bluera-knowledge 0.35.0 → 0.37.0

package/dist/{chunk-L2SC6J4K.js → chunk-724FNI27.js}

@@ -2072,40 +2072,6 @@ import { readFile as readFile2, access } from "fs/promises";
 import { homedir as homedir2 } from "os";
 import { isAbsolute, join as join6, resolve } from "path";
 
-// src/services/reranker-env.ts
-var logger = createLogger("reranker-env");
-function parseRerankerEnvOverrides(strict) {
-  return {
-    enabled: parseEnabled(process.env["BK_RERANKER_ENABLED"], strict),
-    topK: parseTopK(process.env["BK_RERANKER_TOPK"], strict)
-  };
-}
-function parseEnabled(raw, strict) {
-  if (raw === void 0 || raw === "") return void 0;
-  if (raw === "1") return true;
-  if (raw === "0") return false;
-  const msg = `BK_RERANKER_ENABLED must be '0' or '1', got: "${raw}"`;
-  if (strict) throw new Error(msg);
-  logger.warn(msg);
-  return void 0;
-}
-function parseTopK(raw, strict) {
-  if (raw === void 0 || raw === "") return void 0;
-  const parsed = Number.parseInt(raw, 10);
-  if (Number.isNaN(parsed) || parsed < 1) {
-    const msg = `BK_RERANKER_TOPK must be a positive integer, got: "${raw}"`;
-    if (strict) throw new Error(msg);
-    logger.warn(msg);
-    return void 0;
-  }
-  return parsed;
-}
-
-// src/db/embeddings.ts
-import { homedir } from "os";
-import { join as join5 } from "path";
-import { pipeline, env } from "@huggingface/transformers";
-
 // src/models/registry.ts
 var MODEL_REGISTRY = {
   // ============================================================
@@ -2362,7 +2328,7 @@ var MODEL_REGISTRY = {
     normalize: true,
     queryPrefix: "",
     docPrefix: "",
-    category: "other",
+    category: "jina",
     sizeCategory: "small",
     notes: "8192 token context. Good for long documents."
   },
@@ -2374,12 +2340,102 @@ var MODEL_REGISTRY = {
     normalize: true,
     queryPrefix: "",
     docPrefix: "",
-    category: "other",
+    category: "jina",
     sizeCategory: "base",
     notes: "8192 token context. Larger Jina variant."
+  },
+  "jina-embeddings-v2-base-code": {
+    id: "jinaai/jina-embeddings-v2-base-code",
+    name: "Jina Embeddings v2 Base Code",
+    dimensions: 768,
+    pooling: "mean",
+    normalize: true,
+    queryPrefix: "",
+    docPrefix: "",
+    category: "jina",
+    sizeCategory: "base",
+    notes: "161M params. Code-specific (150M+ code QA pairs, 30+ langs). 8K context. May need trust_remote_code."
+  },
+  // ============================================================
+  // Snowflake Arctic Embed - Retrieval-optimized
+  // ============================================================
+  "snowflake-arctic-embed-xs": {
+    id: "Snowflake/snowflake-arctic-embed-xs",
+    name: "Snowflake Arctic Embed XS",
+    dimensions: 384,
+    pooling: "cls",
+    normalize: true,
+    queryPrefix: "Represent this sentence for searching relevant passages: ",
+    docPrefix: "",
+    category: "snowflake",
+    sizeCategory: "small",
+    notes: "22M params. Ultra-small retrieval model. Based on all-MiniLM-L6-v2."
+  },
+  "snowflake-arctic-embed-s": {
+    id: "Snowflake/snowflake-arctic-embed-s",
+    name: "Snowflake Arctic Embed S",
+    dimensions: 384,
+    pooling: "cls",
+    normalize: true,
+    queryPrefix: "Represent this sentence for searching relevant passages: ",
+    docPrefix: "",
+    category: "snowflake",
+    sizeCategory: "small",
+    notes: "33M params. Same size as bge-small, trained for retrieval. Drop-in candidate."
+  },
+  "snowflake-arctic-embed-m-v1.5": {
+    id: "Snowflake/snowflake-arctic-embed-m-v1.5",
+    name: "Snowflake Arctic Embed M v1.5",
+    dimensions: 768,
+    pooling: "cls",
+    normalize: true,
+    queryPrefix: "Represent this sentence for searching relevant passages: ",
+    docPrefix: "",
+    category: "snowflake",
+    sizeCategory: "base",
+    notes: "109M params. BEIR 55.14. Matryoshka (truncate to 256d). 7 ONNX quant variants."
+  },
+  "snowflake-arctic-embed-m-v2.0": {
+    id: "Snowflake/snowflake-arctic-embed-m-v2.0",
+    name: "Snowflake Arctic Embed M v2.0",
+    dimensions: 768,
+    pooling: "cls",
+    normalize: true,
+    queryPrefix: "query: ",
+    docPrefix: "",
+    category: "snowflake",
+    sizeCategory: "base",
+    notes: "305M params. Multilingual, 8K context. Custom GTE arch \u2014 may need trust_remote_code."
+  },
+  // ============================================================
+  // ModernBERT Embedding Models - Latest architecture (2024+)
+  // ============================================================
+  "gte-modernbert-base": {
+    id: "Alibaba-NLP/gte-modernbert-base",
+    name: "GTE ModernBERT Base",
+    dimensions: 768,
+    pooling: "cls",
+    normalize: true,
+    queryPrefix: "",
+    docPrefix: "",
+    category: "gte",
+    sizeCategory: "base",
+    notes: "149M params. CoIR code retrieval 79.31. BEIR 55.33. 8K context. No trust_remote_code. Top candidate."
+  },
+  "modernbert-embed-base": {
+    id: "nomic-ai/modernbert-embed-base",
+    name: "ModernBERT Embed Base (Nomic)",
+    dimensions: 768,
+    pooling: "mean",
+    normalize: true,
+    queryPrefix: "search_query: ",
+    docPrefix: "search_document: ",
+    category: "nomic",
+    sizeCategory: "base",
+    notes: "149M params. BEIR 52.89. Matryoshka (truncate to 256d). 8K context."
   }
 };
-var DEFAULT_MODEL_ID = "bge-small-en-v1.5";
+var DEFAULT_MODEL_ID = "snowflake-arctic-embed-s";
 function getModelConfig(modelId) {
   if (modelId in MODEL_REGISTRY) {
     return MODEL_REGISTRY[modelId];
@@ -2403,8 +2459,75 @@ function getConfiguredModelId() {
   }
   return DEFAULT_MODEL_ID;
 }
+var RERANKER_REGISTRY = {
+  "ms-marco-MiniLM-L-6-v2": {
+    id: "Xenova/ms-marco-MiniLM-L-6-v2",
+    name: "MS MARCO MiniLM L6 v2",
+    notes: "Default reranker. Fast cross-encoder for passage ranking."
+  },
+  "ms-marco-MiniLM-L-12-v2": {
+    id: "Xenova/ms-marco-MiniLM-L-12-v2",
+    name: "MS MARCO MiniLM L12 v2",
+    notes: "Deeper reranker. Better quality but slower."
+  },
+  "bge-reranker-base": {
+    id: "Xenova/bge-reranker-base",
+    name: "BGE Reranker Base",
+    notes: "BGE-family cross-encoder. Better fit for BGE embeddings."
+  },
+  "bge-reranker-large": {
+    id: "Xenova/bge-reranker-large",
+    name: "BGE Reranker Large",
+    notes: "Larger BGE cross-encoder. Higher quality, slower."
+  }
+};
+
+// src/services/reranker-env.ts
+var logger = createLogger("reranker-env");
+function parseRerankerEnvOverrides(strict) {
+  return {
+    enabled: parseEnabled(process.env["BK_RERANKER_ENABLED"], strict),
+    topK: parseTopK(process.env["BK_RERANKER_TOPK"], strict),
+    model: parseModel(process.env["BK_RERANKER_MODEL"], strict)
+  };
+}
+function parseEnabled(raw, strict) {
+  if (raw === void 0 || raw === "") return void 0;
+  if (raw === "1") return true;
+  if (raw === "0") return false;
+  const msg = `BK_RERANKER_ENABLED must be '0' or '1', got: "${raw}"`;
+  if (strict) throw new Error(msg);
+  logger.warn(msg);
+  return void 0;
+}
+function parseTopK(raw, strict) {
+  if (raw === void 0 || raw === "") return void 0;
+  const parsed = Number.parseInt(raw, 10);
+  if (Number.isNaN(parsed) || parsed < 1) {
+    const msg = `BK_RERANKER_TOPK must be a positive integer, got: "${raw}"`;
+    if (strict) throw new Error(msg);
+    logger.warn(msg);
+    return void 0;
+  }
+  return parsed;
+}
+function parseModel(raw, strict) {
+  if (raw === void 0 || raw === "") return void 0;
+  const entry = RERANKER_REGISTRY[raw];
+  if (entry === void 0) {
+    const valid = Object.keys(RERANKER_REGISTRY).join(", ");
+    const msg = `BK_RERANKER_MODEL must be one of [${valid}], got: "${raw}"`;
+    if (strict) throw new Error(msg);
+    logger.warn(msg);
+    return void 0;
+  }
+  return entry.id;
+}
 
 // src/db/embeddings.ts
+import { homedir } from "os";
+import { join as join5 } from "path";
+import { pipeline, env } from "@huggingface/transformers";
 env.cacheDir = join5(homedir(), ".cache", "huggingface-transformers");
 function getFinetunedModelPath() {
   const path4 = process.env["BK_FINETUNED_MODEL"];
@@ -2419,14 +2542,16 @@ function buildEmbeddingConfig(modelId, overrides) {
   const envQueryPrefix = process.env["BK_QUERY_PREFIX"];
   const modelConfig = getModelConfig(modelId);
   if (modelConfig === void 0) {
+    const baseModelId = process.env["BK_BASE_MODEL"];
+    const baseConfig = baseModelId !== void 0 && baseModelId !== "" ? getModelConfig(baseModelId) : void 0;
     return {
       model: modelId,
       batchSize: overrides?.batchSize ?? 32,
       dtype: overrides?.dtype ?? "fp32",
-      pooling: overrides?.pooling ?? envPooling ?? "mean",
-      normalize: overrides?.normalize ?? true,
-      queryPrefix: overrides?.queryPrefix ?? envQueryPrefix ?? "",
-      docPrefix: overrides?.docPrefix ?? "",
+      pooling: overrides?.pooling ?? envPooling ?? baseConfig?.pooling ?? "mean",
+      normalize: overrides?.normalize ?? baseConfig?.normalize ?? true,
+      queryPrefix: overrides?.queryPrefix ?? envQueryPrefix ?? baseConfig?.queryPrefix ?? "",
+      docPrefix: overrides?.docPrefix ?? baseConfig?.docPrefix ?? "",
       maxInFlightBatches: overrides?.maxInFlightBatches ?? 1
     };
   }
@@ -2670,10 +2795,10 @@ var DEFAULT_CONFIG = {
   version: 1,
   dataDir: ".bluera/bluera-knowledge/data",
   embedding: {
-    model: "Xenova/bge-small-en-v1.5",
+    model: "Snowflake/snowflake-arctic-embed-s",
     batchSize: 32,
     dtype: "fp32",
-    pooling: "mean",
+    pooling: "cls",
     normalize: true,
     queryPrefix: "Represent this sentence for searching relevant passages: ",
     docPrefix: "",
@@ -2824,13 +2949,14 @@ var ConfigService = class {
       };
     }
     const rerankerOverrides = parseRerankerEnvOverrides(false);
-    if (rerankerOverrides.enabled !== void 0 || rerankerOverrides.topK !== void 0) {
+    if (rerankerOverrides.enabled !== void 0 || rerankerOverrides.topK !== void 0 || rerankerOverrides.model !== void 0) {
       this.config = {
         ...this.config,
         reranker: {
           ...this.config.reranker,
           ...rerankerOverrides.enabled !== void 0 ? { enabled: rerankerOverrides.enabled } : {},
-          ...rerankerOverrides.topK !== void 0 ? { topK: rerankerOverrides.topK } : {}
+          ...rerankerOverrides.topK !== void 0 ? { topK: rerankerOverrides.topK } : {},
+          ...rerankerOverrides.model !== void 0 ? { model: rerankerOverrides.model } : {}
         }
       };
     }
@@ -3361,11 +3487,15 @@ var DriftService = class {
   }
 };
 
-// src/utils/text-extensions.ts
+// src/services/index.service.ts
+var minimatch = minimatchFn;
+var execFileAsync = promisify(execFile);
+var logger2 = createLogger("index-service");
 var TEXT_EXTENSIONS = /* @__PURE__ */ new Set([
   // Text/docs
   ".txt",
   ".md",
+  ".mdx",
   ".rst",
   ".adoc",
   // JavaScript/TypeScript
@@ -3471,11 +3601,33 @@ var TEXT_EXTENSIONS = /* @__PURE__ */ new Set([
   ".makefile",
   ".cmake"
 ]);
-
-// src/services/index.service.ts
-var minimatch = minimatchFn;
-var execFileAsync = promisify(execFile);
-var logger2 = createLogger("index-service");
+var TEXT_BASENAMES = /* @__PURE__ */ new Set([
+  // Build systems
+  "Dockerfile",
+  "Makefile",
+  "Rakefile",
+  "Gemfile",
+  "Procfile",
+  "Vagrantfile",
+  "Justfile",
+  "Brewfile",
+  "Earthfile",
+  "Tiltfile",
+  "CMakeLists.txt",
+  "BUILD",
+  "BUILD.bazel",
+  "WORKSPACE",
+  // Dotfiles (config)
+  ".gitignore",
+  ".gitattributes",
+  ".editorconfig",
+  ".dockerignore",
+  ".eslintignore",
+  ".prettierignore",
+  ".npmignore",
+  ".env.example",
+  ".env.sample"
+]);
 function isMinifiedFile(filename) {
   const minPatterns = [
     /\.min\.(js|css|mjs|cjs)$/i,
@@ -3542,6 +3694,7 @@ var EXT_TO_LANGUAGE = {
   ".psm1": "powershell",
   ".sql": "sql",
   ".md": "markdown",
+  ".mdx": "mdx",
   ".rst": "restructuredtext",
   ".lua": "lua",
   ".r": "r",
@@ -3581,6 +3734,19 @@ var EXT_TO_LANGUAGE = {
   ".toml": "toml",
   ".xml": "xml"
 };
+var BASENAME_TO_LANGUAGE = {
+  Dockerfile: "dockerfile",
+  Makefile: "makefile",
+  Rakefile: "ruby",
+  Gemfile: "ruby",
+  Brewfile: "ruby",
+  "CMakeLists.txt": "cmake",
+  BUILD: "starlark",
+  "BUILD.bazel": "starlark",
+  WORKSPACE: "starlark",
+  Tiltfile: "starlark",
+  Justfile: "just"
+};
 var ENTRY_POINT_NAMES = /* @__PURE__ */ new Set([
   "index.ts",
   "index.js",
@@ -3969,7 +4135,7 @@ ${rawContent}` : rawContent;
     const ext = extname(filePath).toLowerCase();
     const fileName = basename(filePath).toLowerCase();
     const fileType = this.classifyFileType(ext, fileName, filePath);
-    const language = EXT_TO_LANGUAGE[ext];
+    const language = EXT_TO_LANGUAGE[ext] ?? BASENAME_TO_LANGUAGE[basename(filePath)];
     const normalizedRelPath = relativePath.replaceAll("\\", "/");
     const depth = normalizedRelPath.split("/").length - 1;
     const isEntryPoint = ENTRY_POINT_NAMES.has(basename(filePath));
@@ -4058,7 +4224,8 @@ ${rawContent}` : rawContent;
       candidates = await this.scanDirectory(storePath);
       logger2.debug({ storePath, count: candidates.length }, "Using filesystem walk for discovery");
     }
-    return this.filterFiles(candidates, storePath, ingestConfig);
+    const { files } = await this.filterFiles(candidates, storePath, ingestConfig);
+    return files;
   }
   /**
    * Apply extension filter, ignore patterns, size limit, and per-store ingest filters.
@@ -4070,23 +4237,35 @@ ${rawContent}` : rawContent;
     const skippedMinified = [];
     const skippedBinary = [];
     const skippedExcluded = [];
+    let skippedExtension = 0;
+    let skippedDirSegment = 0;
+    let skippedIgnorePattern = 0;
+    let skippedStatError = 0;
     const maxFileSize = ingestConfig?.maxFileSizeBytes ?? this.maxFileSizeBytes;
-    const skipMinified = ingestConfig?.skipMinified ?? true;
+    const skipMinifiedOpt = ingestConfig?.skipMinified ?? true;
     const skipBinaries = ingestConfig?.skipBinaries ?? true;
     const excludeGlobs = ingestConfig?.excludeGlobs ?? [];
+    const maxFiles = ingestConfig?.maxFiles;
     for (const filePath of files) {
       const ext = extname(filePath).toLowerCase();
       const filename = basename(filePath);
-      if (!TEXT_EXTENSIONS.has(ext)) continue;
+      if (!TEXT_EXTENSIONS.has(ext) && !TEXT_BASENAMES.has(filename)) {
+        skippedExtension++;
+        continue;
+      }
       const relativePath = relative(storePath, filePath).replaceAll("\\", "/");
       const pathSegments = relativePath.split("/");
       const dirSegments = pathSegments.slice(0, -1);
       if (dirSegments.some((segment) => this.ignoreDirs.has(segment))) {
+        skippedDirSegment++;
         continue;
       }
       const shouldIgnore = this.ignoreFilePatterns.some((matcher) => matcher(filename));
-      if (shouldIgnore) continue;
-      if (skipMinified && isMinifiedFile(filename)) {
+      if (shouldIgnore) {
+        skippedIgnorePattern++;
+        continue;
+      }
+      if (skipMinifiedOpt && isMinifiedFile(filename)) {
         skippedMinified.push(filename);
         continue;
       }
@@ -4100,13 +4279,14 @@ ${rawContent}` : rawContent;
       }
       let fileSize;
       try {
-        const stats = await stat2(filePath);
-        fileSize = stats.size;
+        const stats2 = await stat2(filePath);
+        fileSize = stats2.size;
         if (fileSize > maxFileSize) {
           skippedLarge.push({ path: filePath, size: fileSize });
           continue;
         }
       } catch {
+        skippedStatError++;
         continue;
       }
       if (skipBinaries) {
@@ -4116,17 +4296,23 @@ ${rawContent}` : rawContent;
             continue;
           }
         } catch {
+          skippedStatError++;
           continue;
         }
       }
       result.push(filePath);
     }
+    if (maxFiles !== void 0 && result.length > maxFiles) {
+      logger2.info({ storePath, total: result.length, maxFiles }, "Applying maxFiles cap");
+      result.length = maxFiles;
+    }
+    const skippedLargeBytes = skippedLarge.reduce((sum, f) => sum + f.size, 0);
     if (skippedLarge.length > 0) {
       logger2.info(
         {
           storePath,
           count: skippedLarge.length,
-          totalBytes: skippedLarge.reduce((sum, f) => sum + f.size, 0),
+          totalBytes: skippedLargeBytes,
           examples: skippedLarge.slice(0, 5).map((f) => relative(storePath, f.path))
         },
         "Skipped large files"
@@ -4150,7 +4336,21 @@ ${rawContent}` : rawContent;
         "Skipped excluded files"
       );
     }
-    return result;
+    const stats = {
+      candidates: files.length,
+      accepted: result.length,
+      skippedExtension,
+      skippedDirSegment,
+      skippedIgnorePattern,
+      skippedMinified: skippedMinified.length,
+      skippedExcluded: skippedExcluded.length,
+      skippedLargeFiles: skippedLarge.length,
+      skippedLargeBytes,
+      skippedBinary: skippedBinary.length,
+      skippedStatError
+    };
+    logger2.info({ storePath, ...stats }, "File filtering complete");
+    return { files: result, stats };
   }
   async scanDirectory(dir) {
     const files = [];
@@ -4167,7 +4367,7 @@ ${rawContent}` : rawContent;
           continue;
         }
         const ext = extname(entry.name).toLowerCase();
-        if (TEXT_EXTENSIONS.has(ext)) {
+        if (TEXT_EXTENSIONS.has(ext) || TEXT_BASENAMES.has(entry.name)) {
           files.push(fullPath);
         }
       }
@@ -4761,7 +4961,8 @@ function parseSearchEnvOverrides(strict) {
   return {
     rrfK: parseRrfK(process.env["BK_RRF_K"], strict),
     vectorWeight: parseVectorWeight(process.env["BK_RRF_VECTOR_WEIGHT"], strict),
-    candidateMultiplier: parseCandidateMultiplier(process.env["BK_CANDIDATE_MULTIPLIER"], strict)
+    candidateMultiplier: parseCandidateMultiplier(process.env["BK_CANDIDATE_MULTIPLIER"], strict),
+    queryExpansion: parseBoolFlag(process.env["BK_QUERY_EXPANSION"], "BK_QUERY_EXPANSION", strict)
   };
 }
 function parseRrfK(raw, strict) {
@@ -4797,6 +4998,69 @@ function parseCandidateMultiplier(raw, strict) {
   }
   return parsed;
 }
+function parseBoolFlag(raw, name, strict) {
+  if (raw === void 0 || raw === "") return void 0;
+  if (raw === "1") return true;
+  if (raw === "0") return false;
+  const msg = `${name} must be '0' or '1', got: "${raw}"`;
+  if (strict) throw new Error(msg);
+  logger3.warn(msg);
+  return void 0;
+}
+
+// src/utils/code-tokenizer.ts
+function splitIdentifier(identifier) {
+  const segments = identifier.split(/[_.]/).filter((s) => s.length > 0);
+  const words = [];
+  for (const segment of segments) {
+    const parts = segment.replace(/([a-z])([A-Z])/g, "$1\0$2").replace(/([A-Z]{2,})([A-Z][a-z])/g, "$1\0$2").replace(/([a-zA-Z])(\d)/g, "$1\0$2").replace(/(\d)([a-zA-Z])/g, "$1\0$2").split("\0");
+    for (const part of parts) {
+      if (part.length > 0) {
+        words.push(part);
+      }
+    }
+  }
+  return words;
+}
+var IDENTIFIER_PATTERN = /(?:[a-zA-Z_$][\w$]*(?:\.[\w$]+)*)/g;
+function isSplittable(identifier) {
+  if (/[a-z][A-Z]/.test(identifier)) return true;
+  if (identifier.includes("_")) return true;
+  if (identifier.includes(".")) return true;
+  if (/[A-Z]{2,}[a-z]/.test(identifier)) return true;
+  if (/[a-zA-Z]\d|\d[a-zA-Z]/.test(identifier)) return true;
+  return false;
+}
+function extractSplitVariants(text) {
+  const seen = /* @__PURE__ */ new Set();
+  const variants = [];
+  for (const match of text.matchAll(IDENTIFIER_PATTERN)) {
+    const identifier = match[0];
+    if (identifier.length < 3 || seen.has(identifier)) continue;
+    seen.add(identifier);
+    if (!isSplittable(identifier)) continue;
+    const parts = splitIdentifier(identifier);
+    if (parts.length > 1) {
+      variants.push(parts.join(" "));
+      const lower = parts.map((p) => p.toLowerCase()).join(" ");
+      if (lower !== parts.join(" ").toLowerCase()) {
+        variants.push(lower);
+      }
+    }
+  }
+  return variants.join(" ");
+}
+function buildFtsContent(originalContent) {
+  const variants = extractSplitVariants(originalContent);
+  if (variants.length === 0) return originalContent;
+  return `${originalContent}
+${variants}`;
+}
+function normalizeFtsQuery(query) {
+  const variants = extractSplitVariants(query);
+  if (variants.length === 0) return query;
+  return `${query} ${variants}`;
+}
 
 // src/services/search.service.ts
 var logger4 = createLogger("search-service");
@@ -4868,6 +5132,18 @@ var INTENT_FILE_BOOSTS = {
     changelog: 1.1,
     // Often contains bug fixes and known issues
     other: 1
+  },
+  testing: {
+    "documentation-primary": 0.8,
+    documentation: 0.85,
+    example: 1,
+    source: 0.9,
+    "source-internal": 0.85,
+    test: 1.5,
+    // Tests are exactly what the user wants
+    config: 0.7,
+    changelog: 0.6,
+    other: 0.9
   }
 };
 var FRAMEWORK_PATTERNS = [
@@ -4923,12 +5199,23 @@ var CONCEPTUAL_PATTERNS = [
   /\bhow does .* work\b/i,
   /\bwhat('s| is) the (purpose|point|idea)\b/i
 ];
+var TESTING_PATTERNS = [
+  /\b(test|tests|testing)\b/i,
+  /\b(mock|mocking|stub|stubs)\b/i,
+  /\b(assert|assertion|expect)\b/i,
+  /\btest\s*(client|runner|helper|fixture|suite)\b/i,
+  /\b(unit|integration|e2e|end-to-end)\s*test/i,
+  /\b(pytest|jest|vitest|mocha|gotest)\b/i
+];
 function classifyQueryIntents(query) {
   const q = query.toLowerCase();
   const intents = [];
   if (IMPLEMENTATION_PATTERNS.some((p) => p.test(q))) {
     intents.push({ intent: "implementation", confidence: 0.9 });
   }
+  if (TESTING_PATTERNS.some((p) => p.test(q))) {
+    intents.push({ intent: "testing", confidence: 0.85 });
+  }
   if (DEBUGGING_PATTERNS.some((p) => p.test(q))) {
     intents.push({ intent: "debugging", confidence: 0.85 });
   }
@@ -4954,41 +5241,16 @@ function mapSearchIntentToQueryIntent(intent) {
     case "find-pattern":
     case "find-implementation":
     case "find-definition":
-    case "find-files":
       return "implementation";
     case "find-usage":
     case "find-documentation":
       return "how-to";
+    case "find-files":
+      return "implementation";
   }
 }
-var INTENT_EXPANSION_TERMS = {
-  "find-implementation": "source code implementation function class",
-  "find-documentation": "documentation guide tutorial example",
-  "find-usage": "usage example how to use",
-  "find-pattern": "pattern matching code structure",
-  "find-definition": "definition type interface declaration",
-  "find-files": "file module path"
-};
-function expandQueryWithIntent(query, intent) {
-  if (intent === void 0) return query;
-  const expansion = INTENT_EXPANSION_TERMS[intent];
-  return `${query} ${expansion}`;
-}
-function isStrongFtsSignal(query, ftsResults) {
-  if (ftsResults.length < 2) return false;
-  const top = ftsResults[0];
-  const second = ftsResults[1];
-  if (top === void 0 || second === void 0) return false;
-  if (second.score > 0 && top.score / second.score <= 2) return false;
-  const queryLower = query.toLowerCase();
-  const rawFile = top.metadata["file"] ?? top.metadata["path"];
-  const rawName = top.metadata["name"];
-  const filePath = typeof rawFile === "string" ? rawFile : "";
-  const name = typeof rawName === "string" ? rawName : "";
-  return filePath.toLowerCase().includes(queryLower) || name.toLowerCase().includes(queryLower);
-}
 var RRF_PRESETS = {
-  code: { k: 25, vectorWeight: 0.75, ftsWeight: 0.25 },
+  code: { k: 25, vectorWeight: 0.35, ftsWeight: 0.65 },
   web: { k: 30, vectorWeight: 0.7, ftsWeight: 0.3 }
 };
 var DEFAULT_CANDIDATE_MULTIPLIER = 2;
@@ -5075,8 +5337,7 @@ var SearchService = class {
     let rerankTimeMs;
     const fetchLimit = limit * 3;
     if (mode === "vector") {
-      const expandedQuery = expandQueryWithIntent(query.query, query.intent);
-      const rawResults = await this.vectorSearchRaw(expandedQuery, stores, fetchLimit);
+      const rawResults = await this.vectorSearchRaw(query.query, stores, fetchLimit);
       maxRawScore = rawResults.length > 0 ? rawResults[0]?.score ?? 0 : 0;
       allResults = this.normalizeAndFilterScores(rawResults, query.threshold).slice(0, fetchLimit);
     } else if (mode === "fts") {
@@ -5086,8 +5347,7 @@ var SearchService = class {
         query.query,
         stores,
         fetchLimit,
-        query.threshold,
-        query.intent
+        query.threshold
       );
       allResults = hybridResult.results;
       maxRawScore = hybridResult.maxRawScore;
@@ -5226,6 +5486,41 @@ var SearchService = class {
     }
     return normalized;
   }
+  /**
+   * Generate query variants for multi-query expansion.
+   * Strips intent prefixes to create a keyword-focused variant.
+   * Returns original + variants (deduplicated).
+   */
+  expandQuery(query) {
+    const queries = [query];
+    const stripped = query.replace(
+      /^(how to |how do I |how does |implement |usage of |find the |what is |show me |where is |where are )/i,
+      ""
+    ).trim();
+    if (stripped !== query && stripped.length >= 5) {
+      queries.push(stripped);
+    }
+    return queries;
+  }
+  /**
+   * Run vector search across multiple query variants and merge results.
+   * Deduplicates by document ID, keeping the highest score.
+   */
+  async multiQueryVectorSearch(queries, stores, limit) {
+    const allResults = await Promise.all(
+      queries.map((q) => this.vectorSearchRaw(q, stores, limit))
+    );
+    const merged = /* @__PURE__ */ new Map();
+    for (const results of allResults) {
+      for (const r of results) {
+        const existing = merged.get(r.id);
+        if (existing === void 0 || r.score > existing.score) {
+          merged.set(r.id, r);
+        }
+      }
+    }
+    return [...merged.values()].sort((a, b) => b.score - a.score).slice(0, limit);
+  }
   /**
    * Fetch raw vector search results without normalization.
    * Returns results with raw cosine similarity scores [0-1].
@@ -5249,9 +5544,10 @@ var SearchService = class {
   }
   async ftsSearch(query, stores, limit) {
     const results = [];
+    const normalizedQuery = normalizeFtsQuery(query);
     for (const storeId of stores) {
       try {
-        const hits = await this.lanceStore.fullTextSearch(storeId, query, limit);
+        const hits = await this.lanceStore.fullTextSearch(storeId, normalizedQuery, limit);
         results.push(
           ...hits.map((r) => ({
             id: r.id,
@@ -5268,37 +5564,20 @@ var SearchService = class {
   /**
    * Internal hybrid search result with additional metadata for confidence calculation.
    */
-  async hybridSearchWithMetadata(query, stores, limit, threshold, searchIntent) {
+  async hybridSearchWithMetadata(query, stores, limit, threshold) {
     const intents = classifyQueryIntents(query);
     const envOverrides = parseSearchEnvOverrides(false);
     const candidateMultiplier = envOverrides.candidateMultiplier ?? DEFAULT_CANDIDATE_MULTIPLIER;
-    const ftsResults = await this.ftsSearch(query, stores, limit * candidateMultiplier);
-    if (isStrongFtsSignal(query, ftsResults)) {
-      logger4.debug(
-        { query, topScore: ftsResults[0]?.score },
-        "Strong FTS signal \u2014 skipping vector search"
-      );
-      const sorted2 = ftsResults.slice(0, limit).map((r, i) => ({
-        ...r,
-        score: Math.round((1 - i / Math.max(ftsResults.length, 1)) * 1e6) / 1e6
-      }));
-      if (threshold !== void 0) {
-        return { results: sorted2.filter((r) => r.score >= threshold), maxRawScore: 0 };
-      }
-      return { results: sorted2, maxRawScore: 0 };
-    }
-    const expandedQuery = expandQueryWithIntent(query, searchIntent);
-    const rawVectorResults = await this.vectorSearchRaw(
-      expandedQuery,
-      stores,
-      limit * candidateMultiplier
-    );
+    const fetchLimit = limit * candidateMultiplier;
+    const useExpansion = envOverrides.queryExpansion === true;
+    const rawVectorResults = useExpansion ? await this.multiQueryVectorSearch(this.expandQuery(query), stores, fetchLimit) : await this.vectorSearchRaw(query, stores, fetchLimit);
     const rawVectorScores = /* @__PURE__ */ new Map();
     rawVectorResults.forEach((r) => {
       rawVectorScores.set(r.id, r.score);
     });
     const maxRawScore = rawVectorResults.length > 0 ? rawVectorResults[0]?.score ?? 0 : 0;
     const vectorResults = this.normalizeAndFilterScores(rawVectorResults);
+    const ftsResults = await this.ftsSearch(query, stores, limit * candidateMultiplier);
     const vectorRanks = /* @__PURE__ */ new Map();
     const ftsRanks = /* @__PURE__ */ new Map();
     const allDocs = /* @__PURE__ */ new Map();
@@ -5334,6 +5613,8 @@ var SearchService = class {
       const pathKeywordBoost = this.getPathKeywordBoost(query, result2);
       const depthBoost = this.getDepthBoost(result2, getPrimaryIntent(intents));
       const entryPointBoost = this.getEntryPointBoost(result2, getPrimaryIntent(intents));
+      const sectionHeaderBoost = 1;
+      const symbolNameBoost = 1;
       const metadata = {
         vectorRRF,
         ftsRRF,
@@ -5342,7 +5623,9 @@ var SearchService = class {
         urlKeywordBoost,
         pathKeywordBoost,
         depthBoost,
-        entryPointBoost
+        entryPointBoost,
+        sectionHeaderBoost,
+        symbolNameBoost
       };
       if (vectorRank !== Infinity) {
         metadata.vectorRank = vectorRank;
@@ -5355,7 +5638,7 @@ var SearchService = class {
       }
       rrfScores.push({
         id,
-        score: (vectorRRF + ftsRRF) * fileTypeBoost * frameworkBoost * urlKeywordBoost * pathKeywordBoost * depthBoost * entryPointBoost,
+        score: (vectorRRF + ftsRRF) * fileTypeBoost * frameworkBoost * urlKeywordBoost * pathKeywordBoost * depthBoost * entryPointBoost * sectionHeaderBoost * symbolNameBoost,
         result: result2,
         rawVectorScore,
         metadata
@@ -5376,20 +5659,10 @@ var SearchService = class {
       reranked.results.forEach((r) => {
         rerankedScores.set(r.id, r.rerankerScore);
       });
-      const maxRrfScore = sortedAll[0]?.score ?? 1;
-      sorted = sortedAll.map((r, rrfRank) => {
-        const rerankerScore = rerankedScores.get(r.id);
-        if (rerankerScore === void 0) {
-          return { ...r, blendedScore: -Infinity };
-        }
-        const normalizedRrf = maxRrfScore > 0 ? r.score / maxRrfScore : 0;
-        const rrfWeight = rrfRank < 3 ? 0.7 : rrfRank < 10 ? 0.5 : 0.3;
-        const rerankerWeight = 1 - rrfWeight;
-        return {
-          ...r,
-          blendedScore: normalizedRrf * rrfWeight + rerankerScore * rerankerWeight
-        };
-      }).sort((a, b) => b.blendedScore - a.blendedScore).slice(0, limit);
+      sorted = sortedAll.map((r) => ({
+        ...r,
+        rerankerScore: rerankedScores.get(r.id)
+      })).sort((a, b) => (b.rerankerScore ?? -Infinity) - (a.rerankerScore ?? -Infinity)).slice(0, limit);
     } else {
       sorted = sortedAll.slice(0, limit);
     }
@@ -5489,7 +5762,9 @@ var SearchService = class {
     const blendedMultiplier = totalConfidence > 0 ? weightedMultiplier / totalConfidence : 1;
     const finalBoost = baseBoost * blendedMultiplier;
     if (fileType === "test") {
-      return Math.min(finalBoost, 0.6);
+      const primaryIntent = intents[0]?.intent;
+      const cap = primaryIntent === "testing" ? 1.5 : 0.6;
+      return Math.min(finalBoost, cap);
     }
     return finalBoost;
   }
@@ -5646,12 +5921,6 @@ var SearchService = class {
       location: `${path4}${codeUnit ? `:${String(codeUnit.startLine)}` : ""}`,
       relevanceReason: this.generateRelevanceReason(result, query)
     };
-    if (graph) {
-      const relatedFiles = this.getRelatedFilePaths(graph, path4, symbolName);
-      if (relatedFiles.length > 0) {
-        enhanced.summary = { ...enhanced.summary, relatedFiles };
-      }
-    }
     if (detail === "contextual" || detail === "full") {
       const usage = this.getUsageFromGraph(graph, path4, symbolName);
       enhanced.context = {
@@ -5867,23 +6136,9 @@ var SearchService = class {
     };
   }
   /**
-   * Get related file paths from code graph edges for follow-up reads.
-   * Returns unique file paths (max 5) from callers/callees, excluding the result's own file.
+   * Get related code from graph.
+   * Returns callers and callees for the symbol.
    */
-  getRelatedFilePaths(graph, filePath, symbolName) {
-    if (symbolName === "" || symbolName === "(anonymous)") return [];
-    const nodeId = `${filePath}:${symbolName}`;
-    const files = /* @__PURE__ */ new Set();
-    for (const edge of graph.getIncomingEdges(nodeId)) {
-      const [file] = this.parseNodeId(edge.from);
-      if (file && file !== filePath) files.add(file);
-    }
-    for (const edge of graph.getEdges(nodeId)) {
-      const [file] = this.parseNodeId(edge.to);
-      if (file && file !== filePath) files.add(file);
-    }
-    return Array.from(files).slice(0, 5);
-  }
   getRelatedCodeFromGraph(graph, filePath, symbolName) {
     if (!graph || symbolName === "" || symbolName === "(anonymous)") {
       return [];
@@ -5947,7 +6202,9 @@ var IngestConfigSchema = z3.object({
   /** Skip binary files detected by content heuristic - default true */
   skipBinaries: z3.boolean().optional(),
   /** Override max file size for this store (bytes) */
-  maxFileSizeBytes: z3.number().int().positive().optional()
+  maxFileSizeBytes: z3.number().int().positive().optional(),
+  /** Maximum number of files to index (cap applied after all other filters) */
+  maxFiles: z3.number().int().positive().optional()
 });
 var FileStoreDefinitionSchema = BaseStoreDefinitionSchema.extend({
   type: z3.literal("file"),
@@ -6190,7 +6447,7 @@ async function cloneRepository(options) {
   if (partialClone) {
     args.push("--filter=blob:none");
   }
-  args.push("--depth", String(depth), "--single-branch");
+  args.push("--depth", String(depth), "--single-branch", "--no-tags");
   if (branch !== void 0) {
     args.push("--branch", branch);
   }
@@ -6202,12 +6459,17 @@ async function cloneRepository(options) {
       branch,
       depth,
       singleBranch: true,
+      noTags: true,
+      lfsSkipSmudge: true,
       partialClone
     },
     "Starting git clone"
   );
   return new Promise((resolve4) => {
-    const git = spawn("git", args, { stdio: ["ignore", "pipe", "pipe"] });
+    const git = spawn("git", args, {
+      stdio: ["ignore", "pipe", "pipe"],
+      env: { ...process.env, GIT_LFS_SKIP_SMUDGE: "1" }
+    });
     let timedOut = false;
     let forceKillTimeout = null;
     const timeout = setTimeout(() => {
@@ -7212,6 +7474,8 @@ var LanceStore = class {
         const schema = LanceSchema({
           id: new Utf8(),
           content: this.embeddingFunction.sourceField(),
+          fts_content: new Utf8(),
+          // content + identifier-split variants for FTS
           vector: this.embeddingFunction.vectorField(),
           metadata: new Utf8()
         });
@@ -7222,6 +7486,7 @@ var LanceStore = class {
           {
             id: "__init__",
             content: "",
+            fts_content: "",
             vector: new Array(this._dimensions).fill(0),
             metadata: "{}"
           }
@@ -7236,13 +7501,25 @@ var LanceStore = class {
   }
   async addDocuments(storeId, documents) {
     const table = await this.getTable(storeId);
-    const lanceDocuments = documents.map((doc) => ({
-      id: doc.id,
-      content: doc.content,
-      vector: [...doc.vector],
-      metadata: JSON.stringify(doc.metadata)
-    }));
-    await table.add(lanceDocuments);
+    const hasFts = await this.hasFtsContentColumn(storeId);
+    if (hasFts) {
+      const lanceDocuments = documents.map((doc) => ({
+        id: doc.id,
+        content: doc.content,
+        fts_content: buildFtsContent(doc.content),
+        vector: [...doc.vector],
+        metadata: JSON.stringify(doc.metadata)
+      }));
+      await table.add(lanceDocuments);
+    } else {
+      const lanceDocuments = documents.map((doc) => ({
+        id: doc.id,
+        content: doc.content,
+        vector: [...doc.vector],
+        metadata: JSON.stringify(doc.metadata)
+      }));
+      await table.add(lanceDocuments);
+    }
   }
   async deleteDocuments(storeId, documentIds) {
     if (documentIds.length === 0) {
@@ -7297,10 +7574,29 @@ var LanceStore = class {
   }
   async createFtsIndex(storeId) {
     const table = await this.getTable(storeId);
-    await table.createIndex("content", {
-      config: lancedb.Index.fts()
+    const ftsColumn = await this.hasFtsContentColumn(storeId) ? "fts_content" : "content";
+    await table.createIndex(ftsColumn, {
+      config: lancedb.Index.fts({
+        stem: true,
+        removeStopWords: false,
+        lowercase: true,
+        language: "English"
+      })
     });
   }
+  /**
+   * Check if a table has the fts_content column (v3 schema).
+   * Tables created before the FTS improvement only have content.
+   */
+  async hasFtsContentColumn(storeId) {
+    try {
+      const table = await this.getTable(storeId);
+      const schema = await table.schema();
+      return schema.fields.some((f) => f.name === "fts_content");
+    } catch {
+      return false;
+    }
+  }
   async fullTextSearch(storeId, query, limit) {
     const table = await this.getTable(storeId);
     const results = await table.search(query, "fts").limit(limit).toArray();
@@ -7637,7 +7933,6 @@ export {
   ASTParser,
   ok,
   err,
-  TEXT_EXTENSIONS,
   classifyWebContentType,
   isFileStoreDefinition,
   isRepoStoreDefinition,
@@ -7650,4 +7945,4 @@ export {
   createServices,
   destroyServices
 };
-//# sourceMappingURL=chunk-L2SC6J4K.js.map
+//# sourceMappingURL=chunk-724FNI27.js.map