searchsocket 0.6.0 → 0.6.1

package/dist/cli.js

@@ -12,7 +12,7 @@ import { Command, Option } from "commander";
 // package.json
 var package_default = {
   name: "searchsocket",
-  version: "0.6.0",
+  version: "0.6.1",
   description: "Semantic site search and MCP retrieval for SvelteKit static sites",
   license: "MIT",
   author: "Greg Priday <greg@siteorigin.com>",
@@ -874,7 +874,7 @@ var UpstashSearchStore = class {
   }
   async upsertChunks(chunks, scope) {
     if (chunks.length === 0) return;
-    const BATCH_SIZE = 100;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < chunks.length; i += BATCH_SIZE) {
       const batch = chunks.slice(i, i + BATCH_SIZE);
       await this.chunksNs.upsert(
@@ -1018,7 +1018,7 @@ var UpstashSearchStore = class {
   }
   async deleteByIds(ids, _scope) {
     if (ids.length === 0) return;
-    const BATCH_SIZE = 100;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < ids.length; i += BATCH_SIZE) {
       const batch = ids.slice(i, i + BATCH_SIZE);
       await this.chunksNs.delete(batch);
@@ -1046,7 +1046,7 @@ var UpstashSearchStore = class {
       } catch {
       }
       if (ids.length > 0) {
-        const BATCH_SIZE = 100;
+        const BATCH_SIZE = 90;
         for (let i = 0; i < ids.length; i += BATCH_SIZE) {
           const batch = ids.slice(i, i + BATCH_SIZE);
           await ns.delete(batch);
@@ -1085,7 +1085,39 @@ var UpstashSearchStore = class {
     }));
   }
   async getContentHashes(scope) {
+    return this.scanHashes(this.chunksNs, scope);
+  }
+  /**
+   * Fetch content hashes for a specific set of chunk keys using direct fetch()
+   * instead of range(). This avoids potential issues with range() returning
+   * vectors from the wrong namespace on hybrid indexes.
+   */
+  async fetchContentHashesForKeys(keys, scope) {
     const map = /* @__PURE__ */ new Map();
+    if (keys.length === 0) return map;
+    const BATCH_SIZE = 90;
+    for (let i = 0; i < keys.length; i += BATCH_SIZE) {
+      const batch = keys.slice(i, i + BATCH_SIZE);
+      try {
+        const results = await this.chunksNs.fetch(batch, {
+          includeMetadata: true
+        });
+        for (const doc of results) {
+          if (doc && doc.metadata?.projectId === scope.projectId && doc.metadata?.scopeName === scope.scopeName && doc.metadata?.contentHash) {
+            map.set(String(doc.id), doc.metadata.contentHash);
+          }
+        }
+      } catch {
+      }
+    }
+    return map;
+  }
+  /**
+   * Scan all IDs in the chunks namespace for this scope.
+   * Used for deletion detection (finding stale chunk keys).
+   */
+  async scanChunkIds(scope) {
+    const ids = /* @__PURE__ */ new Set();
     let cursor = "0";
     try {
       for (; ; ) {
@@ -1094,6 +1126,28 @@ var UpstashSearchStore = class {
           limit: 100,
           includeMetadata: true
         });
+        for (const doc of result.vectors) {
+          if (doc.metadata?.projectId === scope.projectId && doc.metadata?.scopeName === scope.scopeName) {
+            ids.add(String(doc.id));
+          }
+        }
+        if (!result.nextCursor || result.nextCursor === "0") break;
+        cursor = result.nextCursor;
+      }
+    } catch {
+    }
+    return ids;
+  }
+  async scanHashes(ns, scope) {
+    const map = /* @__PURE__ */ new Map();
+    let cursor = "0";
+    try {
+      for (; ; ) {
+        const result = await ns.range({
+          cursor,
+          limit: 100,
+          includeMetadata: true
+        });
         for (const doc of result.vectors) {
           if (doc.metadata?.projectId === scope.projectId && doc.metadata?.scopeName === scope.scopeName && doc.metadata?.contentHash) {
             map.set(String(doc.id), doc.metadata.contentHash);
@@ -1156,7 +1210,7 @@ var UpstashSearchStore = class {
   }
   async deletePagesByIds(ids, _scope) {
     if (ids.length === 0) return;
-    const BATCH_SIZE = 50;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < ids.length; i += BATCH_SIZE) {
       const batch = ids.slice(i, i + BATCH_SIZE);
       await this.pagesNs.delete(batch);
@@ -1164,7 +1218,7 @@ var UpstashSearchStore = class {
   }
   async upsertPages(pages, scope) {
     if (pages.length === 0) return;
-    const BATCH_SIZE = 50;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < pages.length; i += BATCH_SIZE) {
       const batch = pages.slice(i, i + BATCH_SIZE);
       await this.pagesNs.upsert(
@@ -1308,7 +1362,7 @@ var UpstashSearchStore = class {
       } catch {
       }
       if (ids.length > 0) {
-        const BATCH_SIZE = 100;
+        const BATCH_SIZE = 90;
         for (let i = 0; i < ids.length; i += BATCH_SIZE) {
           const batch = ids.slice(i, i + BATCH_SIZE);
           await ns.delete(batch);
@@ -3375,10 +3429,9 @@ var IndexPipeline = class _IndexPipeline {
       this.logger.info("Dry run \u2014 no writes will be performed");
     }
     const manifestStart = stageStart();
-    const existingHashes = options.force ? /* @__PURE__ */ new Map() : await this.store.getContentHashes(scope);
     const existingPageHashes = options.force ? /* @__PURE__ */ new Map() : await this.store.getPageHashes(scope);
     stageEnd("manifest", manifestStart);
-    this.logger.debug(`Manifest: ${existingHashes.size} existing chunk hashes, ${existingPageHashes.size} existing page hashes loaded`);
+    this.logger.debug(`Manifest: ${existingPageHashes.size} existing page hashes loaded`);
     const sourceStart = stageStart();
     this.logger.info(`Loading pages (source: ${sourceMode})...`);
     let sourcePages;
@@ -3767,6 +3820,11 @@ var IndexPipeline = class _IndexPipeline {
     for (const chunk of chunks) {
       currentChunkMap.set(chunk.chunkKey, chunk);
     }
+    const chunkHashStart = stageStart();
+    const currentChunkKeys = chunks.map((c) => c.chunkKey);
+    const existingHashes = options.force ? /* @__PURE__ */ new Map() : await this.store.fetchContentHashesForKeys(currentChunkKeys, scope);
+    stageEnd("chunk_hashes", chunkHashStart);
+    this.logger.debug(`Fetched ${existingHashes.size} existing chunk hashes for ${currentChunkKeys.length} current keys`);
     let changedChunks = chunks.filter((chunk) => {
       if (options.force) {
         return true;
@@ -3780,7 +3838,8 @@ var IndexPipeline = class _IndexPipeline {
       }
       return existingHash !== chunk.contentHash;
     });
-    const deletes = [...existingHashes.keys()].filter((chunkKey) => !currentChunkMap.has(chunkKey));
+    const existingChunkIds = options.force ? /* @__PURE__ */ new Set() : await this.store.scanChunkIds(scope);
+    const deletes = [...existingChunkIds].filter((chunkKey) => !currentChunkMap.has(chunkKey));
     if (this.hooks.beforeIndex) {
       changedChunks = await this.hooks.beforeIndex(changedChunks);
     }

package/dist/index.cjs

@@ -17285,7 +17285,7 @@ var UpstashSearchStore = class {
   }
   async upsertChunks(chunks, scope) {
     if (chunks.length === 0) return;
-    const BATCH_SIZE = 100;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < chunks.length; i += BATCH_SIZE) {
       const batch = chunks.slice(i, i + BATCH_SIZE);
       await this.chunksNs.upsert(
@@ -17429,7 +17429,7 @@ var UpstashSearchStore = class {
   }
   async deleteByIds(ids, _scope) {
     if (ids.length === 0) return;
-    const BATCH_SIZE = 100;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < ids.length; i += BATCH_SIZE) {
       const batch = ids.slice(i, i + BATCH_SIZE);
       await this.chunksNs.delete(batch);
@@ -17457,7 +17457,7 @@ var UpstashSearchStore = class {
       } catch {
       }
       if (ids.length > 0) {
-        const BATCH_SIZE = 100;
+        const BATCH_SIZE = 90;
         for (let i = 0; i < ids.length; i += BATCH_SIZE) {
           const batch = ids.slice(i, i + BATCH_SIZE);
           await ns.delete(batch);
@@ -17496,7 +17496,39 @@ var UpstashSearchStore = class {
     }));
   }
   async getContentHashes(scope) {
+    return this.scanHashes(this.chunksNs, scope);
+  }
+  /**
+   * Fetch content hashes for a specific set of chunk keys using direct fetch()
+   * instead of range(). This avoids potential issues with range() returning
+   * vectors from the wrong namespace on hybrid indexes.
+   */
+  async fetchContentHashesForKeys(keys, scope) {
     const map = /* @__PURE__ */ new Map();
+    if (keys.length === 0) return map;
+    const BATCH_SIZE = 90;
+    for (let i = 0; i < keys.length; i += BATCH_SIZE) {
+      const batch = keys.slice(i, i + BATCH_SIZE);
+      try {
+        const results = await this.chunksNs.fetch(batch, {
+          includeMetadata: true
+        });
+        for (const doc of results) {
+          if (doc && doc.metadata?.projectId === scope.projectId && doc.metadata?.scopeName === scope.scopeName && doc.metadata?.contentHash) {
+            map.set(String(doc.id), doc.metadata.contentHash);
+          }
+        }
+      } catch {
+      }
+    }
+    return map;
+  }
+  /**
+   * Scan all IDs in the chunks namespace for this scope.
+   * Used for deletion detection (finding stale chunk keys).
+   */
+  async scanChunkIds(scope) {
+    const ids = /* @__PURE__ */ new Set();
     let cursor = "0";
     try {
       for (; ; ) {
@@ -17505,6 +17537,28 @@ var UpstashSearchStore = class {
           limit: 100,
           includeMetadata: true
         });
+        for (const doc of result.vectors) {
+          if (doc.metadata?.projectId === scope.projectId && doc.metadata?.scopeName === scope.scopeName) {
+            ids.add(String(doc.id));
+          }
+        }
+        if (!result.nextCursor || result.nextCursor === "0") break;
+        cursor = result.nextCursor;
+      }
+    } catch {
+    }
+    return ids;
+  }
+  async scanHashes(ns, scope) {
+    const map = /* @__PURE__ */ new Map();
+    let cursor = "0";
+    try {
+      for (; ; ) {
+        const result = await ns.range({
+          cursor,
+          limit: 100,
+          includeMetadata: true
+        });
         for (const doc of result.vectors) {
           if (doc.metadata?.projectId === scope.projectId && doc.metadata?.scopeName === scope.scopeName && doc.metadata?.contentHash) {
             map.set(String(doc.id), doc.metadata.contentHash);
@@ -17567,7 +17621,7 @@ var UpstashSearchStore = class {
   }
   async deletePagesByIds(ids, _scope) {
     if (ids.length === 0) return;
-    const BATCH_SIZE = 50;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < ids.length; i += BATCH_SIZE) {
       const batch = ids.slice(i, i + BATCH_SIZE);
       await this.pagesNs.delete(batch);
@@ -17575,7 +17629,7 @@ var UpstashSearchStore = class {
   }
   async upsertPages(pages, scope) {
     if (pages.length === 0) return;
-    const BATCH_SIZE = 50;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < pages.length; i += BATCH_SIZE) {
       const batch = pages.slice(i, i + BATCH_SIZE);
       await this.pagesNs.upsert(
@@ -17719,7 +17773,7 @@ var UpstashSearchStore = class {
       } catch {
       }
       if (ids.length > 0) {
-        const BATCH_SIZE = 100;
+        const BATCH_SIZE = 90;
         for (let i = 0; i < ids.length; i += BATCH_SIZE) {
           const batch = ids.slice(i, i + BATCH_SIZE);
           await ns.delete(batch);
@@ -20634,10 +20688,9 @@ var IndexPipeline = class _IndexPipeline {
       this.logger.info("Dry run \u2014 no writes will be performed");
     }
     const manifestStart = stageStart();
-    const existingHashes = options.force ? /* @__PURE__ */ new Map() : await this.store.getContentHashes(scope);
     const existingPageHashes = options.force ? /* @__PURE__ */ new Map() : await this.store.getPageHashes(scope);
     stageEnd("manifest", manifestStart);
-    this.logger.debug(`Manifest: ${existingHashes.size} existing chunk hashes, ${existingPageHashes.size} existing page hashes loaded`);
+    this.logger.debug(`Manifest: ${existingPageHashes.size} existing page hashes loaded`);
     const sourceStart = stageStart();
     this.logger.info(`Loading pages (source: ${sourceMode})...`);
     let sourcePages;
@@ -21026,6 +21079,11 @@ var IndexPipeline = class _IndexPipeline {
     for (const chunk of chunks) {
       currentChunkMap.set(chunk.chunkKey, chunk);
     }
+    const chunkHashStart = stageStart();
+    const currentChunkKeys = chunks.map((c) => c.chunkKey);
+    const existingHashes = options.force ? /* @__PURE__ */ new Map() : await this.store.fetchContentHashesForKeys(currentChunkKeys, scope);
+    stageEnd("chunk_hashes", chunkHashStart);
+    this.logger.debug(`Fetched ${existingHashes.size} existing chunk hashes for ${currentChunkKeys.length} current keys`);
     let changedChunks = chunks.filter((chunk) => {
       if (options.force) {
         return true;
@@ -21039,7 +21097,8 @@ var IndexPipeline = class _IndexPipeline {
       }
       return existingHash !== chunk.contentHash;
     });
-    const deletes = [...existingHashes.keys()].filter((chunkKey) => !currentChunkMap.has(chunkKey));
+    const existingChunkIds = options.force ? /* @__PURE__ */ new Set() : await this.store.scanChunkIds(scope);
+    const deletes = [...existingChunkIds].filter((chunkKey) => !currentChunkMap.has(chunkKey));
     if (this.hooks.beforeIndex) {
       changedChunks = await this.hooks.beforeIndex(changedChunks);
     }

package/dist/index.d.cts

@@ -88,6 +88,18 @@ declare class UpstashSearchStore {
     deleteScope(scope: Scope): Promise<void>;
     listScopes(projectId: string): Promise<ScopeInfo[]>;
     getContentHashes(scope: Scope): Promise<Map<string, string>>;
+    /**
+     * Fetch content hashes for a specific set of chunk keys using direct fetch()
+     * instead of range(). This avoids potential issues with range() returning
+     * vectors from the wrong namespace on hybrid indexes.
+     */
+    fetchContentHashesForKeys(keys: string[], scope: Scope): Promise<Map<string, string>>;
+    /**
+     * Scan all IDs in the chunks namespace for this scope.
+     * Used for deletion detection (finding stale chunk keys).
+     */
+    scanChunkIds(scope: Scope): Promise<Set<string>>;
+    private scanHashes;
     listPages(scope: Scope, opts?: {
         cursor?: string;
         limit?: number;

package/dist/index.d.ts

@@ -88,6 +88,18 @@ declare class UpstashSearchStore {
     deleteScope(scope: Scope): Promise<void>;
     listScopes(projectId: string): Promise<ScopeInfo[]>;
     getContentHashes(scope: Scope): Promise<Map<string, string>>;
+    /**
+     * Fetch content hashes for a specific set of chunk keys using direct fetch()
+     * instead of range(). This avoids potential issues with range() returning
+     * vectors from the wrong namespace on hybrid indexes.
+     */
+    fetchContentHashesForKeys(keys: string[], scope: Scope): Promise<Map<string, string>>;
+    /**
+     * Scan all IDs in the chunks namespace for this scope.
+     * Used for deletion detection (finding stale chunk keys).
+     */
+    scanChunkIds(scope: Scope): Promise<Set<string>>;
+    private scanHashes;
     listPages(scope: Scope, opts?: {
         cursor?: string;
         limit?: number;

package/dist/index.js

@@ -17273,7 +17273,7 @@ var UpstashSearchStore = class {
   }
   async upsertChunks(chunks, scope) {
     if (chunks.length === 0) return;
-    const BATCH_SIZE = 100;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < chunks.length; i += BATCH_SIZE) {
       const batch = chunks.slice(i, i + BATCH_SIZE);
       await this.chunksNs.upsert(
@@ -17417,7 +17417,7 @@ var UpstashSearchStore = class {
   }
   async deleteByIds(ids, _scope) {
     if (ids.length === 0) return;
-    const BATCH_SIZE = 100;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < ids.length; i += BATCH_SIZE) {
       const batch = ids.slice(i, i + BATCH_SIZE);
       await this.chunksNs.delete(batch);
@@ -17445,7 +17445,7 @@ var UpstashSearchStore = class {
       } catch {
       }
       if (ids.length > 0) {
-        const BATCH_SIZE = 100;
+        const BATCH_SIZE = 90;
         for (let i = 0; i < ids.length; i += BATCH_SIZE) {
           const batch = ids.slice(i, i + BATCH_SIZE);
           await ns.delete(batch);
@@ -17484,7 +17484,39 @@ var UpstashSearchStore = class {
     }));
   }
   async getContentHashes(scope) {
+    return this.scanHashes(this.chunksNs, scope);
+  }
+  /**
+   * Fetch content hashes for a specific set of chunk keys using direct fetch()
+   * instead of range(). This avoids potential issues with range() returning
+   * vectors from the wrong namespace on hybrid indexes.
+   */
+  async fetchContentHashesForKeys(keys, scope) {
     const map = /* @__PURE__ */ new Map();
+    if (keys.length === 0) return map;
+    const BATCH_SIZE = 90;
+    for (let i = 0; i < keys.length; i += BATCH_SIZE) {
+      const batch = keys.slice(i, i + BATCH_SIZE);
+      try {
+        const results = await this.chunksNs.fetch(batch, {
+          includeMetadata: true
+        });
+        for (const doc of results) {
+          if (doc && doc.metadata?.projectId === scope.projectId && doc.metadata?.scopeName === scope.scopeName && doc.metadata?.contentHash) {
+            map.set(String(doc.id), doc.metadata.contentHash);
+          }
+        }
+      } catch {
+      }
+    }
+    return map;
+  }
+  /**
+   * Scan all IDs in the chunks namespace for this scope.
+   * Used for deletion detection (finding stale chunk keys).
+   */
+  async scanChunkIds(scope) {
+    const ids = /* @__PURE__ */ new Set();
     let cursor = "0";
     try {
       for (; ; ) {
@@ -17493,6 +17525,28 @@ var UpstashSearchStore = class {
           limit: 100,
           includeMetadata: true
         });
+        for (const doc of result.vectors) {
+          if (doc.metadata?.projectId === scope.projectId && doc.metadata?.scopeName === scope.scopeName) {
+            ids.add(String(doc.id));
+          }
+        }
+        if (!result.nextCursor || result.nextCursor === "0") break;
+        cursor = result.nextCursor;
+      }
+    } catch {
+    }
+    return ids;
+  }
+  async scanHashes(ns, scope) {
+    const map = /* @__PURE__ */ new Map();
+    let cursor = "0";
+    try {
+      for (; ; ) {
+        const result = await ns.range({
+          cursor,
+          limit: 100,
+          includeMetadata: true
+        });
         for (const doc of result.vectors) {
           if (doc.metadata?.projectId === scope.projectId && doc.metadata?.scopeName === scope.scopeName && doc.metadata?.contentHash) {
             map.set(String(doc.id), doc.metadata.contentHash);
@@ -17555,7 +17609,7 @@ var UpstashSearchStore = class {
   }
   async deletePagesByIds(ids, _scope) {
     if (ids.length === 0) return;
-    const BATCH_SIZE = 50;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < ids.length; i += BATCH_SIZE) {
       const batch = ids.slice(i, i + BATCH_SIZE);
       await this.pagesNs.delete(batch);
@@ -17563,7 +17617,7 @@ var UpstashSearchStore = class {
   }
   async upsertPages(pages, scope) {
     if (pages.length === 0) return;
-    const BATCH_SIZE = 50;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < pages.length; i += BATCH_SIZE) {
       const batch = pages.slice(i, i + BATCH_SIZE);
       await this.pagesNs.upsert(
@@ -17707,7 +17761,7 @@ var UpstashSearchStore = class {
       } catch {
       }
       if (ids.length > 0) {
-        const BATCH_SIZE = 100;
+        const BATCH_SIZE = 90;
         for (let i = 0; i < ids.length; i += BATCH_SIZE) {
           const batch = ids.slice(i, i + BATCH_SIZE);
           await ns.delete(batch);
@@ -20622,10 +20676,9 @@ var IndexPipeline = class _IndexPipeline {
       this.logger.info("Dry run \u2014 no writes will be performed");
     }
     const manifestStart = stageStart();
-    const existingHashes = options.force ? /* @__PURE__ */ new Map() : await this.store.getContentHashes(scope);
     const existingPageHashes = options.force ? /* @__PURE__ */ new Map() : await this.store.getPageHashes(scope);
     stageEnd("manifest", manifestStart);
-    this.logger.debug(`Manifest: ${existingHashes.size} existing chunk hashes, ${existingPageHashes.size} existing page hashes loaded`);
+    this.logger.debug(`Manifest: ${existingPageHashes.size} existing page hashes loaded`);
     const sourceStart = stageStart();
     this.logger.info(`Loading pages (source: ${sourceMode})...`);
     let sourcePages;
@@ -21014,6 +21067,11 @@ var IndexPipeline = class _IndexPipeline {
     for (const chunk of chunks) {
       currentChunkMap.set(chunk.chunkKey, chunk);
     }
+    const chunkHashStart = stageStart();
+    const currentChunkKeys = chunks.map((c) => c.chunkKey);
+    const existingHashes = options.force ? /* @__PURE__ */ new Map() : await this.store.fetchContentHashesForKeys(currentChunkKeys, scope);
+    stageEnd("chunk_hashes", chunkHashStart);
+    this.logger.debug(`Fetched ${existingHashes.size} existing chunk hashes for ${currentChunkKeys.length} current keys`);
     let changedChunks = chunks.filter((chunk) => {
       if (options.force) {
         return true;
@@ -21027,7 +21085,8 @@ var IndexPipeline = class _IndexPipeline {
       }
       return existingHash !== chunk.contentHash;
     });
-    const deletes = [...existingHashes.keys()].filter((chunkKey) => !currentChunkMap.has(chunkKey));
+    const existingChunkIds = options.force ? /* @__PURE__ */ new Set() : await this.store.scanChunkIds(scope);
+    const deletes = [...existingChunkIds].filter((chunkKey) => !currentChunkMap.has(chunkKey));
     if (this.hooks.beforeIndex) {
       changedChunks = await this.hooks.beforeIndex(changedChunks);
     }

package/dist/sveltekit.cjs

@@ -17316,7 +17316,7 @@ var UpstashSearchStore = class {
   }
   async upsertChunks(chunks, scope) {
     if (chunks.length === 0) return;
-    const BATCH_SIZE = 100;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < chunks.length; i += BATCH_SIZE) {
       const batch = chunks.slice(i, i + BATCH_SIZE);
       await this.chunksNs.upsert(
@@ -17460,7 +17460,7 @@ var UpstashSearchStore = class {
   }
   async deleteByIds(ids, _scope) {
     if (ids.length === 0) return;
-    const BATCH_SIZE = 100;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < ids.length; i += BATCH_SIZE) {
       const batch = ids.slice(i, i + BATCH_SIZE);
       await this.chunksNs.delete(batch);
@@ -17488,7 +17488,7 @@ var UpstashSearchStore = class {
       } catch {
       }
       if (ids.length > 0) {
-        const BATCH_SIZE = 100;
+        const BATCH_SIZE = 90;
         for (let i = 0; i < ids.length; i += BATCH_SIZE) {
           const batch = ids.slice(i, i + BATCH_SIZE);
           await ns.delete(batch);
@@ -17527,7 +17527,39 @@ var UpstashSearchStore = class {
     }));
   }
   async getContentHashes(scope) {
+    return this.scanHashes(this.chunksNs, scope);
+  }
+  /**
+   * Fetch content hashes for a specific set of chunk keys using direct fetch()
+   * instead of range(). This avoids potential issues with range() returning
+   * vectors from the wrong namespace on hybrid indexes.
+   */
+  async fetchContentHashesForKeys(keys, scope) {
     const map = /* @__PURE__ */ new Map();
+    if (keys.length === 0) return map;
+    const BATCH_SIZE = 90;
+    for (let i = 0; i < keys.length; i += BATCH_SIZE) {
+      const batch = keys.slice(i, i + BATCH_SIZE);
+      try {
+        const results = await this.chunksNs.fetch(batch, {
+          includeMetadata: true
+        });
+        for (const doc of results) {
+          if (doc && doc.metadata?.projectId === scope.projectId && doc.metadata?.scopeName === scope.scopeName && doc.metadata?.contentHash) {
+            map.set(String(doc.id), doc.metadata.contentHash);
+          }
+        }
+      } catch {
+      }
+    }
+    return map;
+  }
+  /**
+   * Scan all IDs in the chunks namespace for this scope.
+   * Used for deletion detection (finding stale chunk keys).
+   */
+  async scanChunkIds(scope) {
+    const ids = /* @__PURE__ */ new Set();
     let cursor = "0";
     try {
       for (; ; ) {
@@ -17536,6 +17568,28 @@ var UpstashSearchStore = class {
           limit: 100,
           includeMetadata: true
         });
+        for (const doc of result.vectors) {
+          if (doc.metadata?.projectId === scope.projectId && doc.metadata?.scopeName === scope.scopeName) {
+            ids.add(String(doc.id));
+          }
+        }
+        if (!result.nextCursor || result.nextCursor === "0") break;
+        cursor = result.nextCursor;
+      }
+    } catch {
+    }
+    return ids;
+  }
+  async scanHashes(ns, scope) {
+    const map = /* @__PURE__ */ new Map();
+    let cursor = "0";
+    try {
+      for (; ; ) {
+        const result = await ns.range({
+          cursor,
+          limit: 100,
+          includeMetadata: true
+        });
         for (const doc of result.vectors) {
           if (doc.metadata?.projectId === scope.projectId && doc.metadata?.scopeName === scope.scopeName && doc.metadata?.contentHash) {
             map.set(String(doc.id), doc.metadata.contentHash);
@@ -17598,7 +17652,7 @@ var UpstashSearchStore = class {
   }
   async deletePagesByIds(ids, _scope) {
     if (ids.length === 0) return;
-    const BATCH_SIZE = 50;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < ids.length; i += BATCH_SIZE) {
       const batch = ids.slice(i, i + BATCH_SIZE);
       await this.pagesNs.delete(batch);
@@ -17606,7 +17660,7 @@ var UpstashSearchStore = class {
   }
   async upsertPages(pages, scope) {
     if (pages.length === 0) return;
-    const BATCH_SIZE = 50;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < pages.length; i += BATCH_SIZE) {
       const batch = pages.slice(i, i + BATCH_SIZE);
       await this.pagesNs.upsert(
@@ -17750,7 +17804,7 @@ var UpstashSearchStore = class {
       } catch {
       }
       if (ids.length > 0) {
-        const BATCH_SIZE = 100;
+        const BATCH_SIZE = 90;
         for (let i = 0; i < ids.length; i += BATCH_SIZE) {
           const batch = ids.slice(i, i + BATCH_SIZE);
           await ns.delete(batch);
@@ -21776,10 +21830,9 @@ var IndexPipeline = class _IndexPipeline {
       this.logger.info("Dry run \u2014 no writes will be performed");
     }
     const manifestStart = stageStart();
-    const existingHashes = options.force ? /* @__PURE__ */ new Map() : await this.store.getContentHashes(scope);
     const existingPageHashes = options.force ? /* @__PURE__ */ new Map() : await this.store.getPageHashes(scope);
     stageEnd("manifest", manifestStart);
-    this.logger.debug(`Manifest: ${existingHashes.size} existing chunk hashes, ${existingPageHashes.size} existing page hashes loaded`);
+    this.logger.debug(`Manifest: ${existingPageHashes.size} existing page hashes loaded`);
     const sourceStart = stageStart();
     this.logger.info(`Loading pages (source: ${sourceMode})...`);
     let sourcePages;
@@ -22168,6 +22221,11 @@ var IndexPipeline = class _IndexPipeline {
     for (const chunk of chunks) {
       currentChunkMap.set(chunk.chunkKey, chunk);
     }
+    const chunkHashStart = stageStart();
+    const currentChunkKeys = chunks.map((c) => c.chunkKey);
+    const existingHashes = options.force ? /* @__PURE__ */ new Map() : await this.store.fetchContentHashesForKeys(currentChunkKeys, scope);
+    stageEnd("chunk_hashes", chunkHashStart);
+    this.logger.debug(`Fetched ${existingHashes.size} existing chunk hashes for ${currentChunkKeys.length} current keys`);
     let changedChunks = chunks.filter((chunk) => {
       if (options.force) {
         return true;
@@ -22181,7 +22239,8 @@ var IndexPipeline = class _IndexPipeline {
       }
       return existingHash !== chunk.contentHash;
     });
-    const deletes = [...existingHashes.keys()].filter((chunkKey) => !currentChunkMap.has(chunkKey));
+    const existingChunkIds = options.force ? /* @__PURE__ */ new Set() : await this.store.scanChunkIds(scope);
+    const deletes = [...existingChunkIds].filter((chunkKey) => !currentChunkMap.has(chunkKey));
     if (this.hooks.beforeIndex) {
       changedChunks = await this.hooks.beforeIndex(changedChunks);
     }

package/dist/sveltekit.js

@@ -17304,7 +17304,7 @@ var UpstashSearchStore = class {
   }
   async upsertChunks(chunks, scope) {
     if (chunks.length === 0) return;
-    const BATCH_SIZE = 100;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < chunks.length; i += BATCH_SIZE) {
       const batch = chunks.slice(i, i + BATCH_SIZE);
       await this.chunksNs.upsert(
@@ -17448,7 +17448,7 @@ var UpstashSearchStore = class {
   }
   async deleteByIds(ids, _scope) {
     if (ids.length === 0) return;
-    const BATCH_SIZE = 100;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < ids.length; i += BATCH_SIZE) {
       const batch = ids.slice(i, i + BATCH_SIZE);
       await this.chunksNs.delete(batch);
@@ -17476,7 +17476,7 @@ var UpstashSearchStore = class {
       } catch {
       }
       if (ids.length > 0) {
-        const BATCH_SIZE = 100;
+        const BATCH_SIZE = 90;
         for (let i = 0; i < ids.length; i += BATCH_SIZE) {
           const batch = ids.slice(i, i + BATCH_SIZE);
           await ns.delete(batch);
@@ -17515,7 +17515,39 @@ var UpstashSearchStore = class {
     }));
   }
   async getContentHashes(scope) {
+    return this.scanHashes(this.chunksNs, scope);
+  }
+  /**
+   * Fetch content hashes for a specific set of chunk keys using direct fetch()
+   * instead of range(). This avoids potential issues with range() returning
+   * vectors from the wrong namespace on hybrid indexes.
+   */
+  async fetchContentHashesForKeys(keys, scope) {
     const map = /* @__PURE__ */ new Map();
+    if (keys.length === 0) return map;
+    const BATCH_SIZE = 90;
+    for (let i = 0; i < keys.length; i += BATCH_SIZE) {
+      const batch = keys.slice(i, i + BATCH_SIZE);
+      try {
+        const results = await this.chunksNs.fetch(batch, {
+          includeMetadata: true
+        });
+        for (const doc of results) {
+          if (doc && doc.metadata?.projectId === scope.projectId && doc.metadata?.scopeName === scope.scopeName && doc.metadata?.contentHash) {
+            map.set(String(doc.id), doc.metadata.contentHash);
+          }
+        }
+      } catch {
+      }
+    }
+    return map;
+  }
+  /**
+   * Scan all IDs in the chunks namespace for this scope.
+   * Used for deletion detection (finding stale chunk keys).
+   */
+  async scanChunkIds(scope) {
+    const ids = /* @__PURE__ */ new Set();
     let cursor = "0";
     try {
       for (; ; ) {
@@ -17524,6 +17556,28 @@ var UpstashSearchStore = class {
           limit: 100,
           includeMetadata: true
         });
+        for (const doc of result.vectors) {
+          if (doc.metadata?.projectId === scope.projectId && doc.metadata?.scopeName === scope.scopeName) {
+            ids.add(String(doc.id));
+          }
+        }
+        if (!result.nextCursor || result.nextCursor === "0") break;
+        cursor = result.nextCursor;
+      }
+    } catch {
+    }
+    return ids;
+  }
+  async scanHashes(ns, scope) {
+    const map = /* @__PURE__ */ new Map();
+    let cursor = "0";
+    try {
+      for (; ; ) {
+        const result = await ns.range({
+          cursor,
+          limit: 100,
+          includeMetadata: true
+        });
         for (const doc of result.vectors) {
           if (doc.metadata?.projectId === scope.projectId && doc.metadata?.scopeName === scope.scopeName && doc.metadata?.contentHash) {
             map.set(String(doc.id), doc.metadata.contentHash);
@@ -17586,7 +17640,7 @@ var UpstashSearchStore = class {
   }
   async deletePagesByIds(ids, _scope) {
     if (ids.length === 0) return;
-    const BATCH_SIZE = 50;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < ids.length; i += BATCH_SIZE) {
       const batch = ids.slice(i, i + BATCH_SIZE);
       await this.pagesNs.delete(batch);
@@ -17594,7 +17648,7 @@ var UpstashSearchStore = class {
   }
   async upsertPages(pages, scope) {
     if (pages.length === 0) return;
-    const BATCH_SIZE = 50;
+    const BATCH_SIZE = 90;
     for (let i = 0; i < pages.length; i += BATCH_SIZE) {
       const batch = pages.slice(i, i + BATCH_SIZE);
       await this.pagesNs.upsert(
@@ -17738,7 +17792,7 @@ var UpstashSearchStore = class {
       } catch {
       }
       if (ids.length > 0) {
-        const BATCH_SIZE = 100;
+        const BATCH_SIZE = 90;
         for (let i = 0; i < ids.length; i += BATCH_SIZE) {
           const batch = ids.slice(i, i + BATCH_SIZE);
           await ns.delete(batch);
@@ -21764,10 +21818,9 @@ var IndexPipeline = class _IndexPipeline {
       this.logger.info("Dry run \u2014 no writes will be performed");
     }
     const manifestStart = stageStart();
-    const existingHashes = options.force ? /* @__PURE__ */ new Map() : await this.store.getContentHashes(scope);
     const existingPageHashes = options.force ? /* @__PURE__ */ new Map() : await this.store.getPageHashes(scope);
     stageEnd("manifest", manifestStart);
-    this.logger.debug(`Manifest: ${existingHashes.size} existing chunk hashes, ${existingPageHashes.size} existing page hashes loaded`);
+    this.logger.debug(`Manifest: ${existingPageHashes.size} existing page hashes loaded`);
     const sourceStart = stageStart();
     this.logger.info(`Loading pages (source: ${sourceMode})...`);
     let sourcePages;
@@ -22156,6 +22209,11 @@ var IndexPipeline = class _IndexPipeline {
     for (const chunk of chunks) {
       currentChunkMap.set(chunk.chunkKey, chunk);
     }
+    const chunkHashStart = stageStart();
+    const currentChunkKeys = chunks.map((c) => c.chunkKey);
+    const existingHashes = options.force ? /* @__PURE__ */ new Map() : await this.store.fetchContentHashesForKeys(currentChunkKeys, scope);
+    stageEnd("chunk_hashes", chunkHashStart);
+    this.logger.debug(`Fetched ${existingHashes.size} existing chunk hashes for ${currentChunkKeys.length} current keys`);
     let changedChunks = chunks.filter((chunk) => {
       if (options.force) {
         return true;
@@ -22169,7 +22227,8 @@ var IndexPipeline = class _IndexPipeline {
       }
       return existingHash !== chunk.contentHash;
     });
-    const deletes = [...existingHashes.keys()].filter((chunkKey) => !currentChunkMap.has(chunkKey));
+    const existingChunkIds = options.force ? /* @__PURE__ */ new Set() : await this.store.scanChunkIds(scope);
+    const deletes = [...existingChunkIds].filter((chunkKey) => !currentChunkMap.has(chunkKey));
     if (this.hooks.beforeIndex) {
       changedChunks = await this.hooks.beforeIndex(changedChunks);
     }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "searchsocket",
-  "version": "0.6.0",
+  "version": "0.6.1",
   "description": "Semantic site search and MCP retrieval for SvelteKit static sites",
   "license": "MIT",
   "author": "Greg Priday <greg@siteorigin.com>",
@@ -67,18 +67,9 @@
       "optional": true
     }
   },
-  "scripts": {
-    "build": "tsup",
-    "clean": "rm -rf dist",
-    "typecheck": "tsc --noEmit",
-    "test": "vitest run",
-    "test:watch": "vitest",
-    "test:quality": "SEARCHSOCKET_QUALITY_TESTS=1 vitest run tests/quality.test.ts"
-  },
   "engines": {
     "node": ">=20"
   },
-  "packageManager": "pnpm@10.29.2",
   "dependencies": {
     "@clack/prompts": "^1.2.0",
     "@modelcontextprotocol/sdk": "^1.26.0",
@@ -107,5 +98,13 @@
     "tsup": "^8.5.1",
     "typescript": "^5.9.3",
     "vitest": "^4.0.18"
+  },
+  "scripts": {
+    "build": "tsup",
+    "clean": "rm -rf dist",
+    "typecheck": "tsc --noEmit",
+    "test": "vitest run",
+    "test:watch": "vitest",
+    "test:quality": "SEARCHSOCKET_QUALITY_TESTS=1 vitest run tests/quality.test.ts"
   }
-}
+}