searchsocket 0.3.0 → 0.3.2

package/dist/cli.js

@@ -12,7 +12,7 @@ import { Command } from "commander";
 // package.json
 var package_default = {
   name: "searchsocket",
-  version: "0.3.0",
+  version: "0.3.2",
   description: "Semantic site search and MCP retrieval for SvelteKit static sites",
   license: "MIT",
   author: "Greg Priday <greg@siteorigin.com>",
@@ -189,7 +189,7 @@ var searchSocketConfigSchema = z.object({
   ranking: z.object({
     enableIncomingLinkBoost: z.boolean().optional(),
     enableDepthBoost: z.boolean().optional(),
-    pageWeights: z.record(z.string(), z.number().positive()).optional(),
+    pageWeights: z.record(z.string(), z.number().min(0)).optional(),
     aggregationCap: z.number().int().positive().optional(),
     aggregationDecay: z.number().min(0).max(1).optional(),
     minChunkScoreRatio: z.number().min(0).max(1).optional(),
@@ -913,23 +913,12 @@ var TursoVectorStore = class {
         incoming_links INTEGER NOT NULL DEFAULT 0,
         route_file     TEXT NOT NULL DEFAULT '',
         tags           TEXT NOT NULL DEFAULT '[]',
+        description    TEXT NOT NULL DEFAULT '',
+        keywords       TEXT NOT NULL DEFAULT '[]',
         embedding      F32_BLOB(${dim})
       )`,
       `CREATE INDEX IF NOT EXISTS idx ON chunks (libsql_vector_idx(embedding, 'metric=cosine'))`
     ]);
-    const chunkMigrationCols = [
-      { name: "chunk_text", def: "TEXT NOT NULL DEFAULT ''" },
-      { name: "ordinal", def: "INTEGER NOT NULL DEFAULT 0" }
-    ];
-    for (const col of chunkMigrationCols) {
-      try {
-        await this.client.execute(`ALTER TABLE chunks ADD COLUMN ${col.name} ${col.def}`);
-      } catch (error) {
-        if (error instanceof Error && !error.message.includes("duplicate column")) {
-          throw error;
-        }
-      }
-    }
     this.chunksReady = true;
   }
   async ensurePages() {
@@ -1007,8 +996,8 @@ var TursoVectorStore = class {
         sql: `INSERT OR REPLACE INTO chunks
               (id, project_id, scope_name, url, path, title, section_title,
                heading_path, snippet, chunk_text, ordinal, content_hash, model_id, depth,
-               incoming_links, route_file, tags, embedding)
-              VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, vector(?))`,
+               incoming_links, route_file, tags, description, keywords, embedding)
+              VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, vector(?))`,
         args: [
           r.id,
           r.metadata.projectId,
@@ -1027,6 +1016,8 @@ var TursoVectorStore = class {
           r.metadata.incomingLinks,
           r.metadata.routeFile,
           JSON.stringify(r.metadata.tags),
+          r.metadata.description ?? "",
+          JSON.stringify(r.metadata.keywords ?? []),
           JSON.stringify(r.vector)
         ]
       }));
@@ -1042,6 +1033,7 @@ var TursoVectorStore = class {
                    c.section_title, c.heading_path, c.snippet, c.chunk_text,
                    c.ordinal, c.content_hash,
                    c.model_id, c.depth, c.incoming_links, c.route_file, c.tags,
+                   c.description, c.keywords,
                    vector_distance_cos(c.embedding, vector(?)) AS distance
             FROM vector_top_k('idx', vector(?), ?) AS v
             JOIN chunks AS c ON c.rowid = v.id`,
@@ -1072,6 +1064,12 @@ var TursoVectorStore = class {
       }
       const distance = row.distance;
       const score = 1 - distance;
+      const description = row.description || void 0;
+      const keywords = (() => {
+        const raw = row.keywords || "[]";
+        const parsed = JSON.parse(raw);
+        return parsed.length > 0 ? parsed : void 0;
+      })();
       hits.push({
         id: row.id,
         score,
@@ -1091,7 +1089,9 @@ var TursoVectorStore = class {
           depth: row.depth,
           incomingLinks: row.incoming_links,
           routeFile: row.route_file,
-          tags
+          tags,
+          description,
+          keywords
         }
       });
     }
@@ -1636,7 +1636,9 @@ function chunkMirrorPage(page, config, scope) {
       incomingLinks: page.incomingLinks,
       routeFile: page.routeFile,
       tags: page.tags,
-      contentHash: ""
+      contentHash: "",
+      description: page.description,
+      keywords: page.keywords
     };
     const embeddingText = buildEmbeddingText(summaryChunk, config.chunking.prependTitle);
     summaryChunk.contentHash = sha256(normalizeText(embeddingText));
@@ -1663,7 +1665,9 @@ function chunkMirrorPage(page, config, scope) {
       incomingLinks: page.incomingLinks,
       routeFile: page.routeFile,
       tags: page.tags,
-      contentHash: ""
+      contentHash: "",
+      description: page.description,
+      keywords: page.keywords
     };
     const embeddingText = buildEmbeddingText(chunk, config.chunking.prependTitle);
     chunk.contentHash = sha256(normalizeText(embeddingText));
@@ -1696,7 +1700,7 @@ function extractFromHtml(url, html, config) {
   const $ = load(html);
   const normalizedUrl = normalizeUrlPath(url);
   const pageBaseUrl = new URL(`https://searchsocket.local${normalizedUrl}`);
-  const title = normalizeText($("title").first().text() || "") || normalizeText($(`${config.extract.mainSelector} h1`).first().text() || "") || normalizedUrl;
+  const title = $("meta[property='og:title']").attr("content")?.trim() || normalizeText($(`${config.extract.mainSelector} h1`).first().text() || "") || $("meta[name='twitter:title']").attr("content")?.trim() || normalizeText($("title").first().text() || "") || normalizedUrl;
   if (config.extract.respectRobotsNoindex) {
     const robots = $("meta[name='robots']").attr("content") ?? "";
     if (/\bnoindex\b/i.test(robots)) {
@@ -2774,7 +2778,9 @@ var IndexPipeline = class _IndexPipeline {
             depth: chunk.depth,
             incomingLinks: chunk.incomingLinks,
             routeFile: chunk.routeFile,
-            tags: chunk.tags
+            tags: chunk.tags,
+            description: chunk.description,
+            keywords: chunk.keywords
           }
         });
       }
@@ -3004,6 +3010,7 @@ function aggregateByPage(ranked, config) {
     }
     let pageScore = maxScore + aggregationBonus * config.ranking.weights.aggregation;
     const pageWeight = findPageWeight(url, config.ranking.pageWeights);
+    if (pageWeight === 0) continue;
     if (pageWeight !== 1) {
       pageScore *= pageWeight;
     }
@@ -3226,16 +3233,29 @@ var SearchEngine = class _SearchEngine {
       if (group) group.push(entry);
       else pageGroups.set(url, [entry]);
     }
+    const MAX_CHUNKS_PER_PAGE = 5;
+    const MIN_CHUNKS_PER_PAGE = 1;
+    const MIN_CHUNK_SCORE_RATIO = 0.5;
     const pageCandidates = [];
     for (const [url, chunks] of pageGroups) {
-      const sorted = [...chunks].sort(
-        (a, b) => (a.hit.metadata.ordinal ?? 0) - (b.hit.metadata.ordinal ?? 0)
-      );
-      const title = sorted[0].hit.metadata.title;
-      const body = sorted.map((c) => c.hit.metadata.chunkText || c.hit.metadata.snippet).join("\n\n");
-      pageCandidates.push({ id: url, text: `${title}
-
-${body}` });
+      const byScore = [...chunks].sort((a, b) => b.finalScore - a.finalScore);
+      const bestScore = byScore[0].finalScore;
+      const scoreFloor = Number.isFinite(bestScore) ? bestScore * MIN_CHUNK_SCORE_RATIO : Number.NEGATIVE_INFINITY;
+      const selected = byScore.filter(
+        (c, i) => i < MIN_CHUNKS_PER_PAGE || c.finalScore >= scoreFloor
+      ).slice(0, MAX_CHUNKS_PER_PAGE);
+      selected.sort((a, b) => (a.hit.metadata.ordinal ?? 0) - (b.hit.metadata.ordinal ?? 0));
+      const first = selected[0].hit.metadata;
+      const parts = [first.title];
+      if (first.description) {
+        parts.push(first.description);
+      }
+      if (first.keywords && first.keywords.length > 0) {
+        parts.push(first.keywords.join(", "));
+      }
+      const body = selected.map((c) => c.hit.metadata.chunkText || c.hit.metadata.snippet).join("\n\n");
+      parts.push(body);
+      pageCandidates.push({ id: url, text: parts.join("\n\n") });
     }
     const reranked = await this.reranker.rerank(
       query,

package/dist/client.d.cts

@@ -1,4 +1,4 @@
-import { S as SearchRequest, a as SearchResponse } from './types-DAXk6A3Y.cjs';
+import { S as SearchRequest, a as SearchResponse } from './types-BrG6XTUU.cjs';
 
 interface SearchClientOptions {
     endpoint?: string;

package/dist/client.d.ts

@@ -1,4 +1,4 @@
-import { S as SearchRequest, a as SearchResponse } from './types-DAXk6A3Y.js';
+import { S as SearchRequest, a as SearchResponse } from './types-BrG6XTUU.js';
 
 interface SearchClientOptions {
     endpoint?: string;

package/dist/index.cjs

@@ -16688,7 +16688,7 @@ var searchSocketConfigSchema = zod.z.object({
   ranking: zod.z.object({
     enableIncomingLinkBoost: zod.z.boolean().optional(),
     enableDepthBoost: zod.z.boolean().optional(),
-    pageWeights: zod.z.record(zod.z.string(), zod.z.number().positive()).optional(),
+    pageWeights: zod.z.record(zod.z.string(), zod.z.number().min(0)).optional(),
     aggregationCap: zod.z.number().int().positive().optional(),
     aggregationDecay: zod.z.number().min(0).max(1).optional(),
     minChunkScoreRatio: zod.z.number().min(0).max(1).optional(),
@@ -17435,23 +17435,12 @@ var TursoVectorStore = class {
         incoming_links INTEGER NOT NULL DEFAULT 0,
         route_file     TEXT NOT NULL DEFAULT '',
         tags           TEXT NOT NULL DEFAULT '[]',
+        description    TEXT NOT NULL DEFAULT '',
+        keywords       TEXT NOT NULL DEFAULT '[]',
         embedding      F32_BLOB(${dim})
       )`,
       `CREATE INDEX IF NOT EXISTS idx ON chunks (libsql_vector_idx(embedding, 'metric=cosine'))`
     ]);
-    const chunkMigrationCols = [
-      { name: "chunk_text", def: "TEXT NOT NULL DEFAULT ''" },
-      { name: "ordinal", def: "INTEGER NOT NULL DEFAULT 0" }
-    ];
-    for (const col of chunkMigrationCols) {
-      try {
-        await this.client.execute(`ALTER TABLE chunks ADD COLUMN ${col.name} ${col.def}`);
-      } catch (error) {
-        if (error instanceof Error && !error.message.includes("duplicate column")) {
-          throw error;
-        }
-      }
-    }
     this.chunksReady = true;
   }
   async ensurePages() {
@@ -17529,8 +17518,8 @@ var TursoVectorStore = class {
         sql: `INSERT OR REPLACE INTO chunks
               (id, project_id, scope_name, url, path, title, section_title,
                heading_path, snippet, chunk_text, ordinal, content_hash, model_id, depth,
-               incoming_links, route_file, tags, embedding)
-              VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, vector(?))`,
+               incoming_links, route_file, tags, description, keywords, embedding)
+              VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, vector(?))`,
         args: [
           r.id,
           r.metadata.projectId,
@@ -17549,6 +17538,8 @@ var TursoVectorStore = class {
           r.metadata.incomingLinks,
           r.metadata.routeFile,
           JSON.stringify(r.metadata.tags),
+          r.metadata.description ?? "",
+          JSON.stringify(r.metadata.keywords ?? []),
           JSON.stringify(r.vector)
         ]
       }));
@@ -17564,6 +17555,7 @@ var TursoVectorStore = class {
                    c.section_title, c.heading_path, c.snippet, c.chunk_text,
                    c.ordinal, c.content_hash,
                    c.model_id, c.depth, c.incoming_links, c.route_file, c.tags,
+                   c.description, c.keywords,
                    vector_distance_cos(c.embedding, vector(?)) AS distance
             FROM vector_top_k('idx', vector(?), ?) AS v
             JOIN chunks AS c ON c.rowid = v.id`,
@@ -17594,6 +17586,12 @@ var TursoVectorStore = class {
       }
       const distance = row.distance;
       const score = 1 - distance;
+      const description = row.description || void 0;
+      const keywords = (() => {
+        const raw = row.keywords || "[]";
+        const parsed = JSON.parse(raw);
+        return parsed.length > 0 ? parsed : void 0;
+      })();
       hits.push({
         id: row.id,
         score,
@@ -17613,7 +17611,9 @@ var TursoVectorStore = class {
           depth: row.depth,
           incomingLinks: row.incoming_links,
           routeFile: row.route_file,
-          tags
+          tags,
+          description,
+          keywords
         }
       });
     }
@@ -18152,7 +18152,9 @@ function chunkMirrorPage(page, config, scope) {
       incomingLinks: page.incomingLinks,
       routeFile: page.routeFile,
       tags: page.tags,
-      contentHash: ""
+      contentHash: "",
+      description: page.description,
+      keywords: page.keywords
     };
     const embeddingText = buildEmbeddingText(summaryChunk, config.chunking.prependTitle);
     summaryChunk.contentHash = sha256(normalizeText(embeddingText));
@@ -18179,7 +18181,9 @@ function chunkMirrorPage(page, config, scope) {
       incomingLinks: page.incomingLinks,
       routeFile: page.routeFile,
       tags: page.tags,
-      contentHash: ""
+      contentHash: "",
+      description: page.description,
+      keywords: page.keywords
     };
     const embeddingText = buildEmbeddingText(chunk, config.chunking.prependTitle);
     chunk.contentHash = sha256(normalizeText(embeddingText));
@@ -19034,7 +19038,7 @@ function extractFromHtml(url, html, config) {
   const $ = cheerio.load(html);
   const normalizedUrl = normalizeUrlPath(url);
   const pageBaseUrl = new URL(`https://searchsocket.local${normalizedUrl}`);
-  const title = normalizeText($("title").first().text() || "") || normalizeText($(`${config.extract.mainSelector} h1`).first().text() || "") || normalizedUrl;
+  const title = $("meta[property='og:title']").attr("content")?.trim() || normalizeText($(`${config.extract.mainSelector} h1`).first().text() || "") || $("meta[name='twitter:title']").attr("content")?.trim() || normalizeText($("title").first().text() || "") || normalizedUrl;
   if (config.extract.respectRobotsNoindex) {
     const robots = $("meta[name='robots']").attr("content") ?? "";
     if (/\bnoindex\b/i.test(robots)) {
@@ -20150,7 +20154,9 @@ var IndexPipeline = class _IndexPipeline {
             depth: chunk.depth,
             incomingLinks: chunk.incomingLinks,
             routeFile: chunk.routeFile,
-            tags: chunk.tags
+            tags: chunk.tags,
+            description: chunk.description,
+            keywords: chunk.keywords
           }
         });
       }
@@ -20273,6 +20279,7 @@ function aggregateByPage(ranked, config) {
     }
     let pageScore = maxScore + aggregationBonus * config.ranking.weights.aggregation;
     const pageWeight = findPageWeight(url, config.ranking.pageWeights);
+    if (pageWeight === 0) continue;
     if (pageWeight !== 1) {
       pageScore *= pageWeight;
     }
@@ -20495,16 +20502,29 @@ var SearchEngine = class _SearchEngine {
       if (group) group.push(entry);
       else pageGroups.set(url, [entry]);
     }
+    const MAX_CHUNKS_PER_PAGE = 5;
+    const MIN_CHUNKS_PER_PAGE = 1;
+    const MIN_CHUNK_SCORE_RATIO = 0.5;
     const pageCandidates = [];
     for (const [url, chunks] of pageGroups) {
-      const sorted = [...chunks].sort(
-        (a, b) => (a.hit.metadata.ordinal ?? 0) - (b.hit.metadata.ordinal ?? 0)
-      );
-      const title = sorted[0].hit.metadata.title;
-      const body = sorted.map((c) => c.hit.metadata.chunkText || c.hit.metadata.snippet).join("\n\n");
-      pageCandidates.push({ id: url, text: `${title}
-
-${body}` });
+      const byScore = [...chunks].sort((a, b) => b.finalScore - a.finalScore);
+      const bestScore = byScore[0].finalScore;
+      const scoreFloor = Number.isFinite(bestScore) ? bestScore * MIN_CHUNK_SCORE_RATIO : Number.NEGATIVE_INFINITY;
+      const selected = byScore.filter(
+        (c, i) => i < MIN_CHUNKS_PER_PAGE || c.finalScore >= scoreFloor
+      ).slice(0, MAX_CHUNKS_PER_PAGE);
+      selected.sort((a, b) => (a.hit.metadata.ordinal ?? 0) - (b.hit.metadata.ordinal ?? 0));
+      const first = selected[0].hit.metadata;
+      const parts = [first.title];
+      if (first.description) {
+        parts.push(first.description);
+      }
+      if (first.keywords && first.keywords.length > 0) {
+        parts.push(first.keywords.join(", "));
+      }
+      const body = selected.map((c) => c.hit.metadata.chunkText || c.hit.metadata.snippet).join("\n\n");
+      parts.push(body);
+      pageCandidates.push({ id: url, text: parts.join("\n\n") });
     }
     const reranked = await this.reranker.rerank(
       query,

package/dist/index.d.cts

@@ -1,5 +1,5 @@
-import { R as ResolvedSearchSocketConfig, b as SearchSocketConfig, c as Scope, E as EmbeddingsProvider, d as Reranker, e as RerankCandidate, V as VectorStore, I as IndexOptions, f as IndexStats, S as SearchRequest, a as SearchResponse } from './types-DAXk6A3Y.cjs';
-export { C as Chunk, Q as QueryOpts, g as VectorHit, h as VectorRecord } from './types-DAXk6A3Y.cjs';
+import { R as ResolvedSearchSocketConfig, b as SearchSocketConfig, c as Scope, E as EmbeddingsProvider, d as Reranker, e as RerankCandidate, V as VectorStore, I as IndexOptions, f as IndexStats, S as SearchRequest, a as SearchResponse } from './types-BrG6XTUU.cjs';
+export { C as Chunk, Q as QueryOpts, g as VectorHit, h as VectorRecord } from './types-BrG6XTUU.cjs';
 export { searchsocketHandle, searchsocketVitePlugin } from './sveltekit.cjs';
 export { createSearchClient } from './client.cjs';
 

package/dist/index.d.ts

@@ -1,5 +1,5 @@
-import { R as ResolvedSearchSocketConfig, b as SearchSocketConfig, c as Scope, E as EmbeddingsProvider, d as Reranker, e as RerankCandidate, V as VectorStore, I as IndexOptions, f as IndexStats, S as SearchRequest, a as SearchResponse } from './types-DAXk6A3Y.js';
-export { C as Chunk, Q as QueryOpts, g as VectorHit, h as VectorRecord } from './types-DAXk6A3Y.js';
+import { R as ResolvedSearchSocketConfig, b as SearchSocketConfig, c as Scope, E as EmbeddingsProvider, d as Reranker, e as RerankCandidate, V as VectorStore, I as IndexOptions, f as IndexStats, S as SearchRequest, a as SearchResponse } from './types-BrG6XTUU.js';
+export { C as Chunk, Q as QueryOpts, g as VectorHit, h as VectorRecord } from './types-BrG6XTUU.js';
 export { searchsocketHandle, searchsocketVitePlugin } from './sveltekit.js';
 export { createSearchClient } from './client.js';
 

package/dist/index.js

@@ -16676,7 +16676,7 @@ var searchSocketConfigSchema = z.object({
   ranking: z.object({
     enableIncomingLinkBoost: z.boolean().optional(),
     enableDepthBoost: z.boolean().optional(),
-    pageWeights: z.record(z.string(), z.number().positive()).optional(),
+    pageWeights: z.record(z.string(), z.number().min(0)).optional(),
     aggregationCap: z.number().int().positive().optional(),
     aggregationDecay: z.number().min(0).max(1).optional(),
     minChunkScoreRatio: z.number().min(0).max(1).optional(),
@@ -17423,23 +17423,12 @@ var TursoVectorStore = class {
         incoming_links INTEGER NOT NULL DEFAULT 0,
         route_file     TEXT NOT NULL DEFAULT '',
         tags           TEXT NOT NULL DEFAULT '[]',
+        description    TEXT NOT NULL DEFAULT '',
+        keywords       TEXT NOT NULL DEFAULT '[]',
         embedding      F32_BLOB(${dim})
       )`,
       `CREATE INDEX IF NOT EXISTS idx ON chunks (libsql_vector_idx(embedding, 'metric=cosine'))`
     ]);
-    const chunkMigrationCols = [
-      { name: "chunk_text", def: "TEXT NOT NULL DEFAULT ''" },
-      { name: "ordinal", def: "INTEGER NOT NULL DEFAULT 0" }
-    ];
-    for (const col of chunkMigrationCols) {
-      try {
-        await this.client.execute(`ALTER TABLE chunks ADD COLUMN ${col.name} ${col.def}`);
-      } catch (error) {
-        if (error instanceof Error && !error.message.includes("duplicate column")) {
-          throw error;
-        }
-      }
-    }
     this.chunksReady = true;
   }
   async ensurePages() {
@@ -17517,8 +17506,8 @@ var TursoVectorStore = class {
         sql: `INSERT OR REPLACE INTO chunks
               (id, project_id, scope_name, url, path, title, section_title,
                heading_path, snippet, chunk_text, ordinal, content_hash, model_id, depth,
-               incoming_links, route_file, tags, embedding)
-              VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, vector(?))`,
+               incoming_links, route_file, tags, description, keywords, embedding)
+              VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, vector(?))`,
         args: [
           r.id,
           r.metadata.projectId,
@@ -17537,6 +17526,8 @@ var TursoVectorStore = class {
           r.metadata.incomingLinks,
           r.metadata.routeFile,
           JSON.stringify(r.metadata.tags),
+          r.metadata.description ?? "",
+          JSON.stringify(r.metadata.keywords ?? []),
           JSON.stringify(r.vector)
         ]
       }));
@@ -17552,6 +17543,7 @@ var TursoVectorStore = class {
                    c.section_title, c.heading_path, c.snippet, c.chunk_text,
                    c.ordinal, c.content_hash,
                    c.model_id, c.depth, c.incoming_links, c.route_file, c.tags,
+                   c.description, c.keywords,
                    vector_distance_cos(c.embedding, vector(?)) AS distance
             FROM vector_top_k('idx', vector(?), ?) AS v
             JOIN chunks AS c ON c.rowid = v.id`,
@@ -17582,6 +17574,12 @@ var TursoVectorStore = class {
       }
       const distance = row.distance;
       const score = 1 - distance;
+      const description = row.description || void 0;
+      const keywords = (() => {
+        const raw = row.keywords || "[]";
+        const parsed = JSON.parse(raw);
+        return parsed.length > 0 ? parsed : void 0;
+      })();
       hits.push({
         id: row.id,
         score,
@@ -17601,7 +17599,9 @@ var TursoVectorStore = class {
           depth: row.depth,
           incomingLinks: row.incoming_links,
           routeFile: row.route_file,
-          tags
+          tags,
+          description,
+          keywords
         }
       });
     }
@@ -18140,7 +18140,9 @@ function chunkMirrorPage(page, config, scope) {
       incomingLinks: page.incomingLinks,
       routeFile: page.routeFile,
       tags: page.tags,
-      contentHash: ""
+      contentHash: "",
+      description: page.description,
+      keywords: page.keywords
     };
     const embeddingText = buildEmbeddingText(summaryChunk, config.chunking.prependTitle);
     summaryChunk.contentHash = sha256(normalizeText(embeddingText));
@@ -18167,7 +18169,9 @@ function chunkMirrorPage(page, config, scope) {
       incomingLinks: page.incomingLinks,
       routeFile: page.routeFile,
       tags: page.tags,
-      contentHash: ""
+      contentHash: "",
+      description: page.description,
+      keywords: page.keywords
     };
     const embeddingText = buildEmbeddingText(chunk, config.chunking.prependTitle);
     chunk.contentHash = sha256(normalizeText(embeddingText));
@@ -19022,7 +19026,7 @@ function extractFromHtml(url, html, config) {
   const $ = load(html);
   const normalizedUrl = normalizeUrlPath(url);
   const pageBaseUrl = new URL(`https://searchsocket.local${normalizedUrl}`);
-  const title = normalizeText($("title").first().text() || "") || normalizeText($(`${config.extract.mainSelector} h1`).first().text() || "") || normalizedUrl;
+  const title = $("meta[property='og:title']").attr("content")?.trim() || normalizeText($(`${config.extract.mainSelector} h1`).first().text() || "") || $("meta[name='twitter:title']").attr("content")?.trim() || normalizeText($("title").first().text() || "") || normalizedUrl;
   if (config.extract.respectRobotsNoindex) {
     const robots = $("meta[name='robots']").attr("content") ?? "";
     if (/\bnoindex\b/i.test(robots)) {
@@ -20138,7 +20142,9 @@ var IndexPipeline = class _IndexPipeline {
             depth: chunk.depth,
             incomingLinks: chunk.incomingLinks,
             routeFile: chunk.routeFile,
-            tags: chunk.tags
+            tags: chunk.tags,
+            description: chunk.description,
+            keywords: chunk.keywords
           }
         });
       }
@@ -20261,6 +20267,7 @@ function aggregateByPage(ranked, config) {
     }
     let pageScore = maxScore + aggregationBonus * config.ranking.weights.aggregation;
     const pageWeight = findPageWeight(url, config.ranking.pageWeights);
+    if (pageWeight === 0) continue;
     if (pageWeight !== 1) {
       pageScore *= pageWeight;
     }
@@ -20483,16 +20490,29 @@ var SearchEngine = class _SearchEngine {
       if (group) group.push(entry);
       else pageGroups.set(url, [entry]);
     }
+    const MAX_CHUNKS_PER_PAGE = 5;
+    const MIN_CHUNKS_PER_PAGE = 1;
+    const MIN_CHUNK_SCORE_RATIO = 0.5;
     const pageCandidates = [];
     for (const [url, chunks] of pageGroups) {
-      const sorted = [...chunks].sort(
-        (a, b) => (a.hit.metadata.ordinal ?? 0) - (b.hit.metadata.ordinal ?? 0)
-      );
-      const title = sorted[0].hit.metadata.title;
-      const body = sorted.map((c) => c.hit.metadata.chunkText || c.hit.metadata.snippet).join("\n\n");
-      pageCandidates.push({ id: url, text: `${title}
-
-${body}` });
+      const byScore = [...chunks].sort((a, b) => b.finalScore - a.finalScore);
+      const bestScore = byScore[0].finalScore;
+      const scoreFloor = Number.isFinite(bestScore) ? bestScore * MIN_CHUNK_SCORE_RATIO : Number.NEGATIVE_INFINITY;
+      const selected = byScore.filter(
+        (c, i) => i < MIN_CHUNKS_PER_PAGE || c.finalScore >= scoreFloor
+      ).slice(0, MAX_CHUNKS_PER_PAGE);
+      selected.sort((a, b) => (a.hit.metadata.ordinal ?? 0) - (b.hit.metadata.ordinal ?? 0));
+      const first = selected[0].hit.metadata;
+      const parts = [first.title];
+      if (first.description) {
+        parts.push(first.description);
+      }
+      if (first.keywords && first.keywords.length > 0) {
+        parts.push(first.keywords.join(", "));
+      }
+      const body = selected.map((c) => c.hit.metadata.chunkText || c.hit.metadata.snippet).join("\n\n");
+      parts.push(body);
+      pageCandidates.push({ id: url, text: parts.join("\n\n") });
     }
     const reranked = await this.reranker.rerank(
       query,

package/dist/sveltekit.cjs

@@ -16684,7 +16684,7 @@ var searchSocketConfigSchema = zod.z.object({
   ranking: zod.z.object({
     enableIncomingLinkBoost: zod.z.boolean().optional(),
     enableDepthBoost: zod.z.boolean().optional(),
-    pageWeights: zod.z.record(zod.z.string(), zod.z.number().positive()).optional(),
+    pageWeights: zod.z.record(zod.z.string(), zod.z.number().min(0)).optional(),
     aggregationCap: zod.z.number().int().positive().optional(),
     aggregationDecay: zod.z.number().min(0).max(1).optional(),
     minChunkScoreRatio: zod.z.number().min(0).max(1).optional(),
@@ -17468,23 +17468,12 @@ var TursoVectorStore = class {
         incoming_links INTEGER NOT NULL DEFAULT 0,
         route_file     TEXT NOT NULL DEFAULT '',
         tags           TEXT NOT NULL DEFAULT '[]',
+        description    TEXT NOT NULL DEFAULT '',
+        keywords       TEXT NOT NULL DEFAULT '[]',
         embedding      F32_BLOB(${dim})
       )`,
       `CREATE INDEX IF NOT EXISTS idx ON chunks (libsql_vector_idx(embedding, 'metric=cosine'))`
     ]);
-    const chunkMigrationCols = [
-      { name: "chunk_text", def: "TEXT NOT NULL DEFAULT ''" },
-      { name: "ordinal", def: "INTEGER NOT NULL DEFAULT 0" }
-    ];
-    for (const col of chunkMigrationCols) {
-      try {
-        await this.client.execute(`ALTER TABLE chunks ADD COLUMN ${col.name} ${col.def}`);
-      } catch (error) {
-        if (error instanceof Error && !error.message.includes("duplicate column")) {
-          throw error;
-        }
-      }
-    }
     this.chunksReady = true;
   }
   async ensurePages() {
@@ -17562,8 +17551,8 @@ var TursoVectorStore = class {
         sql: `INSERT OR REPLACE INTO chunks
               (id, project_id, scope_name, url, path, title, section_title,
                heading_path, snippet, chunk_text, ordinal, content_hash, model_id, depth,
-               incoming_links, route_file, tags, embedding)
-              VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, vector(?))`,
+               incoming_links, route_file, tags, description, keywords, embedding)
+              VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, vector(?))`,
         args: [
           r.id,
           r.metadata.projectId,
@@ -17582,6 +17571,8 @@ var TursoVectorStore = class {
           r.metadata.incomingLinks,
           r.metadata.routeFile,
           JSON.stringify(r.metadata.tags),
+          r.metadata.description ?? "",
+          JSON.stringify(r.metadata.keywords ?? []),
           JSON.stringify(r.vector)
         ]
       }));
@@ -17597,6 +17588,7 @@ var TursoVectorStore = class {
                    c.section_title, c.heading_path, c.snippet, c.chunk_text,
                    c.ordinal, c.content_hash,
                    c.model_id, c.depth, c.incoming_links, c.route_file, c.tags,
+                   c.description, c.keywords,
                    vector_distance_cos(c.embedding, vector(?)) AS distance
             FROM vector_top_k('idx', vector(?), ?) AS v
             JOIN chunks AS c ON c.rowid = v.id`,
@@ -17627,6 +17619,12 @@ var TursoVectorStore = class {
       }
       const distance = row.distance;
       const score = 1 - distance;
+      const description = row.description || void 0;
+      const keywords = (() => {
+        const raw = row.keywords || "[]";
+        const parsed = JSON.parse(raw);
+        return parsed.length > 0 ? parsed : void 0;
+      })();
       hits.push({
         id: row.id,
         score,
@@ -17646,7 +17644,9 @@ var TursoVectorStore = class {
           depth: row.depth,
           incomingLinks: row.incoming_links,
           routeFile: row.route_file,
-          tags
+          tags,
+          description,
+          keywords
         }
       });
     }
@@ -17940,6 +17940,7 @@ function aggregateByPage(ranked, config) {
     }
     let pageScore = maxScore + aggregationBonus * config.ranking.weights.aggregation;
     const pageWeight = findPageWeight(url, config.ranking.pageWeights);
+    if (pageWeight === 0) continue;
     if (pageWeight !== 1) {
       pageScore *= pageWeight;
     }
@@ -18162,16 +18163,29 @@ var SearchEngine = class _SearchEngine {
       if (group) group.push(entry);
       else pageGroups.set(url, [entry]);
     }
+    const MAX_CHUNKS_PER_PAGE = 5;
+    const MIN_CHUNKS_PER_PAGE = 1;
+    const MIN_CHUNK_SCORE_RATIO = 0.5;
     const pageCandidates = [];
     for (const [url, chunks] of pageGroups) {
-      const sorted = [...chunks].sort(
-        (a, b) => (a.hit.metadata.ordinal ?? 0) - (b.hit.metadata.ordinal ?? 0)
-      );
-      const title = sorted[0].hit.metadata.title;
-      const body = sorted.map((c) => c.hit.metadata.chunkText || c.hit.metadata.snippet).join("\n\n");
-      pageCandidates.push({ id: url, text: `${title}
-
-${body}` });
+      const byScore = [...chunks].sort((a, b) => b.finalScore - a.finalScore);
+      const bestScore = byScore[0].finalScore;
+      const scoreFloor = Number.isFinite(bestScore) ? bestScore * MIN_CHUNK_SCORE_RATIO : Number.NEGATIVE_INFINITY;
+      const selected = byScore.filter(
+        (c, i) => i < MIN_CHUNKS_PER_PAGE || c.finalScore >= scoreFloor
+      ).slice(0, MAX_CHUNKS_PER_PAGE);
+      selected.sort((a, b) => (a.hit.metadata.ordinal ?? 0) - (b.hit.metadata.ordinal ?? 0));
+      const first = selected[0].hit.metadata;
+      const parts = [first.title];
+      if (first.description) {
+        parts.push(first.description);
+      }
+      if (first.keywords && first.keywords.length > 0) {
+        parts.push(first.keywords.join(", "));
+      }
+      const body = selected.map((c) => c.hit.metadata.chunkText || c.hit.metadata.snippet).join("\n\n");
+      parts.push(body);
+      pageCandidates.push({ id: url, text: parts.join("\n\n") });
     }
     const reranked = await this.reranker.rerank(
       query,
@@ -18681,7 +18695,9 @@ function chunkMirrorPage(page, config, scope) {
       incomingLinks: page.incomingLinks,
       routeFile: page.routeFile,
       tags: page.tags,
-      contentHash: ""
+      contentHash: "",
+      description: page.description,
+      keywords: page.keywords
     };
     const embeddingText = buildEmbeddingText(summaryChunk, config.chunking.prependTitle);
     summaryChunk.contentHash = sha256(normalizeText(embeddingText));
@@ -18708,7 +18724,9 @@ function chunkMirrorPage(page, config, scope) {
       incomingLinks: page.incomingLinks,
       routeFile: page.routeFile,
       tags: page.tags,
-      contentHash: ""
+      contentHash: "",
+      description: page.description,
+      keywords: page.keywords
     };
     const embeddingText = buildEmbeddingText(chunk, config.chunking.prependTitle);
     chunk.contentHash = sha256(normalizeText(embeddingText));
@@ -19563,7 +19581,7 @@ function extractFromHtml(url, html, config) {
   const $ = cheerio.load(html);
   const normalizedUrl = normalizeUrlPath(url);
   const pageBaseUrl = new URL(`https://searchsocket.local${normalizedUrl}`);
-  const title = normalizeText($("title").first().text() || "") || normalizeText($(`${config.extract.mainSelector} h1`).first().text() || "") || normalizedUrl;
+  const title = $("meta[property='og:title']").attr("content")?.trim() || normalizeText($(`${config.extract.mainSelector} h1`).first().text() || "") || $("meta[name='twitter:title']").attr("content")?.trim() || normalizeText($("title").first().text() || "") || normalizedUrl;
   if (config.extract.respectRobotsNoindex) {
     const robots = $("meta[name='robots']").attr("content") ?? "";
     if (/\bnoindex\b/i.test(robots)) {
@@ -20671,7 +20689,9 @@ var IndexPipeline = class _IndexPipeline {
             depth: chunk.depth,
             incomingLinks: chunk.incomingLinks,
             routeFile: chunk.routeFile,
-            tags: chunk.tags
+            tags: chunk.tags,
+            description: chunk.description,
+            keywords: chunk.keywords
           }
         });
       }

package/dist/sveltekit.d.cts

@@ -1,4 +1,4 @@
-import { R as ResolvedSearchSocketConfig, b as SearchSocketConfig } from './types-DAXk6A3Y.cjs';
+import { R as ResolvedSearchSocketConfig, b as SearchSocketConfig } from './types-BrG6XTUU.cjs';
 
 interface SearchSocketHandleOptions {
     configPath?: string;

package/dist/sveltekit.d.ts

@@ -1,4 +1,4 @@
-import { R as ResolvedSearchSocketConfig, b as SearchSocketConfig } from './types-DAXk6A3Y.js';
+import { R as ResolvedSearchSocketConfig, b as SearchSocketConfig } from './types-BrG6XTUU.js';
 
 interface SearchSocketHandleOptions {
     configPath?: string;

package/dist/sveltekit.js

@@ -16672,7 +16672,7 @@ var searchSocketConfigSchema = z.object({
   ranking: z.object({
     enableIncomingLinkBoost: z.boolean().optional(),
     enableDepthBoost: z.boolean().optional(),
-    pageWeights: z.record(z.string(), z.number().positive()).optional(),
+    pageWeights: z.record(z.string(), z.number().min(0)).optional(),
     aggregationCap: z.number().int().positive().optional(),
     aggregationDecay: z.number().min(0).max(1).optional(),
     minChunkScoreRatio: z.number().min(0).max(1).optional(),
@@ -17456,23 +17456,12 @@ var TursoVectorStore = class {
         incoming_links INTEGER NOT NULL DEFAULT 0,
         route_file     TEXT NOT NULL DEFAULT '',
         tags           TEXT NOT NULL DEFAULT '[]',
+        description    TEXT NOT NULL DEFAULT '',
+        keywords       TEXT NOT NULL DEFAULT '[]',
         embedding      F32_BLOB(${dim})
       )`,
       `CREATE INDEX IF NOT EXISTS idx ON chunks (libsql_vector_idx(embedding, 'metric=cosine'))`
     ]);
-    const chunkMigrationCols = [
-      { name: "chunk_text", def: "TEXT NOT NULL DEFAULT ''" },
-      { name: "ordinal", def: "INTEGER NOT NULL DEFAULT 0" }
-    ];
-    for (const col of chunkMigrationCols) {
-      try {
-        await this.client.execute(`ALTER TABLE chunks ADD COLUMN ${col.name} ${col.def}`);
-      } catch (error) {
-        if (error instanceof Error && !error.message.includes("duplicate column")) {
-          throw error;
-        }
-      }
-    }
     this.chunksReady = true;
   }
   async ensurePages() {
@@ -17550,8 +17539,8 @@ var TursoVectorStore = class {
         sql: `INSERT OR REPLACE INTO chunks
               (id, project_id, scope_name, url, path, title, section_title,
                heading_path, snippet, chunk_text, ordinal, content_hash, model_id, depth,
-               incoming_links, route_file, tags, embedding)
-              VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, vector(?))`,
+               incoming_links, route_file, tags, description, keywords, embedding)
+              VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, vector(?))`,
         args: [
           r.id,
           r.metadata.projectId,
@@ -17570,6 +17559,8 @@ var TursoVectorStore = class {
           r.metadata.incomingLinks,
           r.metadata.routeFile,
           JSON.stringify(r.metadata.tags),
+          r.metadata.description ?? "",
+          JSON.stringify(r.metadata.keywords ?? []),
           JSON.stringify(r.vector)
         ]
       }));
@@ -17585,6 +17576,7 @@ var TursoVectorStore = class {
                    c.section_title, c.heading_path, c.snippet, c.chunk_text,
                    c.ordinal, c.content_hash,
                    c.model_id, c.depth, c.incoming_links, c.route_file, c.tags,
+                   c.description, c.keywords,
                    vector_distance_cos(c.embedding, vector(?)) AS distance
             FROM vector_top_k('idx', vector(?), ?) AS v
             JOIN chunks AS c ON c.rowid = v.id`,
@@ -17615,6 +17607,12 @@ var TursoVectorStore = class {
       }
       const distance = row.distance;
       const score = 1 - distance;
+      const description = row.description || void 0;
+      const keywords = (() => {
+        const raw = row.keywords || "[]";
+        const parsed = JSON.parse(raw);
+        return parsed.length > 0 ? parsed : void 0;
+      })();
       hits.push({
         id: row.id,
         score,
@@ -17634,7 +17632,9 @@ var TursoVectorStore = class {
           depth: row.depth,
           incomingLinks: row.incoming_links,
           routeFile: row.route_file,
-          tags
+          tags,
+          description,
+          keywords
         }
       });
     }
@@ -17928,6 +17928,7 @@ function aggregateByPage(ranked, config) {
     }
     let pageScore = maxScore + aggregationBonus * config.ranking.weights.aggregation;
     const pageWeight = findPageWeight(url, config.ranking.pageWeights);
+    if (pageWeight === 0) continue;
     if (pageWeight !== 1) {
       pageScore *= pageWeight;
     }
@@ -18150,16 +18151,29 @@ var SearchEngine = class _SearchEngine {
       if (group) group.push(entry);
       else pageGroups.set(url, [entry]);
     }
+    const MAX_CHUNKS_PER_PAGE = 5;
+    const MIN_CHUNKS_PER_PAGE = 1;
+    const MIN_CHUNK_SCORE_RATIO = 0.5;
     const pageCandidates = [];
     for (const [url, chunks] of pageGroups) {
-      const sorted = [...chunks].sort(
-        (a, b) => (a.hit.metadata.ordinal ?? 0) - (b.hit.metadata.ordinal ?? 0)
-      );
-      const title = sorted[0].hit.metadata.title;
-      const body = sorted.map((c) => c.hit.metadata.chunkText || c.hit.metadata.snippet).join("\n\n");
-      pageCandidates.push({ id: url, text: `${title}
-
-${body}` });
+      const byScore = [...chunks].sort((a, b) => b.finalScore - a.finalScore);
+      const bestScore = byScore[0].finalScore;
+      const scoreFloor = Number.isFinite(bestScore) ? bestScore * MIN_CHUNK_SCORE_RATIO : Number.NEGATIVE_INFINITY;
+      const selected = byScore.filter(
+        (c, i) => i < MIN_CHUNKS_PER_PAGE || c.finalScore >= scoreFloor
+      ).slice(0, MAX_CHUNKS_PER_PAGE);
+      selected.sort((a, b) => (a.hit.metadata.ordinal ?? 0) - (b.hit.metadata.ordinal ?? 0));
+      const first = selected[0].hit.metadata;
+      const parts = [first.title];
+      if (first.description) {
+        parts.push(first.description);
+      }
+      if (first.keywords && first.keywords.length > 0) {
+        parts.push(first.keywords.join(", "));
+      }
+      const body = selected.map((c) => c.hit.metadata.chunkText || c.hit.metadata.snippet).join("\n\n");
+      parts.push(body);
+      pageCandidates.push({ id: url, text: parts.join("\n\n") });
     }
     const reranked = await this.reranker.rerank(
       query,
@@ -18669,7 +18683,9 @@ function chunkMirrorPage(page, config, scope) {
       incomingLinks: page.incomingLinks,
       routeFile: page.routeFile,
       tags: page.tags,
-      contentHash: ""
+      contentHash: "",
+      description: page.description,
+      keywords: page.keywords
     };
     const embeddingText = buildEmbeddingText(summaryChunk, config.chunking.prependTitle);
     summaryChunk.contentHash = sha256(normalizeText(embeddingText));
@@ -18696,7 +18712,9 @@ function chunkMirrorPage(page, config, scope) {
       incomingLinks: page.incomingLinks,
       routeFile: page.routeFile,
       tags: page.tags,
-      contentHash: ""
+      contentHash: "",
+      description: page.description,
+      keywords: page.keywords
     };
     const embeddingText = buildEmbeddingText(chunk, config.chunking.prependTitle);
     chunk.contentHash = sha256(normalizeText(embeddingText));
@@ -19551,7 +19569,7 @@ function extractFromHtml(url, html, config) {
   const $ = load(html);
   const normalizedUrl = normalizeUrlPath(url);
   const pageBaseUrl = new URL(`https://searchsocket.local${normalizedUrl}`);
-  const title = normalizeText($("title").first().text() || "") || normalizeText($(`${config.extract.mainSelector} h1`).first().text() || "") || normalizedUrl;
+  const title = $("meta[property='og:title']").attr("content")?.trim() || normalizeText($(`${config.extract.mainSelector} h1`).first().text() || "") || $("meta[name='twitter:title']").attr("content")?.trim() || normalizeText($("title").first().text() || "") || normalizedUrl;
   if (config.extract.respectRobotsNoindex) {
     const robots = $("meta[name='robots']").attr("content") ?? "";
     if (/\bnoindex\b/i.test(robots)) {
@@ -20659,7 +20677,9 @@ var IndexPipeline = class _IndexPipeline {
             depth: chunk.depth,
             incomingLinks: chunk.incomingLinks,
             routeFile: chunk.routeFile,
-            tags: chunk.tags
+            tags: chunk.tags,
+            description: chunk.description,
+            keywords: chunk.keywords
           }
         });
       }

package/dist/{types-DAXk6A3Y.d.cts → types-BrG6XTUU.d.cts}

@@ -261,6 +261,8 @@ interface Chunk {
     routeFile: string;
     tags: string[];
     contentHash: string;
+    description?: string;
+    keywords?: string[];
 }
 interface VectorRecord {
     id: string;
@@ -282,6 +284,8 @@ interface VectorRecord {
         incomingLinks: number;
         routeFile: string;
         tags: string[];
+        description?: string;
+        keywords?: string[];
     };
 }
 interface QueryOpts {

package/dist/{types-DAXk6A3Y.d.ts → types-BrG6XTUU.d.ts}

@@ -261,6 +261,8 @@ interface Chunk {
     routeFile: string;
     tags: string[];
     contentHash: string;
+    description?: string;
+    keywords?: string[];
 }
 interface VectorRecord {
     id: string;
@@ -282,6 +284,8 @@ interface VectorRecord {
         incomingLinks: number;
         routeFile: string;
         tags: string[];
+        description?: string;
+        keywords?: string[];
     };
 }
 interface QueryOpts {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "searchsocket",
-  "version": "0.3.0",
+  "version": "0.3.2",
   "description": "Semantic site search and MCP retrieval for SvelteKit static sites",
   "license": "MIT",
   "author": "Greg Priday <greg@siteorigin.com>",