]: any },
+  ): Promise<SearchResult[]> {
+    const db = await IndexedDbManager.create(DBname, objectStoreName);
+    const generator = db.dbGenerator();
+    const results: { similarity: number; object: any }[] = [];
+
+    for await (const record of generator) {
+      if (Object.keys(filter).every((key) => record[key] === filter[key])) {
+        const similarity = cosineSimilarity(queryEmbedding, record.embedding);
+        results.push({ similarity, object: record });
+      }
+    }
+    results.sort((a, b) => b.similarity - a.similarity);
+    return results.slice(0, topK);
+  }
+
+  async deleteIndexedDB(DBname: string = 'clientVectorDB'): Promise<void> {
+    if (typeof indexedDB === 'undefined') {
+      console.error('IndexedDB is not defined');
+      throw new Error('IndexedDB is not supported');
+    }
+    return new Promise((resolve, reject) => {
+      const request = indexedDB.deleteDatabase(DBname);
+
+      request.onsuccess = () => {
+        console.log(`Database '${DBname}' deleted`);
+        resolve();
+      };
+      request.onerror = (event) => {
+        console.error('Failed to delete database', event);
+        reject(new Error('Failed to delete database'));
+      };
+    });
+  }
+
+  async deleteIndexedDBObjectStore(
+    DBname: string = 'clientVectorDB',
+    objectStoreName: string = 'ClientEmbeddingStore',
+  ): Promise<void> {
+    const db = await IndexedDbManager.create(DBname, objectStoreName);
+
+    try {
+      await db.deleteIndexedDBObjectStoreFromDB(DBname, objectStoreName);
+      console.log(
+        `Object store '${objectStoreName}' deleted from database '${DBname}'`,
+      );
+    } catch (error) {
+      console.error('Error deleting object store:', error);
+      throw new Error('Error deleting object store');
+    }
+  }
+
+  async getAllObjectsFromIndexedDB(
+    DBname: string = 'clientVectorDB',
+    objectStoreName: string = 'ClientEmbeddingStore',
+  ): Promise<any[]> {
+    const db = await IndexedDbManager.create(DBname, objectStoreName);
+    const objects: any[] = [];
+    for await (const record of db.dbGenerator()) {
+      objects.push(record);
+    }
+    return objects;
+  }
+}
+```

src/plugins/built-in/globalSearch/commands.ts

@@ -28,7 +28,7 @@ const staticCommands: StaticCommandItem[] = [
       window.location.hash = '?page=/home';
       loadHomePage();
     },
-    priority: 10
+    priority: 4
   },
   {
     id: 'messages',
@@ -40,7 +40,7 @@ const staticCommands: StaticCommandItem[] = [
     action: () => {
       window.location.hash = '?page=/messages';
     },
-    priority: 10
+    priority: 4
   },
   {
     id: 'timetable',
@@ -52,7 +52,7 @@ const staticCommands: StaticCommandItem[] = [
     action: () => {
       window.location.hash = '?page=/timetable';
     },
-    priority: 10
+    priority: 4
   },
   {
     id: 'assessments',
@@ -64,7 +64,7 @@ const staticCommands: StaticCommandItem[] = [
     action: () => {
       window.location.hash = '?page=/assessments';
     },
-    priority: 10
+    priority: 4
   },
   {
     id: 'toggle-dark-mode',
@@ -72,7 +72,7 @@ const staticCommands: StaticCommandItem[] = [
     category: 'action',
     text: 'Toggle Dark Mode',
     action: () => settingsState.DarkMode = !settingsState.DarkMode,
-    priority: 5,
+    priority: 2,
     keywords: ['theme', 'appearance']
   }
 ];

src/plugins/built-in/globalSearch/components/AssessmentItem.svelte

@@ -1,12 +1,13 @@
 <script lang="ts">
-  import { highlightMatch, highlightSnippet } from '../highlightUtils';
+  import { highlightMatch, highlightSnippet, stripHtmlButKeepHighlights } from '../highlightUtils';
   import type { DynamicContentItem } from '../dynamicSearch';
+  import type { FuseResultMatch } from '../types';
 
-  const { item, isSelected, searchTerm, result } = $props<{
+  const { item, isSelected, searchTerm, matches } = $props<{
     item: DynamicContentItem;
     isSelected: boolean;
     searchTerm: string;
-    result: { matches: string[] };
+    matches?: readonly FuseResultMatch[];
   }>();
     
   /* const dueDate = $derived(item.metadata?.dueDate 
@@ -28,12 +29,11 @@
 <button
   class="w-full flex flex-col px-2 py-1.5 rounded-lg select-none cursor-pointer group 
   {isSelected ? 'bg-zinc-900/5 dark:bg-white/10 text-zinc-900 dark:text-white' : 'hover:bg-zinc-500/5 dark:hover:bg-white/5 text-zinc-800 dark:text-zinc-200'}"
-  onclick={() => { item.action(); }}
   >
   <div class="flex items-center w-full">
-      <div class="flex-none w-8 h-8 text-xl font-IconFamily flex items-center justify-center {isSelected ? 'text-zinc-900 dark:text-white' : 'text-zinc-600 dark:text-zinc-400'}">{item.icon}</div>
+      <div class="flex-none w-8 h-8 text-xl font-IconFamily flex items-center justify-center {isSelected ? 'text-zinc-900 dark:text-white' : 'text-zinc-600 dark:text-zinc-400'}">{item.metadata?.icon || '\ue924'}</div>
       <span class="ml-4 text-lg truncate">
-      {@html highlightMatch(item.text, searchTerm, result.matches)}
+      {@html stripHtmlButKeepHighlights(highlightMatch(item.text, searchTerm, matches))}
       </span>
       <span class="flex-none ml-auto text-xs text-zinc-500 dark:text-zinc-400">
       {item.category} 
@@ -41,16 +41,22 @@
   </div>
   {#if item.content}
       <div class="mt-1 ml-12 text-sm text-zinc-600 dark:text-zinc-400 line-clamp-2 text-start">
-      {@html highlightSnippet(item.content, searchTerm, result.matches)}
+      {@html stripHtmlButKeepHighlights(highlightSnippet(item.content, searchTerm, matches))}
       </div>
   {/if}
 </button>
 
 <style>
-  .highlight {
+  :global(.highlight) {
     background-color: rgba(255, 213, 0, 0.3);
     font-weight: 500;
     border-radius: 2px;
+    padding: 0 1px;
+    margin: 0 -1px;
+  }
+  
+  .dark :global(.highlight) {
+    background-color: rgba(255, 230, 100, 0.4);
   }
   
   .due-badge {

src/plugins/built-in/globalSearch/highlightUtils.ts

@@ -1,5 +1,48 @@
 import type { FuseResultMatch, MatchIndices } from './types';
 
+/**
+ * Simple utility to remove HTML tags from a string.
+ */
+export function stripHtmlTags(html: string): string {
+  if (!html) return '';
+  return html.replace(/<[^>]*>/g, '').replace('\n', ' ');
+}
+
+/**
+ * Removes HTML tags from a string, but preserves <span class="highlight"> tags.
+ */
+export function stripHtmlButKeepHighlights(html: string): string {
+  if (!html) return '';
+  // Use a placeholder for highlight tags, strip others, then restore placeholders.
+  const highlightOpenPlaceholder = '__HIGHLIGHT_OPEN__';
+  const highlightClosePlaceholder = '__HIGHLIGHT_CLOSE__';
+
+  let processed = html.replace(/<span class="highlight">/g, highlightOpenPlaceholder);
+  processed = processed.replace(/<\/span>/g, (match, offset, fullString) => {
+    // Only replace </span> if it likely corresponds to our highlight span
+    // This is imperfect but helps avoid replacing unrelated spans.
+    // Look backwards for the nearest opening placeholder.
+    const lastPlaceholder = fullString.lastIndexOf(highlightOpenPlaceholder, offset);
+    if (lastPlaceholder !== -1) {
+      // Check if there's another opening tag between the placeholder and the closing span
+      const interveningContent = fullString.substring(lastPlaceholder + highlightOpenPlaceholder.length, offset);
+      if (!/<span/i.test(interveningContent)) {
+        return highlightClosePlaceholder;
+      }
+    }
+    return match; // Keep the original </span> if unsure
+  });
+  
+  // Strip all remaining HTML tags
+  processed = processed.replace(/<[^>]*>/g, '');
+
+  // Restore the highlight tags
+  processed = processed.replace(new RegExp(highlightOpenPlaceholder, 'g'), '<span class="highlight">');
+  processed = processed.replace(new RegExp(highlightClosePlaceholder, 'g'), '</span>');
+
+  return processed;
+}
+
 export function highlightMatch(
   text: string, 
   term: string, 

src/plugins/built-in/globalSearch/indexing/indexer.ts

@@ -2,6 +2,7 @@ import { getAll, put, clear, remove } from './db';
 import { jobs } from './jobs';
 import { renderComponentMap } from './renderComponents';
 import type { IndexItem, HydratedIndexItem, Job, JobContext } from './types';
+import { processItems } from '../vectorSearch';
 
 const META_STORE = 'meta';
 const LOCK_KEY = 'bsq-indexer-lock';
@@ -94,6 +95,8 @@ export async function runIndexing(): Promise<void> {
   let completedJobs = 0;
   dispatchProgress(completedJobs, jobIds.length, true);
 
+  const allNewItems: HydratedIndexItem[] = [];
+
   for (const jobId of jobIds) {
     const job = jobs[jobId];
     const lastRun = await getLastRunMeta(jobId);
@@ -136,6 +139,13 @@ export async function runIndexing(): Promise<void> {
       await setStoredItems(merged);
       await updateLastRunMeta(jobId);
 
+      // Add to our collection of new items for vector processing
+      const hydratedItems = merged.map(item => ({
+        ...item,
+        renderComponent: renderComponentMap[job.renderComponentId]
+      }));
+      allNewItems.push(...hydratedItems);
+
       console.debug(`%c[Indexer] ✅ ${job.label}: ${newItems.length} items indexed`, 'color: #00c46f');
     } catch (err) {
       console.debug(`%c[Indexer] ❌ ${job.label} failed:`, 'color: red');
@@ -146,6 +156,12 @@ export async function runIndexing(): Promise<void> {
     dispatchProgress(completedJobs, jobIds.length, true);
   }
 
+  // Process all new items through vector search
+  if (allNewItems.length > 0) {
+    console.debug(`%c[Indexer] Processing ${allNewItems.length} items for vector search...`, 'color: #4ea1ff');
+    await processItems(allNewItems);
+  }
+
   stopHeartbeat();
   dispatchProgress(completedJobs, jobIds.length, false);
 }

src/plugins/built-in/globalSearch/indexing/jobs.ts

@@ -228,14 +228,6 @@ export const jobs: Record<string, Job> = {
           }
 
           offset += limit;
-
-          // If we've processed 500 messages and haven't found any existing ones,
-          // assume these are all new (first run) and stop here to avoid overwhelming
-          if (offset >= 500 && consecutiveExisting === 0) {
-            console.debug('[Messages Job] Processed 500 new messages, stopping for now');
-            hasMore = false;
-            break;
-          }
         } catch (error) {
           console.error('Error fetching messages:', error);
           break;

src/plugins/built-in/globalSearch/searchUtils.ts

@@ -1,8 +1,9 @@
 import Fuse, { type FuseResult } from 'fuse.js';
 import { getStaticCommands, type StaticCommandItem } from './commands';
-import { type DynamicContentItem, getDynamicItems } from './dynamicSearch';
+import { getDynamicItems } from './dynamicSearch';
 import type { CombinedResult } from './types';
 import type { HydratedIndexItem } from './indexing/types';
+import { searchVectors, type VectorSearchResult } from './vectorSearch';
 
 // This function is likely no longer needed as items are pre-processed by the indexer
 /* export function prepareDynamicItems(items: DynamicContentItem[]): DynamicContentItem[] {
@@ -47,9 +48,9 @@ export function createSearchIndexes() {
     includeScore: true,
     includeMatches: true,
     threshold: 0.6,
-    minMatchCharLength: 1,
-    ignoreLocation: true,
-    useExtendedSearch: false
+    minMatchCharLength: 3,
+    distance: 50,
+    useExtendedSearch: false,
   };
   
   return {
@@ -141,18 +142,74 @@ export function searchDynamicItems(
   });
 }
 
-export function performSearch(
+export async function performSearch(
   query: string,
   commandsFuse: Fuse<StaticCommandItem>,
   dynamicContentFuse: Fuse<HydratedIndexItem>,
   commandIdToItemMap: Map<string, StaticCommandItem>,
   dynamicIdToItemMap: Map<string, HydratedIndexItem>,
-  showRecentFirst: boolean // Pass sorting preference
-): CombinedResult[] {
+  showRecentFirst: boolean
+): Promise<CombinedResult[]> {
+  const startTime = performance.now();
+  
+  // Get all results first
   const commandResults = searchCommands(commandsFuse, query, commandIdToItemMap);
+  const commandEndTime = performance.now();
   const dynamicResults = searchDynamicItems(dynamicContentFuse, query, dynamicIdToItemMap, 10, showRecentFirst);
+  const fuseEndTime = performance.now();
 
-  const results = [...commandResults, ...dynamicResults];
+  // Get vector results in parallel
+  const vectorResults = await searchVectors(query, 10);
+  const vectorEndTime = performance.now();
+
+  console.log('Vector results:', vectorResults);
+
+  // Log timings
+  console.log(`Command search took ${commandEndTime - startTime} milliseconds`);
+  console.log(`Dynamic search took ${fuseEndTime - commandEndTime} milliseconds`);
+  console.log(`Vector search took ${vectorEndTime - fuseEndTime} milliseconds`);
+
+  // Create a map to store our final results, using ID as key to avoid duplicates
+  const resultMap = new Map<string, CombinedResult>();
+
+  // Add command results first (they keep their original scores)
+  commandResults.forEach(r => resultMap.set(r.id, r));
+
+  // Process dynamic results and vector results together
+  const seenIds = new Set<string>();
+
+  // Add dynamic results first
+  dynamicResults.forEach(r => {
+    seenIds.add(r.id);
+    const vectorMatch = vectorResults.find(v => v.object.id === r.id);
+    if (vectorMatch) {
+      // If we found it in both searches, combine the scores
+      resultMap.set(r.id, {
+        ...r,
+        score: r.score + (vectorMatch.similarity * 0.6) // Boost exact matches
+      });
+    } else {
+      // If only in Fuse results, keep as is
+      resultMap.set(r.id, r);
+    }
+  });
+
+  // Now add any vector results we haven't seen yet
+  vectorResults.forEach(v => {
+    const id = v.object.id;
+    if (!seenIds.has(id)) {
+      // This is a semantic match that Fuse missed - add it with the vector similarity as score
+      resultMap.set(id, {
+        id,
+        type: 'dynamic' as const,
+        score: v.similarity * 0.9, // High base score for semantic matches
+        item: v.object
+      });
+    }
+  });
+
+  // Convert to array and sort by score
+  const results = Array.from(resultMap.values());
   results.sort((a, b) => b.score - a.score);
   
   return results;

src/plugins/built-in/globalSearch/vectorSearch.ts

@@ -0,0 +1,86 @@
+import { getEmbedding, EmbeddingIndex, initializeModel } from 'client-vector-search';
+import type { HydratedIndexItem } from './indexing/types';
+import type { SearchResult } from 'client-vector-search';
+
+let vectorIndex: EmbeddingIndex | null = null;
+
+export async function initVectorSearch() {
+  try {
+    await initializeModel();
+    vectorIndex = new EmbeddingIndex([]);
+    // Load existing items from IndexedDB
+    const stored = await vectorIndex.getAllObjectsFromIndexedDB();
+    if (stored.length > 0) {
+      stored.forEach(item => vectorIndex!.add(item));
+      console.debug('Vector index loaded from IndexedDB');
+    }
+  } catch (e) {
+    console.debug('Creating new vector index');
+    vectorIndex = new EmbeddingIndex([]);
+  }
+}
+
+export async function vectorizeItem(item: HydratedIndexItem): Promise<HydratedIndexItem & { embedding: number[] }> {
+  const textToEmbed = [
+    item.text,
+    item.content,
+    item.category,
+    item.metadata?.author,
+    item.metadata?.subject
+  ].filter(Boolean).join(' ');
+
+  const embedding = await getEmbedding(textToEmbed);
+  return { ...item, embedding };
+}
+
+export async function processItems(items: HydratedIndexItem[]) {
+  if (!vectorIndex) await initVectorSearch();
+  
+  const unprocessedItems = items.filter(item => {
+    try {
+      return !vectorIndex!.get({ id: item.id });
+    } catch {
+      return true;
+    }
+  });
+
+  if (unprocessedItems.length === 0) {
+    console.debug('No new items to vectorize');
+    return;
+  }
+
+  console.debug(`Vectorizing ${unprocessedItems.length} new items...`);
+  
+  // Process in batches to avoid UI freeze
+  const BATCH_SIZE = 5;
+  for (let i = 0; i < unprocessedItems.length; i += BATCH_SIZE) {
+    const batch = unprocessedItems.slice(i, i + BATCH_SIZE);
+    const vectorized = await Promise.all(batch.map(vectorizeItem));
+    
+    for (const item of vectorized) {
+      vectorIndex!.add(item);
+    }
+    
+    // Save periodically to avoid losing progress
+    await vectorIndex!.saveIndex('indexedDB');
+    
+    // Log progress
+    console.debug(`Vectorized ${Math.min(i + BATCH_SIZE, unprocessedItems.length)}/${unprocessedItems.length} items`);
+  }
+}
+
+export interface VectorSearchResult extends SearchResult {
+  object: HydratedIndexItem & { embedding: number[] };
+}
+
+export async function searchVectors(query: string, topK: number = 10): Promise<VectorSearchResult[]> {
+  if (!vectorIndex) await initVectorSearch();
+  
+  const queryEmbedding = await getEmbedding(query);
+  const results = await vectorIndex!.search(queryEmbedding, { 
+    topK,
+    useStorage: 'indexedDB'
+  });
+  
+  return results as VectorSearchResult[];
+}