searchsocket 0.3.2 → 0.3.3

package/dist/cli.js

@@ -12,7 +12,7 @@ import { Command } from "commander";
 // package.json
 var package_default = {
   name: "searchsocket",
-  version: "0.3.2",
+  version: "0.3.3",
   description: "Semantic site search and MCP retrieval for SvelteKit static sites",
   license: "MIT",
   author: "Greg Priday <greg@siteorigin.com>",
@@ -2854,7 +2854,7 @@ var JinaReranker = class {
   constructor(options) {
     this.apiKey = options.apiKey;
     this.model = options.model;
-    this.maxRetries = options.maxRetries ?? 4;
+    this.maxRetries = options.maxRetries ?? 2;
   }
   async rerank(query, candidates, topN) {
     if (candidates.length === 0) {
@@ -2864,7 +2864,8 @@ var JinaReranker = class {
       model: this.model,
       query,
       documents: candidates.map((candidate) => candidate.text),
-      top_n: topN ?? candidates.length
+      top_n: topN ?? candidates.length,
+      return_documents: false
     };
     let attempt = 0;
     while (attempt <= this.maxRetries) {
@@ -3236,6 +3237,7 @@ var SearchEngine = class _SearchEngine {
     const MAX_CHUNKS_PER_PAGE = 5;
     const MIN_CHUNKS_PER_PAGE = 1;
     const MIN_CHUNK_SCORE_RATIO = 0.5;
+    const MAX_DOC_CHARS = 2e3;
     const pageCandidates = [];
     for (const [url, chunks] of pageGroups) {
       const byScore = [...chunks].sort((a, b) => b.finalScore - a.finalScore);
@@ -3255,12 +3257,18 @@ var SearchEngine = class _SearchEngine {
       }
       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") });
+      let text = parts.join("\n\n");
+      if (text.length > MAX_DOC_CHARS) {
+        text = text.slice(0, MAX_DOC_CHARS);
+      }
+      pageCandidates.push({ id: url, text });
     }
+    const maxCandidates = Math.max(topK, this.config.rerank.topN);
+    const cappedCandidates = pageCandidates.slice(0, maxCandidates);
     const reranked = await this.reranker.rerank(
       query,
-      pageCandidates,
-      Math.max(topK, this.config.rerank.topN)
+      cappedCandidates,
+      maxCandidates
     );
     const scoreByUrl = new Map(reranked.map((e) => [e.id, e.score]));
     return ranked.map((entry) => {

package/dist/index.cjs

@@ -17273,7 +17273,7 @@ var JinaReranker = class {
   constructor(options) {
     this.apiKey = options.apiKey;
     this.model = options.model;
-    this.maxRetries = options.maxRetries ?? 4;
+    this.maxRetries = options.maxRetries ?? 2;
   }
   async rerank(query, candidates, topN) {
     if (candidates.length === 0) {
@@ -17283,7 +17283,8 @@ var JinaReranker = class {
       model: this.model,
       query,
       documents: candidates.map((candidate) => candidate.text),
-      top_n: topN ?? candidates.length
+      top_n: topN ?? candidates.length,
+      return_documents: false
     };
     let attempt = 0;
     while (attempt <= this.maxRetries) {
@@ -20505,6 +20506,7 @@ var SearchEngine = class _SearchEngine {
     const MAX_CHUNKS_PER_PAGE = 5;
     const MIN_CHUNKS_PER_PAGE = 1;
     const MIN_CHUNK_SCORE_RATIO = 0.5;
+    const MAX_DOC_CHARS = 2e3;
     const pageCandidates = [];
     for (const [url, chunks] of pageGroups) {
       const byScore = [...chunks].sort((a, b) => b.finalScore - a.finalScore);
@@ -20524,12 +20526,18 @@ var SearchEngine = class _SearchEngine {
       }
       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") });
+      let text = parts.join("\n\n");
+      if (text.length > MAX_DOC_CHARS) {
+        text = text.slice(0, MAX_DOC_CHARS);
+      }
+      pageCandidates.push({ id: url, text });
     }
+    const maxCandidates = Math.max(topK, this.config.rerank.topN);
+    const cappedCandidates = pageCandidates.slice(0, maxCandidates);
     const reranked = await this.reranker.rerank(
       query,
-      pageCandidates,
-      Math.max(topK, this.config.rerank.topN)
+      cappedCandidates,
+      maxCandidates
     );
     const scoreByUrl = new Map(reranked.map((e) => [e.id, e.score]));
     return ranked.map((entry) => {

package/dist/index.js

@@ -17261,7 +17261,7 @@ var JinaReranker = class {
   constructor(options) {
     this.apiKey = options.apiKey;
     this.model = options.model;
-    this.maxRetries = options.maxRetries ?? 4;
+    this.maxRetries = options.maxRetries ?? 2;
   }
   async rerank(query, candidates, topN) {
     if (candidates.length === 0) {
@@ -17271,7 +17271,8 @@ var JinaReranker = class {
       model: this.model,
       query,
       documents: candidates.map((candidate) => candidate.text),
-      top_n: topN ?? candidates.length
+      top_n: topN ?? candidates.length,
+      return_documents: false
     };
     let attempt = 0;
     while (attempt <= this.maxRetries) {
@@ -20493,6 +20494,7 @@ var SearchEngine = class _SearchEngine {
     const MAX_CHUNKS_PER_PAGE = 5;
     const MIN_CHUNKS_PER_PAGE = 1;
     const MIN_CHUNK_SCORE_RATIO = 0.5;
+    const MAX_DOC_CHARS = 2e3;
     const pageCandidates = [];
     for (const [url, chunks] of pageGroups) {
       const byScore = [...chunks].sort((a, b) => b.finalScore - a.finalScore);
@@ -20512,12 +20514,18 @@ var SearchEngine = class _SearchEngine {
       }
       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") });
+      let text = parts.join("\n\n");
+      if (text.length > MAX_DOC_CHARS) {
+        text = text.slice(0, MAX_DOC_CHARS);
+      }
+      pageCandidates.push({ id: url, text });
     }
+    const maxCandidates = Math.max(topK, this.config.rerank.topN);
+    const cappedCandidates = pageCandidates.slice(0, maxCandidates);
     const reranked = await this.reranker.rerank(
       query,
-      pageCandidates,
-      Math.max(topK, this.config.rerank.topN)
+      cappedCandidates,
+      maxCandidates
     );
     const scoreByUrl = new Map(reranked.map((e) => [e.id, e.score]));
     return ranked.map((entry) => {

package/dist/sveltekit.cjs

@@ -17254,7 +17254,7 @@ var JinaReranker = class {
   constructor(options) {
     this.apiKey = options.apiKey;
     this.model = options.model;
-    this.maxRetries = options.maxRetries ?? 4;
+    this.maxRetries = options.maxRetries ?? 2;
   }
   async rerank(query, candidates, topN) {
     if (candidates.length === 0) {
@@ -17264,7 +17264,8 @@ var JinaReranker = class {
       model: this.model,
       query,
       documents: candidates.map((candidate) => candidate.text),
-      top_n: topN ?? candidates.length
+      top_n: topN ?? candidates.length,
+      return_documents: false
     };
     let attempt = 0;
     while (attempt <= this.maxRetries) {
@@ -18166,6 +18167,7 @@ var SearchEngine = class _SearchEngine {
     const MAX_CHUNKS_PER_PAGE = 5;
     const MIN_CHUNKS_PER_PAGE = 1;
     const MIN_CHUNK_SCORE_RATIO = 0.5;
+    const MAX_DOC_CHARS = 2e3;
     const pageCandidates = [];
     for (const [url, chunks] of pageGroups) {
       const byScore = [...chunks].sort((a, b) => b.finalScore - a.finalScore);
@@ -18185,12 +18187,18 @@ var SearchEngine = class _SearchEngine {
       }
       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") });
+      let text = parts.join("\n\n");
+      if (text.length > MAX_DOC_CHARS) {
+        text = text.slice(0, MAX_DOC_CHARS);
+      }
+      pageCandidates.push({ id: url, text });
     }
+    const maxCandidates = Math.max(topK, this.config.rerank.topN);
+    const cappedCandidates = pageCandidates.slice(0, maxCandidates);
     const reranked = await this.reranker.rerank(
       query,
-      pageCandidates,
-      Math.max(topK, this.config.rerank.topN)
+      cappedCandidates,
+      maxCandidates
     );
     const scoreByUrl = new Map(reranked.map((e) => [e.id, e.score]));
     return ranked.map((entry) => {

package/dist/sveltekit.js

@@ -17242,7 +17242,7 @@ var JinaReranker = class {
   constructor(options) {
     this.apiKey = options.apiKey;
     this.model = options.model;
-    this.maxRetries = options.maxRetries ?? 4;
+    this.maxRetries = options.maxRetries ?? 2;
   }
   async rerank(query, candidates, topN) {
     if (candidates.length === 0) {
@@ -17252,7 +17252,8 @@ var JinaReranker = class {
       model: this.model,
       query,
       documents: candidates.map((candidate) => candidate.text),
-      top_n: topN ?? candidates.length
+      top_n: topN ?? candidates.length,
+      return_documents: false
     };
     let attempt = 0;
     while (attempt <= this.maxRetries) {
@@ -18154,6 +18155,7 @@ var SearchEngine = class _SearchEngine {
     const MAX_CHUNKS_PER_PAGE = 5;
     const MIN_CHUNKS_PER_PAGE = 1;
     const MIN_CHUNK_SCORE_RATIO = 0.5;
+    const MAX_DOC_CHARS = 2e3;
     const pageCandidates = [];
     for (const [url, chunks] of pageGroups) {
       const byScore = [...chunks].sort((a, b) => b.finalScore - a.finalScore);
@@ -18173,12 +18175,18 @@ var SearchEngine = class _SearchEngine {
       }
       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") });
+      let text = parts.join("\n\n");
+      if (text.length > MAX_DOC_CHARS) {
+        text = text.slice(0, MAX_DOC_CHARS);
+      }
+      pageCandidates.push({ id: url, text });
     }
+    const maxCandidates = Math.max(topK, this.config.rerank.topN);
+    const cappedCandidates = pageCandidates.slice(0, maxCandidates);
     const reranked = await this.reranker.rerank(
       query,
-      pageCandidates,
-      Math.max(topK, this.config.rerank.topN)
+      cappedCandidates,
+      maxCandidates
     );
     const scoreByUrl = new Map(reranked.map((e) => [e.id, e.score]));
     return ranked.map((entry) => {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "searchsocket",
-  "version": "0.3.2",
+  "version": "0.3.3",
   "description": "Semantic site search and MCP retrieval for SvelteKit static sites",
   "license": "MIT",
   "author": "Greg Priday <greg@siteorigin.com>",