searchsocket 0.6.2 → 0.6.3

package/dist/cli.js

@@ -12,7 +12,7 @@ import { Command, Option } from "commander";
 // package.json
 var package_default = {
   name: "searchsocket",
-  version: "0.6.2",
+  version: "0.6.3",
   description: "Semantic site search and MCP retrieval for SvelteKit static sites",
   license: "MIT",
   author: "Greg Priday <greg@siteorigin.com>",
@@ -863,6 +863,26 @@ import path11 from "path";
 
 // src/vector/upstash.ts
 import { QueryMode, FusionAlgorithm } from "@upstash/vector";
+function reconstructMarkdownFromChunks(chunks, pageTitle) {
+  if (chunks.length === 0) return "";
+  const parts = [];
+  for (const chunk of chunks) {
+    let text2 = chunk.chunkText;
+    const prefixWithSection = `${pageTitle} \u2014 ${chunk.sectionTitle}
+
+`;
+    const prefixWithoutSection = `${pageTitle}
+
+`;
+    if (chunk.sectionTitle && text2.startsWith(prefixWithSection)) {
+      text2 = text2.slice(prefixWithSection.length);
+    } else if (text2.startsWith(prefixWithoutSection)) {
+      text2 = text2.slice(prefixWithoutSection.length);
+    }
+    parts.push(text2.trim());
+  }
+  return parts.join("\n\n");
+}
 var UpstashSearchStore = class {
   index;
   pagesNs;
@@ -1242,10 +1262,12 @@ var UpstashSearchStore = class {
       });
       const doc = results[0];
       if (!doc || !doc.metadata) return null;
+      const chunks = await this.getChunksForPage(url, scope);
+      const markdown = reconstructMarkdownFromChunks(chunks, doc.metadata.title);
       return {
         url: doc.metadata.url,
         title: doc.metadata.title,
-        markdown: doc.metadata.markdown,
+        markdown,
         projectId: doc.metadata.projectId,
         scopeName: doc.metadata.scopeName,
         routeFile: doc.metadata.routeFile,
@@ -1265,6 +1287,37 @@ var UpstashSearchStore = class {
       return null;
     }
   }
+  /**
+   * Fetch all chunks belonging to a specific page URL, sorted by ordinal.
+   * Used to reconstruct full page markdown from chunk content.
+   */
+  async getChunksForPage(url, scope) {
+    const chunks = [];
+    let cursor = "0";
+    try {
+      for (; ; ) {
+        const result = await this.chunksNs.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?.url === url) {
+            chunks.push({
+              chunkText: doc.metadata.chunkText ?? "",
+              ordinal: doc.metadata.ordinal ?? 0,
+              sectionTitle: doc.metadata.sectionTitle ?? "",
+              headingPath: doc.metadata.headingPath ? String(doc.metadata.headingPath).split(" > ").filter(Boolean) : []
+            });
+          }
+        }
+        if (!result.nextCursor || result.nextCursor === "0") break;
+        cursor = result.nextCursor;
+      }
+    } catch {
+    }
+    return chunks.sort((a, b) => a.ordinal - b.ordinal);
+  }
   async fetchPageWithVector(url, scope) {
     try {
       const results = await this.pagesNs.fetch([url], {
@@ -3737,7 +3790,6 @@ var IndexPipeline = class _IndexPipeline {
             keywords: r.keywords ?? [],
             summary: r.summary ?? "",
             tags: r.tags,
-            markdown: r.markdown,
             routeFile: r.routeFile,
             routeResolution: r.routeResolution,
             incomingLinks: r.incomingLinks,
@@ -3764,7 +3816,6 @@ var IndexPipeline = class _IndexPipeline {
               keywords: r.keywords ?? [],
               summary: r.summary ?? "",
               tags: r.tags,
-              markdown: r.markdown,
               routeFile: r.routeFile,
               routeResolution: r.routeResolution,
               incomingLinks: r.incomingLinks,
@@ -3848,6 +3899,7 @@ var IndexPipeline = class _IndexPipeline {
     let documentsUpserted = 0;
     if (!options.dryRun && changedChunks.length > 0) {
       this.logger.info(`Upserting ${changedChunks.length} chunk${changedChunks.length === 1 ? "" : "s"} to Upstash Vector...`);
+      const CHUNK_TEXT_MAX_CHARS = 3e4;
       const docs = changedChunks.map((chunk) => {
         const embeddingText = buildEmbeddingText(chunk, this.config.chunking.prependTitle);
         if (embeddingText.length > 2e3) {
@@ -3855,6 +3907,7 @@ var IndexPipeline = class _IndexPipeline {
             `Chunk ${chunk.chunkKey} text is ${embeddingText.length} chars (~${Math.round(embeddingText.length / 4)} tokens), which may exceed the 512-token model limit and be silently truncated.`
           );
         }
+        const cappedText = embeddingText.length > CHUNK_TEXT_MAX_CHARS ? embeddingText.slice(0, CHUNK_TEXT_MAX_CHARS) : embeddingText;
         return {
           id: chunk.chunkKey,
           data: embeddingText,
@@ -3865,7 +3918,7 @@ var IndexPipeline = class _IndexPipeline {
             sectionTitle: chunk.sectionTitle ?? "",
             headingPath: chunk.headingPath.join(" > "),
             snippet: chunk.snippet,
-            chunkText: embeddingText,
+            chunkText: cappedText,
             tags: chunk.tags,
             ordinal: chunk.ordinal,
             contentHash: chunk.contentHash,

package/dist/index.cjs

@@ -17274,6 +17274,26 @@ function ensureStateDirs(cwd, stateDir, scope) {
   fs__default.default.mkdirSync(statePath, { recursive: true });
   return { statePath };
 }
+function reconstructMarkdownFromChunks(chunks, pageTitle) {
+  if (chunks.length === 0) return "";
+  const parts = [];
+  for (const chunk of chunks) {
+    let text = chunk.chunkText;
+    const prefixWithSection = `${pageTitle} \u2014 ${chunk.sectionTitle}
+
+`;
+    const prefixWithoutSection = `${pageTitle}
+
+`;
+    if (chunk.sectionTitle && text.startsWith(prefixWithSection)) {
+      text = text.slice(prefixWithSection.length);
+    } else if (text.startsWith(prefixWithoutSection)) {
+      text = text.slice(prefixWithoutSection.length);
+    }
+    parts.push(text.trim());
+  }
+  return parts.join("\n\n");
+}
 var UpstashSearchStore = class {
   index;
   pagesNs;
@@ -17653,10 +17673,12 @@ var UpstashSearchStore = class {
       });
       const doc = results[0];
       if (!doc || !doc.metadata) return null;
+      const chunks = await this.getChunksForPage(url, scope);
+      const markdown = reconstructMarkdownFromChunks(chunks, doc.metadata.title);
       return {
         url: doc.metadata.url,
         title: doc.metadata.title,
-        markdown: doc.metadata.markdown,
+        markdown,
         projectId: doc.metadata.projectId,
         scopeName: doc.metadata.scopeName,
         routeFile: doc.metadata.routeFile,
@@ -17676,6 +17698,37 @@ var UpstashSearchStore = class {
       return null;
     }
   }
+  /**
+   * Fetch all chunks belonging to a specific page URL, sorted by ordinal.
+   * Used to reconstruct full page markdown from chunk content.
+   */
+  async getChunksForPage(url, scope) {
+    const chunks = [];
+    let cursor = "0";
+    try {
+      for (; ; ) {
+        const result = await this.chunksNs.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?.url === url) {
+            chunks.push({
+              chunkText: doc.metadata.chunkText ?? "",
+              ordinal: doc.metadata.ordinal ?? 0,
+              sectionTitle: doc.metadata.sectionTitle ?? "",
+              headingPath: doc.metadata.headingPath ? String(doc.metadata.headingPath).split(" > ").filter(Boolean) : []
+            });
+          }
+        }
+        if (!result.nextCursor || result.nextCursor === "0") break;
+        cursor = result.nextCursor;
+      }
+    } catch {
+    }
+    return chunks.sort((a, b) => a.ordinal - b.ordinal);
+  }
   async fetchPageWithVector(url, scope) {
     try {
       const results = await this.pagesNs.fetch([url], {
@@ -20996,7 +21049,6 @@ var IndexPipeline = class _IndexPipeline {
             keywords: r.keywords ?? [],
             summary: r.summary ?? "",
             tags: r.tags,
-            markdown: r.markdown,
             routeFile: r.routeFile,
             routeResolution: r.routeResolution,
             incomingLinks: r.incomingLinks,
@@ -21023,7 +21075,6 @@ var IndexPipeline = class _IndexPipeline {
               keywords: r.keywords ?? [],
               summary: r.summary ?? "",
               tags: r.tags,
-              markdown: r.markdown,
               routeFile: r.routeFile,
               routeResolution: r.routeResolution,
               incomingLinks: r.incomingLinks,
@@ -21107,6 +21158,7 @@ var IndexPipeline = class _IndexPipeline {
     let documentsUpserted = 0;
     if (!options.dryRun && changedChunks.length > 0) {
       this.logger.info(`Upserting ${changedChunks.length} chunk${changedChunks.length === 1 ? "" : "s"} to Upstash Vector...`);
+      const CHUNK_TEXT_MAX_CHARS = 3e4;
       const docs = changedChunks.map((chunk) => {
         const embeddingText = buildEmbeddingText(chunk, this.config.chunking.prependTitle);
         if (embeddingText.length > 2e3) {
@@ -21114,6 +21166,7 @@ var IndexPipeline = class _IndexPipeline {
             `Chunk ${chunk.chunkKey} text is ${embeddingText.length} chars (~${Math.round(embeddingText.length / 4)} tokens), which may exceed the 512-token model limit and be silently truncated.`
           );
         }
+        const cappedText = embeddingText.length > CHUNK_TEXT_MAX_CHARS ? embeddingText.slice(0, CHUNK_TEXT_MAX_CHARS) : embeddingText;
         return {
           id: chunk.chunkKey,
           data: embeddingText,
@@ -21124,7 +21177,7 @@ var IndexPipeline = class _IndexPipeline {
             sectionTitle: chunk.sectionTitle ?? "",
             headingPath: chunk.headingPath.join(" > "),
             snippet: chunk.snippet,
-            chunkText: embeddingText,
+            chunkText: cappedText,
             tags: chunk.tags,
             ordinal: chunk.ordinal,
             contentHash: chunk.contentHash,

package/dist/index.d.cts

@@ -40,7 +40,6 @@ interface PageVectorMetadata {
     keywords: string[];
     summary: string;
     tags: string[];
-    markdown: string;
     routeFile: string;
     routeResolution: string;
     incomingLinks: number;
@@ -121,6 +120,16 @@ declare class UpstashSearchStore {
         metadata: Record<string, unknown>;
     }>, scope: Scope): Promise<void>;
     getPage(url: string, scope: Scope): Promise<PageRecord | null>;
+    /**
+     * Fetch all chunks belonging to a specific page URL, sorted by ordinal.
+     * Used to reconstruct full page markdown from chunk content.
+     */
+    getChunksForPage(url: string, scope: Scope): Promise<Array<{
+        chunkText: string;
+        ordinal: number;
+        sectionTitle: string;
+        headingPath: string[];
+    }>>;
     fetchPageWithVector(url: string, scope: Scope): Promise<{
         metadata: PageVectorMetadata;
         vector: number[];

package/dist/index.d.ts

@@ -40,7 +40,6 @@ interface PageVectorMetadata {
     keywords: string[];
     summary: string;
     tags: string[];
-    markdown: string;
     routeFile: string;
     routeResolution: string;
     incomingLinks: number;
@@ -121,6 +120,16 @@ declare class UpstashSearchStore {
         metadata: Record<string, unknown>;
     }>, scope: Scope): Promise<void>;
     getPage(url: string, scope: Scope): Promise<PageRecord | null>;
+    /**
+     * Fetch all chunks belonging to a specific page URL, sorted by ordinal.
+     * Used to reconstruct full page markdown from chunk content.
+     */
+    getChunksForPage(url: string, scope: Scope): Promise<Array<{
+        chunkText: string;
+        ordinal: number;
+        sectionTitle: string;
+        headingPath: string[];
+    }>>;
     fetchPageWithVector(url: string, scope: Scope): Promise<{
         metadata: PageVectorMetadata;
         vector: number[];

package/dist/index.js

@@ -17262,6 +17262,26 @@ function ensureStateDirs(cwd, stateDir, scope) {
   fs.mkdirSync(statePath, { recursive: true });
   return { statePath };
 }
+function reconstructMarkdownFromChunks(chunks, pageTitle) {
+  if (chunks.length === 0) return "";
+  const parts = [];
+  for (const chunk of chunks) {
+    let text = chunk.chunkText;
+    const prefixWithSection = `${pageTitle} \u2014 ${chunk.sectionTitle}
+
+`;
+    const prefixWithoutSection = `${pageTitle}
+
+`;
+    if (chunk.sectionTitle && text.startsWith(prefixWithSection)) {
+      text = text.slice(prefixWithSection.length);
+    } else if (text.startsWith(prefixWithoutSection)) {
+      text = text.slice(prefixWithoutSection.length);
+    }
+    parts.push(text.trim());
+  }
+  return parts.join("\n\n");
+}
 var UpstashSearchStore = class {
   index;
   pagesNs;
@@ -17641,10 +17661,12 @@ var UpstashSearchStore = class {
       });
       const doc = results[0];
       if (!doc || !doc.metadata) return null;
+      const chunks = await this.getChunksForPage(url, scope);
+      const markdown = reconstructMarkdownFromChunks(chunks, doc.metadata.title);
       return {
         url: doc.metadata.url,
         title: doc.metadata.title,
-        markdown: doc.metadata.markdown,
+        markdown,
         projectId: doc.metadata.projectId,
         scopeName: doc.metadata.scopeName,
         routeFile: doc.metadata.routeFile,
@@ -17664,6 +17686,37 @@ var UpstashSearchStore = class {
       return null;
     }
   }
+  /**
+   * Fetch all chunks belonging to a specific page URL, sorted by ordinal.
+   * Used to reconstruct full page markdown from chunk content.
+   */
+  async getChunksForPage(url, scope) {
+    const chunks = [];
+    let cursor = "0";
+    try {
+      for (; ; ) {
+        const result = await this.chunksNs.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?.url === url) {
+            chunks.push({
+              chunkText: doc.metadata.chunkText ?? "",
+              ordinal: doc.metadata.ordinal ?? 0,
+              sectionTitle: doc.metadata.sectionTitle ?? "",
+              headingPath: doc.metadata.headingPath ? String(doc.metadata.headingPath).split(" > ").filter(Boolean) : []
+            });
+          }
+        }
+        if (!result.nextCursor || result.nextCursor === "0") break;
+        cursor = result.nextCursor;
+      }
+    } catch {
+    }
+    return chunks.sort((a, b) => a.ordinal - b.ordinal);
+  }
   async fetchPageWithVector(url, scope) {
     try {
       const results = await this.pagesNs.fetch([url], {
@@ -20984,7 +21037,6 @@ var IndexPipeline = class _IndexPipeline {
             keywords: r.keywords ?? [],
             summary: r.summary ?? "",
             tags: r.tags,
-            markdown: r.markdown,
             routeFile: r.routeFile,
             routeResolution: r.routeResolution,
             incomingLinks: r.incomingLinks,
@@ -21011,7 +21063,6 @@ var IndexPipeline = class _IndexPipeline {
               keywords: r.keywords ?? [],
               summary: r.summary ?? "",
               tags: r.tags,
-              markdown: r.markdown,
               routeFile: r.routeFile,
               routeResolution: r.routeResolution,
               incomingLinks: r.incomingLinks,
@@ -21095,6 +21146,7 @@ var IndexPipeline = class _IndexPipeline {
     let documentsUpserted = 0;
     if (!options.dryRun && changedChunks.length > 0) {
       this.logger.info(`Upserting ${changedChunks.length} chunk${changedChunks.length === 1 ? "" : "s"} to Upstash Vector...`);
+      const CHUNK_TEXT_MAX_CHARS = 3e4;
       const docs = changedChunks.map((chunk) => {
         const embeddingText = buildEmbeddingText(chunk, this.config.chunking.prependTitle);
         if (embeddingText.length > 2e3) {
@@ -21102,6 +21154,7 @@ var IndexPipeline = class _IndexPipeline {
             `Chunk ${chunk.chunkKey} text is ${embeddingText.length} chars (~${Math.round(embeddingText.length / 4)} tokens), which may exceed the 512-token model limit and be silently truncated.`
           );
         }
+        const cappedText = embeddingText.length > CHUNK_TEXT_MAX_CHARS ? embeddingText.slice(0, CHUNK_TEXT_MAX_CHARS) : embeddingText;
         return {
           id: chunk.chunkKey,
           data: embeddingText,
@@ -21112,7 +21165,7 @@ var IndexPipeline = class _IndexPipeline {
             sectionTitle: chunk.sectionTitle ?? "",
             headingPath: chunk.headingPath.join(" > "),
             snippet: chunk.snippet,
-            chunkText: embeddingText,
+            chunkText: cappedText,
             tags: chunk.tags,
             ordinal: chunk.ordinal,
             contentHash: chunk.contentHash,

package/dist/sveltekit.cjs

@@ -17305,6 +17305,26 @@ function joinUrl(baseUrl, route) {
   const routePart = ensureLeadingSlash(route);
   return `${base}${routePart}`;
 }
+function reconstructMarkdownFromChunks(chunks, pageTitle) {
+  if (chunks.length === 0) return "";
+  const parts = [];
+  for (const chunk of chunks) {
+    let text = chunk.chunkText;
+    const prefixWithSection = `${pageTitle} \u2014 ${chunk.sectionTitle}
+
+`;
+    const prefixWithoutSection = `${pageTitle}
+
+`;
+    if (chunk.sectionTitle && text.startsWith(prefixWithSection)) {
+      text = text.slice(prefixWithSection.length);
+    } else if (text.startsWith(prefixWithoutSection)) {
+      text = text.slice(prefixWithoutSection.length);
+    }
+    parts.push(text.trim());
+  }
+  return parts.join("\n\n");
+}
 var UpstashSearchStore = class {
   index;
   pagesNs;
@@ -17684,10 +17704,12 @@ var UpstashSearchStore = class {
       });
       const doc = results[0];
       if (!doc || !doc.metadata) return null;
+      const chunks = await this.getChunksForPage(url, scope);
+      const markdown = reconstructMarkdownFromChunks(chunks, doc.metadata.title);
       return {
         url: doc.metadata.url,
         title: doc.metadata.title,
-        markdown: doc.metadata.markdown,
+        markdown,
         projectId: doc.metadata.projectId,
         scopeName: doc.metadata.scopeName,
         routeFile: doc.metadata.routeFile,
@@ -17707,6 +17729,37 @@ var UpstashSearchStore = class {
       return null;
     }
   }
+  /**
+   * Fetch all chunks belonging to a specific page URL, sorted by ordinal.
+   * Used to reconstruct full page markdown from chunk content.
+   */
+  async getChunksForPage(url, scope) {
+    const chunks = [];
+    let cursor = "0";
+    try {
+      for (; ; ) {
+        const result = await this.chunksNs.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?.url === url) {
+            chunks.push({
+              chunkText: doc.metadata.chunkText ?? "",
+              ordinal: doc.metadata.ordinal ?? 0,
+              sectionTitle: doc.metadata.sectionTitle ?? "",
+              headingPath: doc.metadata.headingPath ? String(doc.metadata.headingPath).split(" > ").filter(Boolean) : []
+            });
+          }
+        }
+        if (!result.nextCursor || result.nextCursor === "0") break;
+        cursor = result.nextCursor;
+      }
+    } catch {
+    }
+    return chunks.sort((a, b) => a.ordinal - b.ordinal);
+  }
   async fetchPageWithVector(url, scope) {
     try {
       const results = await this.pagesNs.fetch([url], {
@@ -22138,7 +22191,6 @@ var IndexPipeline = class _IndexPipeline {
             keywords: r.keywords ?? [],
             summary: r.summary ?? "",
             tags: r.tags,
-            markdown: r.markdown,
             routeFile: r.routeFile,
             routeResolution: r.routeResolution,
             incomingLinks: r.incomingLinks,
@@ -22165,7 +22217,6 @@ var IndexPipeline = class _IndexPipeline {
               keywords: r.keywords ?? [],
               summary: r.summary ?? "",
               tags: r.tags,
-              markdown: r.markdown,
               routeFile: r.routeFile,
               routeResolution: r.routeResolution,
               incomingLinks: r.incomingLinks,
@@ -22249,6 +22300,7 @@ var IndexPipeline = class _IndexPipeline {
     let documentsUpserted = 0;
     if (!options.dryRun && changedChunks.length > 0) {
       this.logger.info(`Upserting ${changedChunks.length} chunk${changedChunks.length === 1 ? "" : "s"} to Upstash Vector...`);
+      const CHUNK_TEXT_MAX_CHARS = 3e4;
       const docs = changedChunks.map((chunk) => {
         const embeddingText = buildEmbeddingText(chunk, this.config.chunking.prependTitle);
         if (embeddingText.length > 2e3) {
@@ -22256,6 +22308,7 @@ var IndexPipeline = class _IndexPipeline {
             `Chunk ${chunk.chunkKey} text is ${embeddingText.length} chars (~${Math.round(embeddingText.length / 4)} tokens), which may exceed the 512-token model limit and be silently truncated.`
           );
         }
+        const cappedText = embeddingText.length > CHUNK_TEXT_MAX_CHARS ? embeddingText.slice(0, CHUNK_TEXT_MAX_CHARS) : embeddingText;
         return {
           id: chunk.chunkKey,
           data: embeddingText,
@@ -22266,7 +22319,7 @@ var IndexPipeline = class _IndexPipeline {
             sectionTitle: chunk.sectionTitle ?? "",
             headingPath: chunk.headingPath.join(" > "),
             snippet: chunk.snippet,
-            chunkText: embeddingText,
+            chunkText: cappedText,
             tags: chunk.tags,
             ordinal: chunk.ordinal,
             contentHash: chunk.contentHash,

package/dist/sveltekit.js

@@ -17293,6 +17293,26 @@ function joinUrl(baseUrl, route) {
   const routePart = ensureLeadingSlash(route);
   return `${base}${routePart}`;
 }
+function reconstructMarkdownFromChunks(chunks, pageTitle) {
+  if (chunks.length === 0) return "";
+  const parts = [];
+  for (const chunk of chunks) {
+    let text = chunk.chunkText;
+    const prefixWithSection = `${pageTitle} \u2014 ${chunk.sectionTitle}
+
+`;
+    const prefixWithoutSection = `${pageTitle}
+
+`;
+    if (chunk.sectionTitle && text.startsWith(prefixWithSection)) {
+      text = text.slice(prefixWithSection.length);
+    } else if (text.startsWith(prefixWithoutSection)) {
+      text = text.slice(prefixWithoutSection.length);
+    }
+    parts.push(text.trim());
+  }
+  return parts.join("\n\n");
+}
 var UpstashSearchStore = class {
   index;
   pagesNs;
@@ -17672,10 +17692,12 @@ var UpstashSearchStore = class {
       });
       const doc = results[0];
       if (!doc || !doc.metadata) return null;
+      const chunks = await this.getChunksForPage(url, scope);
+      const markdown = reconstructMarkdownFromChunks(chunks, doc.metadata.title);
       return {
         url: doc.metadata.url,
         title: doc.metadata.title,
-        markdown: doc.metadata.markdown,
+        markdown,
         projectId: doc.metadata.projectId,
         scopeName: doc.metadata.scopeName,
         routeFile: doc.metadata.routeFile,
@@ -17695,6 +17717,37 @@ var UpstashSearchStore = class {
       return null;
     }
   }
+  /**
+   * Fetch all chunks belonging to a specific page URL, sorted by ordinal.
+   * Used to reconstruct full page markdown from chunk content.
+   */
+  async getChunksForPage(url, scope) {
+    const chunks = [];
+    let cursor = "0";
+    try {
+      for (; ; ) {
+        const result = await this.chunksNs.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?.url === url) {
+            chunks.push({
+              chunkText: doc.metadata.chunkText ?? "",
+              ordinal: doc.metadata.ordinal ?? 0,
+              sectionTitle: doc.metadata.sectionTitle ?? "",
+              headingPath: doc.metadata.headingPath ? String(doc.metadata.headingPath).split(" > ").filter(Boolean) : []
+            });
+          }
+        }
+        if (!result.nextCursor || result.nextCursor === "0") break;
+        cursor = result.nextCursor;
+      }
+    } catch {
+    }
+    return chunks.sort((a, b) => a.ordinal - b.ordinal);
+  }
   async fetchPageWithVector(url, scope) {
     try {
       const results = await this.pagesNs.fetch([url], {
@@ -22126,7 +22179,6 @@ var IndexPipeline = class _IndexPipeline {
             keywords: r.keywords ?? [],
             summary: r.summary ?? "",
             tags: r.tags,
-            markdown: r.markdown,
             routeFile: r.routeFile,
             routeResolution: r.routeResolution,
             incomingLinks: r.incomingLinks,
@@ -22153,7 +22205,6 @@ var IndexPipeline = class _IndexPipeline {
               keywords: r.keywords ?? [],
               summary: r.summary ?? "",
               tags: r.tags,
-              markdown: r.markdown,
               routeFile: r.routeFile,
               routeResolution: r.routeResolution,
               incomingLinks: r.incomingLinks,
@@ -22237,6 +22288,7 @@ var IndexPipeline = class _IndexPipeline {
     let documentsUpserted = 0;
     if (!options.dryRun && changedChunks.length > 0) {
       this.logger.info(`Upserting ${changedChunks.length} chunk${changedChunks.length === 1 ? "" : "s"} to Upstash Vector...`);
+      const CHUNK_TEXT_MAX_CHARS = 3e4;
       const docs = changedChunks.map((chunk) => {
         const embeddingText = buildEmbeddingText(chunk, this.config.chunking.prependTitle);
         if (embeddingText.length > 2e3) {
@@ -22244,6 +22296,7 @@ var IndexPipeline = class _IndexPipeline {
             `Chunk ${chunk.chunkKey} text is ${embeddingText.length} chars (~${Math.round(embeddingText.length / 4)} tokens), which may exceed the 512-token model limit and be silently truncated.`
           );
         }
+        const cappedText = embeddingText.length > CHUNK_TEXT_MAX_CHARS ? embeddingText.slice(0, CHUNK_TEXT_MAX_CHARS) : embeddingText;
         return {
           id: chunk.chunkKey,
           data: embeddingText,
@@ -22254,7 +22307,7 @@ var IndexPipeline = class _IndexPipeline {
             sectionTitle: chunk.sectionTitle ?? "",
             headingPath: chunk.headingPath.join(" > "),
             snippet: chunk.snippet,
-            chunkText: embeddingText,
+            chunkText: cappedText,
             tags: chunk.tags,
             ordinal: chunk.ordinal,
             contentHash: chunk.contentHash,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "searchsocket",
-  "version": "0.6.2",
+  "version": "0.6.3",
   "description": "Semantic site search and MCP retrieval for SvelteKit static sites",
   "license": "MIT",
   "author": "Greg Priday <greg@siteorigin.com>",