@tobilu/qmd 2.0.0 → 2.1.0

package/dist/db.js

@@ -4,6 +4,11 @@
  * Provides a unified Database export that works under both Bun (bun:sqlite)
  * and Node.js (better-sqlite3). The APIs are nearly identical — the main
  * difference is the import path.
+ *
+ * On macOS, Apple's system SQLite is compiled with SQLITE_OMIT_LOAD_EXTENSION,
+ * which prevents loading native extensions like sqlite-vec. When running under
+ * Bun we call Database.setCustomSQLite() to swap in Homebrew's full-featured
+ * SQLite build before creating any database instances.
  */
 export const isBun = typeof globalThis.Bun !== "undefined";
 let _Database;
@@ -11,9 +16,35 @@ let _sqliteVecLoad;
 if (isBun) {
     // Dynamic string prevents tsc from resolving bun:sqlite on Node.js builds
     const bunSqlite = "bun:" + "sqlite";
-    _Database = (await import(/* @vite-ignore */ bunSqlite)).Database;
-    const { getLoadablePath } = await import("sqlite-vec");
-    _sqliteVecLoad = (db) => db.loadExtension(getLoadablePath());
+    const BunDatabase = (await import(/* @vite-ignore */ bunSqlite)).Database;
+    // See: https://bun.com/docs/runtime/sqlite#setcustomsqlite
+    if (process.platform === "darwin") {
+        const homebrewPaths = [
+            "/opt/homebrew/opt/sqlite/lib/libsqlite3.dylib", // Apple Silicon
+            "/usr/local/opt/sqlite/lib/libsqlite3.dylib", // Intel
+        ];
+        for (const p of homebrewPaths) {
+            try {
+                BunDatabase.setCustomSQLite(p);
+                break;
+            }
+            catch { }
+        }
+    }
+    _Database = BunDatabase;
+    // setCustomSQLite may have silently failed — test that extensions actually work.
+    try {
+        const { getLoadablePath } = await import("sqlite-vec");
+        const vecPath = getLoadablePath();
+        const testDb = new BunDatabase(":memory:");
+        testDb.loadExtension(vecPath);
+        testDb.close();
+        _sqliteVecLoad = (db) => db.loadExtension(vecPath);
+    }
+    catch {
+        // Vector search won't work, but BM25 and other operations are unaffected.
+        _sqliteVecLoad = null;
+    }
 }
 else {
     _Database = (await import("better-sqlite3")).default;
@@ -28,7 +59,17 @@ export function openDatabase(path) {
 }
 /**
  * Load the sqlite-vec extension into a database.
+ *
+ * Throws with platform-specific fix instructions when the extension is
+ * unavailable.
  */
 export function loadSqliteVec(db) {
+    if (!_sqliteVecLoad) {
+        const hint = isBun && process.platform === "darwin"
+            ? "On macOS with Bun, install Homebrew SQLite: brew install sqlite\n" +
+                "Or install qmd with npm instead: npm install -g @tobilu/qmd"
+            : "Ensure the sqlite-vec native module is installed correctly.";
+        throw new Error(`sqlite-vec extension is unavailable. ${hint}`);
+    }
     _sqliteVecLoad(db);
 }

package/dist/embedded-skills.d.ts

@@ -0,0 +1,6 @@
+export type EmbeddedSkillFile = {
+    relativePath: string;
+    content: string;
+};
+export declare function getEmbeddedQmdSkillFiles(): EmbeddedSkillFile[];
+export declare function getEmbeddedQmdSkillContent(): string;

package/dist/embedded-skills.js

@@ -0,0 +1,14 @@
+// Generated from skills/qmd source files. Keep this in sync when updating the packaged skill.
+const EMBEDDED_QMD_SKILL_BASE64 = {
+    "SKILL.md": "LS0tCm5hbWU6IHFtZApkZXNjcmlwdGlvbjogU2VhcmNoIG1hcmtkb3duIGtub3dsZWRnZSBiYXNlcywgbm90ZXMsIGFuZCBkb2N1bWVudGF0aW9uIHVzaW5nIFFNRC4gVXNlIHdoZW4gdXNlcnMgYXNrIHRvIHNlYXJjaCBub3RlcywgZmluZCBkb2N1bWVudHMsIG9yIGxvb2sgdXAgaW5mb3JtYXRpb24uCmxpY2Vuc2U6IE1JVApjb21wYXRpYmlsaXR5OiBSZXF1aXJlcyBxbWQgQ0xJIG9yIE1DUCBzZXJ2ZXIuIEluc3RhbGwgdmlhIGBucG0gaW5zdGFsbCAtZyBAdG9iaWx1L3FtZGAuCm1ldGFkYXRhOgogIGF1dGhvcjogdG9iaQogIHZlcnNpb246ICIyLjAuMCIKYWxsb3dlZC10b29sczogQmFzaChxbWQ6KiksIG1jcF9fcW1kX18qCi0tLQoKIyBRTUQgLSBRdWljayBNYXJrZG93biBTZWFyY2gKCkxvY2FsIHNlYXJjaCBlbmdpbmUgZm9yIG1hcmtkb3duIGNvbnRlbnQuCgojIyBTdGF0dXMKCiFgcW1kIHN0YXR1cyAyPi9kZXYvbnVsbCB8fCBlY2hvICJOb3QgaW5zdGFsbGVkOiBucG0gaW5zdGFsbCAtZyBAdG9iaWx1L3FtZCJgCgojIyBNQ1A6IGBxdWVyeWAKCmBgYGpzb24KewogICJzZWFyY2hlcyI6IFsKICAgIHsgInR5cGUiOiAibGV4IiwgInF1ZXJ5IjogIkNBUCB0aGVvcmVtIGNvbnNpc3RlbmN5IiB9LAogICAgeyAidHlwZSI6ICJ2ZWMiLCAicXVlcnkiOiAidHJhZGVvZmYgYmV0d2VlbiBjb25zaXN0ZW5jeSBhbmQgYXZhaWxhYmlsaXR5IiB9CiAgXSwKICAiY29sbGVjdGlvbnMiOiBbImRvY3MiXSwKICAibGltaXQiOiAxMAp9CmBgYAoKIyMjIFF1ZXJ5IFR5cGVzCgp8IFR5cGUgfCBNZXRob2QgfCBJbnB1dCB8CnwtLS0tLS18LS0tLS0tLS18LS0tLS0tLXwKfCBgbGV4YCB8IEJNMjUgfCBLZXl3b3JkcyDigJQgZXhhY3QgdGVybXMsIG5hbWVzLCBjb2RlIHwKfCBgdmVjYCB8IFZlY3RvciB8IFF1ZXN0aW9uIOKAlCBuYXR1cmFsIGxhbmd1YWdlIHwKfCBgaHlkZWAgfCBWZWN0b3IgfCBBbnN3ZXIg4oCUIGh5cG90aGV0aWNhbCByZXN1bHQgKDUwLTEwMCB3b3JkcykgfAoKIyMjIFdyaXRpbmcgR29vZCBRdWVyaWVzCgoqKmxleCAoa2V5d29yZCkqKgotIDItNSB0ZXJtcywgbm8gZmlsbGVyIHdvcmRzCi0gRXhhY3QgcGhyYXNlOiBgImNvbm5lY3Rpb24gcG9vbCJgIChxdW90ZWQpCi0gRXhjbHVkZSB0ZXJtczogYHBlcmZvcm1hbmNlIC1zcG9ydHNgIChtaW51cyBwcmVmaXgpCi0gQ29kZSBpZGVudGlmaWVycyB3b3JrOiBgaGFuZGxlRXJyb3IgYXN5bmNgCgoqKnZlYyAoc2VtYW50aWMpKioKLSBGdWxsIG5hdHVyYWwgbGFuZ3VhZ2UgcXVlc3Rpb24KLSBCZSBzcGVjaWZpYzogYCJob3cgZG9lcyB0aGUgcmF0ZSBsaW1pdGVyIGhhbmRsZSBidXJzdCB0cmFmZmljImAKLSBJbmNsdWRlIGNvbnRleHQ6IGAiaW4gdGhlIHBheW1lbnQgc2VydmljZSwgaG93IGFyZSByZWZ1bmRzIHByb2Nlc3NlZCJgCgoqKmh5ZGUgKGh5cG90aGV0aWNhbCBkb2N1bWVudCkqKgotIFdyaXRlIDUwLTEwMCB3b3JkcyBvZiB3aGF0IHRoZSAqYW5zd2VyKiBsb29rcyBsaWtlCi0gVXNlIHRoZSB2b2NhYnVsYXJ5IHlvdSBleHBlY3QgaW4gdGhlIHJlc3VsdAoKKipleHBhbmQgKGF1dG8tZXhwYW5kKSoqCi0gVXNlIGEgc2luZ2xlLWxpbmUgcXVlcnkgKGltcGxpY2l0KSBvciBgZXhwYW5kOiBxdWVzdGlvbmAgb24gaXRzIG93biBsaW5lCi0gTGV0cyB0aGUgbG9jYWwgTExNIGdlbmVyYXRlIGxleC92ZWMvaHlkZSB2YXJpYXRpb25zCi0gRG8gbm90IG1peCBgZXhwYW5kOmAgd2l0aCBvdGhlciB0eXBlZCBsaW5lcyDigJQgaXQncyBlaXRoZXIgYSBzdGFuZGFsb25lIGV4cGFuZCBxdWVyeSBvciBhIGZ1bGwgcXVlcnkgZG9jdW1lbnQKCiMjIyBJbnRlbnQgKERpc2FtYmlndWF0aW9uKQoKV2hlbiBhIHF1ZXJ5IHRlcm0gaXMgYW1iaWd1b3VzLCBhZGQgYGludGVudGAgdG8gc3RlZXIgcmVzdWx0czoKCmBgYGpzb24KewogICJzZWFyY2hlcyI6IFsKICAgIHsgInR5cGUiOiAibGV4IiwgInF1ZXJ5IjogInBlcmZvcm1hbmNlIiB9CiAgXSwKICAiaW50ZW50IjogIndlYiBwYWdlIGxvYWQgdGltZXMgYW5kIENvcmUgV2ViIFZpdGFscyIKfQpgYGAKCkludGVudCBhZmZlY3RzIGV4cGFuc2lvbiwgcmVyYW5raW5nLCBjaHVuayBzZWxlY3Rpb24sIGFuZCBzbmlwcGV0IGV4dHJhY3Rpb24uIEl0IGRvZXMgbm90IHNlYXJjaCBvbiBpdHMgb3duIOKAlCBpdCdzIGEgc3RlZXJpbmcgc2lnbmFsIHRoYXQgZGlzYW1iaWd1YXRlcyBxdWVyaWVzIGxpa2UgInBlcmZvcm1hbmNlIiAod2ViLXBlcmYgdnMgdGVhbSBoZWFsdGggdnMgZml0bmVzcykuCgojIyMgQ29tYmluaW5nIFR5cGVzCgp8IEdvYWwgfCBBcHByb2FjaCB8CnwtLS0tLS18LS0tLS0tLS0tLXwKfCBLbm93IGV4YWN0IHRlcm1zIHwgYGxleGAgb25seSB8CnwgRG9uJ3Qga25vdyB2b2NhYnVsYXJ5IHwgVXNlIGEgc2luZ2xlLWxpbmUgcXVlcnkgKGltcGxpY2l0IGBleHBhbmQ6YCkgb3IgYHZlY2AgfAp8IEJlc3QgcmVjYWxsIHwgYGxleGAgKyBgdmVjYCB8CnwgQ29tcGxleCB0b3BpYyB8IGBsZXhgICsgYHZlY2AgKyBgaHlkZWAgfAp8IEFtYmlndW91cyBxdWVyeSB8IEFkZCBgaW50ZW50YCB0byBhbnkgY29tYmluYXRpb24gYWJvdmUgfAoKRmlyc3QgcXVlcnkgZ2V0cyAyeCB3ZWlnaHQgaW4gZnVzaW9uIOKAlCBwdXQgeW91ciBiZXN0IGd1ZXNzIGZpcnN0LgoKIyMjIExleCBRdWVyeSBTeW50YXgKCnwgU3ludGF4IHwgTWVhbmluZyB8IEV4YW1wbGUgfAp8LS0tLS0tLS18LS0tLS0tLS0tfC0tLS0tLS0tLXwKfCBgdGVybWAgfCBQcmVmaXggbWF0Y2ggfCBgcGVyZmAgbWF0Y2hlcyAicGVyZm9ybWFuY2UiIHwKfCBgInBocmFzZSJgIHwgRXhhY3QgcGhyYXNlIHwgYCJyYXRlIGxpbWl0ZXIiYCB8CnwgYC10ZXJtYCB8IEV4Y2x1ZGUgfCBgcGVyZm9ybWFuY2UgLXNwb3J0c2AgfAoKTm90ZTogYC10ZXJtYCBvbmx5IHdvcmtzIGluIGxleCBxdWVyaWVzLCBub3QgdmVjL2h5ZGUuCgojIyMgQ29sbGVjdGlvbiBGaWx0ZXJpbmcKCmBgYGpzb24KeyAiY29sbGVjdGlvbnMiOiBbImRvY3MiXSB9ICAgICAgICAgICAgICAvLyBTaW5nbGUKeyAiY29sbGVjdGlvbnMiOiBbImRvY3MiLCAibm90ZXMiXSB9ICAgICAvLyBNdWx0aXBsZSAoT1IpCmBgYAoKT21pdCB0byBzZWFyY2ggYWxsIGNvbGxlY3Rpb25zLgoKIyMgT3RoZXIgTUNQIFRvb2xzCgp8IFRvb2wgfCBVc2UgfAp8LS0tLS0tfC0tLS0tfAp8IGBnZXRgIHwgUmV0cmlldmUgZG9jIGJ5IHBhdGggb3IgYCNkb2NpZGAgfAp8IGBtdWx0aV9nZXRgIHwgUmV0cmlldmUgbXVsdGlwbGUgYnkgZ2xvYi9saXN0IHwKfCBgc3RhdHVzYCB8IENvbGxlY3Rpb25zIGFuZCBoZWFsdGggfAoKIyMgQ0xJCgpgYGBiYXNoCnFtZCBxdWVyeSAicXVlc3Rpb24iICAgICAgICAgICAgICAjIEF1dG8tZXhwYW5kICsgcmVyYW5rCnFtZCBxdWVyeSAkJ2xleDogWFxudmVjOiBZJyAgICAgICAjIFN0cnVjdHVyZWQKcW1kIHF1ZXJ5ICQnZXhwYW5kOiBxdWVzdGlvbicgICAgICMgRXhwbGljaXQgZXhwYW5kCnFtZCBxdWVyeSAtLWpzb24gLS1leHBsYWluICJxIiAgICAjIFNob3cgc2NvcmUgdHJhY2VzIChSUkYgKyByZXJhbmsgYmxlbmQpCnFtZCBzZWFyY2ggImtleXdvcmRzIiAgICAgICAgICAgICAjIEJNMjUgb25seSAobm8gTExNKQpxbWQgZ2V0ICIjYWJjMTIzIiAgICAgICAgICAgICAgICAgIyBCeSBkb2NpZApxbWQgbXVsdGktZ2V0ICJqb3VybmFscy8yMDI2LSoubWQiIC1sIDQwICAjIEJhdGNoIHB1bGwgc25pcHBldHMgYnkgZ2xvYgpxbWQgbXVsdGktZ2V0IG5vdGVzL2Zvby5tZCxub3Rlcy9iYXIubWQgICAjIENvbW1hLXNlcGFyYXRlZCBsaXN0LCBwcmVzZXJ2ZXMgb3JkZXIKYGBgCgojIyBIVFRQIEFQSQoKYGBgYmFzaApjdXJsIC1YIFBPU1QgaHR0cDovL2xvY2FsaG9zdDo4MTgxL3F1ZXJ5IFwKICAtSCAiQ29udGVudC1UeXBlOiBhcHBsaWNhdGlvbi9qc29uIiBcCiAgLWQgJ3sic2VhcmNoZXMiOiBbeyJ0eXBlIjogImxleCIsICJxdWVyeSI6ICJ0ZXN0In1dfScKYGBgCgojIyBTZXR1cAoKYGBgYmFzaApucG0gaW5zdGFsbCAtZyBAdG9iaWx1L3FtZApxbWQgY29sbGVjdGlvbiBhZGQgfi9ub3RlcyAtLW5hbWUgbm90ZXMKcW1kIGVtYmVkCmBgYAo=",
+    "references/mcp-setup.md": "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"
+};
+export function getEmbeddedQmdSkillFiles() {
+    return Object.entries(EMBEDDED_QMD_SKILL_BASE64).map(([relativePath, encoded]) => ({
+        relativePath,
+        content: Buffer.from(encoded, 'base64').toString('utf8'),
+    }));
+}
+export function getEmbeddedQmdSkillContent() {
+    return Buffer.from(EMBEDDED_QMD_SKILL_BASE64["SKILL.md"], "base64").toString("utf8");
+}

package/dist/index.d.ts

@@ -16,11 +16,12 @@
  *   const results = await store.search({ query: "how does auth work?" })
  *   await store.close()
  */
-import { extractSnippet, addLineNumbers, DEFAULT_MULTI_GET_MAX_BYTES, type Store as InternalStore, type DocumentResult, type DocumentNotFound, type SearchResult, type HybridQueryResult, type HybridQueryOptions, type HybridQueryExplain, type ExpandedQuery, type StructuredSearchOptions, type MultiGetResult, type IndexStatus, type IndexHealthInfo, type SearchHooks, type ReindexProgress, type ReindexResult, type EmbedProgress, type EmbedResult } from "./store.js";
+import { extractSnippet, addLineNumbers, DEFAULT_MULTI_GET_MAX_BYTES, type Store as InternalStore, type DocumentResult, type DocumentNotFound, type SearchResult, type HybridQueryResult, type HybridQueryOptions, type HybridQueryExplain, type ExpandedQuery, type StructuredSearchOptions, type MultiGetResult, type IndexStatus, type IndexHealthInfo, type SearchHooks, type ReindexProgress, type ReindexResult, type EmbedProgress, type EmbedResult, type ChunkStrategy } from "./store.js";
 import { type Collection, type CollectionConfig, type NamedCollection, type ContextMap } from "./collections.js";
 export type { DocumentResult, DocumentNotFound, SearchResult, HybridQueryResult, HybridQueryOptions, HybridQueryExplain, ExpandedQuery, StructuredSearchOptions, MultiGetResult, IndexStatus, IndexHealthInfo, SearchHooks, ReindexProgress, ReindexResult, EmbedProgress, EmbedResult, Collection, CollectionConfig, NamedCollection, ContextMap, };
 export type { InternalStore };
 export { extractSnippet, addLineNumbers, DEFAULT_MULTI_GET_MAX_BYTES };
+export type { ChunkStrategy } from "./store.js";
 export { getDefaultDbPath } from "./store.js";
 export { Maintenance } from "./maintenance.js";
 /**
@@ -65,6 +66,8 @@ export interface SearchOptions {
     minScore?: number;
     /** Include explain traces */
     explain?: boolean;
+    /** Chunk strategy: "auto" (default, uses AST for code files) or "regex" (legacy) */
+    chunkStrategy?: ChunkStrategy;
 }
 /**
  * Options for searchLex() — BM25 keyword search.
@@ -183,6 +186,9 @@ export interface QMDStore {
     embed(options?: {
         force?: boolean;
         model?: string;
+        maxDocsPerBatch?: number;
+        maxBatchBytes?: number;
+        chunkStrategy?: ChunkStrategy;
         onProgress?: (info: EmbedProgress) => void;
     }): Promise<EmbedResult>;
     /** Get index status (document counts, collections, embedding state) */

package/dist/index.js

@@ -19,7 +19,7 @@
 import { createStore as createStoreInternal, hybridQuery, structuredSearch, extractSnippet, addLineNumbers, DEFAULT_EMBED_MODEL, DEFAULT_MULTI_GET_MAX_BYTES, reindexCollection, generateEmbeddings, listCollections as storeListCollections, syncConfigToDb, getStoreCollections, getStoreCollection, getStoreGlobalContext, getStoreContexts, upsertStoreCollection, deleteStoreCollection, renameStoreCollection, updateStoreContext, removeStoreContext, setStoreGlobalContext, vacuumDatabase, cleanupOrphanedContent, cleanupOrphanedVectors, deleteLLMCache, deleteInactiveDocuments, clearAllEmbeddings, } from "./store.js";
 import { LlamaCpp, } from "./llm.js";
 import { setConfigSource, loadConfig, addCollection as collectionsAddCollection, removeCollection as collectionsRemoveCollection, renameCollection as collectionsRenameCollection, addContext as collectionsAddContext, removeContext as collectionsRemoveContext, setGlobalContext as collectionsSetGlobalContext, } from "./collections.js";
-// Re-export utility functions used by frontends
+// Re-export utility functions and types used by frontends
 export { extractSnippet, addLineNumbers, DEFAULT_MULTI_GET_MAX_BYTES };
 // Re-export getDefaultDbPath for CLI/MCP that need the default database location
 export { getDefaultDbPath } from "./store.js";
@@ -63,21 +63,26 @@ export async function createStore(options) {
     // Track whether we have a YAML config path for write-through
     const hasYamlConfig = !!options.configPath;
     // Sync config into SQLite store_collections
+    let config;
     if (options.configPath) {
         // YAML mode: inject config source for write-through, sync to DB
         setConfigSource({ configPath: options.configPath });
-        const config = loadConfig();
+        config = loadConfig();
         syncConfigToDb(db, config);
     }
     else if (options.config) {
         // Inline config mode: inject config source for mutations, sync to DB
         setConfigSource({ config: options.config });
-        syncConfigToDb(db, options.config);
+        config = options.config;
+        syncConfigToDb(db, config);
     }
     // else: DB-only mode — no external config, use existing store_collections
     // Create a per-store LlamaCpp instance — lazy-loads models on first use,
     // auto-unloads after 5 min inactivity to free VRAM.
     const llm = new LlamaCpp({
+        embedModel: config?.models?.embed,
+        generateModel: config?.models?.generate,
+        rerankModel: config?.models?.rerank,
         inactivityTimeoutMs: 5 * 60 * 1000,
         disposeModelsOnInactivity: true,
     });
@@ -105,6 +110,7 @@ export async function createStore(options) {
                     explain: opts.explain,
                     intent: opts.intent,
                     skipRerank,
+                    chunkStrategy: opts.chunkStrategy,
                 });
             }
             // Simple query string — use hybridQuery (expand + search + rerank)
@@ -115,6 +121,7 @@ export async function createStore(options) {
                 explain: opts.explain,
                 intent: opts.intent,
                 skipRerank,
+                chunkStrategy: opts.chunkStrategy,
             });
         },
         searchLex: async (q, opts) => internal.searchFTS(q, opts?.limit, opts?.collection),
@@ -210,6 +217,9 @@ export async function createStore(options) {
             return generateEmbeddings(internal, {
                 force: embedOpts?.force,
                 model: embedOpts?.model,
+                maxDocsPerBatch: embedOpts?.maxDocsPerBatch,
+                maxBatchBytes: embedOpts?.maxBatchBytes,
+                chunkStrategy: embedOpts?.chunkStrategy,
                 onProgress: embedOpts?.onProgress,
             });
         },

package/dist/llm.d.ts

@@ -105,7 +105,7 @@ export type LLMSessionOptions = {
  */
 export interface ILLMSession {
     embed(text: string, options?: EmbedOptions): Promise<EmbeddingResult | null>;
-    embedBatch(texts: string[]): Promise<(EmbeddingResult | null)[]>;
+    embedBatch(texts: string[], options?: EmbedOptions): Promise<(EmbeddingResult | null)[]>;
     expandQuery(query: string, options?: {
         context?: string;
         includeLexical?: boolean;
@@ -137,7 +137,7 @@ export type RerankDocument = {
 };
 export declare const LFM2_GENERATE_MODEL = "hf:LiquidAI/LFM2-1.2B-GGUF/LFM2-1.2B-Q4_K_M.gguf";
 export declare const LFM2_INSTRUCT_MODEL = "hf:LiquidAI/LFM2.5-1.2B-Instruct-GGUF/LFM2.5-1.2B-Instruct-Q4_K_M.gguf";
-export declare const DEFAULT_EMBED_MODEL_URI: string;
+export declare const DEFAULT_EMBED_MODEL_URI = "hf:ggml-org/embeddinggemma-300M-GGUF/embeddinggemma-300M-Q8_0.gguf";
 export declare const DEFAULT_RERANK_MODEL_URI = "hf:ggml-org/Qwen3-Reranker-0.6B-Q8_0-GGUF/qwen3-reranker-0.6b-q8_0.gguf";
 export declare const DEFAULT_GENERATE_MODEL_URI = "hf:tobil/qmd-query-expansion-1.7B-gguf/qmd-query-expansion-1.7B-q4_k_m.gguf";
 export declare const DEFAULT_MODEL_CACHE_DIR: string;
@@ -232,6 +232,7 @@ export declare class LlamaCpp implements LLM {
     private disposeModelsOnInactivity;
     private disposed;
     constructor(config?: LlamaCppConfig);
+    get embedModelName(): string;
     /**
      * Reset the inactivity timer. Called after each model operation.
      * When timer fires, models are unloaded to free memory (if no active sessions).
@@ -306,6 +307,7 @@ export declare class LlamaCpp implements LLM {
      * - Combined: drops from 11.6 GB (auto, no flash) to 568 MB per context (20×)
      */
     private static readonly RERANK_CONTEXT_SIZE;
+    private static readonly EMBED_CONTEXT_SIZE;
     private ensureRerankContexts;
     /**
      * Tokenize text using the embedding model's tokenizer
@@ -320,12 +322,19 @@ export declare class LlamaCpp implements LLM {
      * Detokenize token IDs back to text
      */
     detokenize(tokens: readonly LlamaToken[]): Promise<string>;
+    /**
+     * Truncate text to fit within the embedding model's context window.
+     * Uses the model's own tokenizer for accurate token counting, then
+     * detokenizes back to text if truncation is needed.
+     * Returns the (possibly truncated) text and whether truncation occurred.
+     */
+    private truncateToContextSize;
     embed(text: string, options?: EmbedOptions): Promise<EmbeddingResult | null>;
     /**
      * Batch embed multiple texts efficiently
      * Uses Promise.all for parallel embedding - node-llama-cpp handles batching internally
      */
-    embedBatch(texts: string[]): Promise<(EmbeddingResult | null)[]>;
+    embedBatch(texts: string[], options?: EmbedOptions): Promise<(EmbeddingResult | null)[]>;
     generate(prompt: string, options?: GenerateOptions): Promise<GenerateResult | null>;
     modelExists(modelUri: string): Promise<ModelInfo>;
     expandQuery(query: string, options?: {

package/dist/llm.js

@@ -47,8 +47,8 @@ export function formatDocForEmbedding(text, title, modelUri) {
 // =============================================================================
 // HuggingFace model URIs for node-llama-cpp
 // Format: hf:<user>/<repo>/<file>
-// Override via QMD_EMBED_MODEL env var (e.g. hf:Qwen/Qwen3-Embedding-0.6B-GGUF/qwen3-embedding-0.6b-q8_0.gguf)
-const DEFAULT_EMBED_MODEL = process.env.QMD_EMBED_MODEL ?? "hf:ggml-org/embeddinggemma-300M-GGUF/embeddinggemma-300M-Q8_0.gguf";
+// Override via QMD_EMBED_MODEL env var (e.g. hf:Qwen/Qwen3-Embedding-0.6B-GGUF/Qwen3-Embedding-0.6B-Q8_0.gguf)
+const DEFAULT_EMBED_MODEL = "hf:ggml-org/embeddinggemma-300M-GGUF/embeddinggemma-300M-Q8_0.gguf";
 const DEFAULT_RERANK_MODEL = "hf:ggml-org/Qwen3-Reranker-0.6B-Q8_0-GGUF/qwen3-reranker-0.6b-q8_0.gguf";
 // const DEFAULT_GENERATE_MODEL = "hf:ggml-org/Qwen3-0.6B-GGUF/Qwen3-0.6B-Q8_0.gguf";
 const DEFAULT_GENERATE_MODEL = "hf:tobil/qmd-query-expansion-1.7B-gguf/qmd-query-expansion-1.7B-q4_k_m.gguf";
@@ -61,7 +61,9 @@ export const DEFAULT_EMBED_MODEL_URI = DEFAULT_EMBED_MODEL;
 export const DEFAULT_RERANK_MODEL_URI = DEFAULT_RERANK_MODEL;
 export const DEFAULT_GENERATE_MODEL_URI = DEFAULT_GENERATE_MODEL;
 // Local model cache directory
-const MODEL_CACHE_DIR = join(homedir(), ".cache", "qmd", "models");
+const MODEL_CACHE_DIR = process.env.XDG_CACHE_HOME
+    ? join(process.env.XDG_CACHE_HOME, "qmd", "models")
+    : join(homedir(), ".cache", "qmd", "models");
 export const DEFAULT_MODEL_CACHE_DIR = MODEL_CACHE_DIR;
 function parseHfUri(model) {
     if (!model.startsWith("hf:"))
@@ -187,14 +189,17 @@ export class LlamaCpp {
     // Track disposal state to prevent double-dispose
     disposed = false;
     constructor(config = {}) {
-        this.embedModelUri = config.embedModel || DEFAULT_EMBED_MODEL;
-        this.generateModelUri = config.generateModel || DEFAULT_GENERATE_MODEL;
-        this.rerankModelUri = config.rerankModel || DEFAULT_RERANK_MODEL;
+        this.embedModelUri = config.embedModel || process.env.QMD_EMBED_MODEL || DEFAULT_EMBED_MODEL;
+        this.generateModelUri = config.generateModel || process.env.QMD_GENERATE_MODEL || DEFAULT_GENERATE_MODEL;
+        this.rerankModelUri = config.rerankModel || process.env.QMD_RERANK_MODEL || DEFAULT_RERANK_MODEL;
         this.modelCacheDir = config.modelCacheDir || MODEL_CACHE_DIR;
         this.expandContextSize = resolveExpandContextSize(config.expandContextSize);
         this.inactivityTimeoutMs = config.inactivityTimeoutMs ?? DEFAULT_INACTIVITY_TIMEOUT_MS;
         this.disposeModelsOnInactivity = config.disposeModelsOnInactivity ?? false;
     }
+    get embedModelName() {
+        return this.embedModelUri;
+    }
     /**
      * Reset the inactivity timer. Called after each model operation.
      * When timer fires, models are unloaded to free memory (if no active sessions).
@@ -289,11 +294,29 @@ export class LlamaCpp {
      */
     async ensureLlama() {
         if (!this.llama) {
-            const llama = await getLlama({
-                // attempt to build
+            // Allow override via QMD_LLAMA_GPU: "false" | "off" | "none" forces CPU
+            const gpuOverride = (process.env.QMD_LLAMA_GPU ?? "").toLowerCase();
+            const forceCpu = ["false", "off", "none", "disable", "disabled", "0"].includes(gpuOverride);
+            const loadLlama = async (gpu) => await getLlama({
                 build: "autoAttempt",
-                logLevel: LlamaLogLevel.error
+                logLevel: LlamaLogLevel.error,
+                gpu,
             });
+            let llama;
+            if (forceCpu) {
+                llama = await loadLlama(false);
+            }
+            else {
+                try {
+                    llama = await loadLlama("auto");
+                }
+                catch (err) {
+                    // GPU backend (e.g. Vulkan on headless/driverless machines) can throw at init.
+                    // Fall back to CPU so qmd still works.
+                    process.stderr.write(`QMD Warning: GPU init failed (${err instanceof Error ? err.message : String(err)}), falling back to CPU.\n`);
+                    llama = await loadLlama(false);
+                }
+            }
             if (llama.gpu === false) {
                 process.stderr.write("QMD Warning: no GPU acceleration, running on CPU (slow). Run 'qmd status' for details.\n");
             }
@@ -394,6 +417,7 @@ export class LlamaCpp {
             for (let i = 0; i < n; i++) {
                 try {
                     this.embedContexts.push(await model.createEmbeddingContext({
+                        contextSize: LlamaCpp.EMBED_CONTEXT_SIZE,
                         ...(threads > 0 ? { threads } : {}),
                     }));
                 }
@@ -484,9 +508,20 @@ export class LlamaCpp {
      * - Combined: drops from 11.6 GB (auto, no flash) to 568 MB per context (20×)
      */
     // Qwen3 reranker template adds ~200 tokens overhead (system prompt, tags, etc.)
-    // Chunks are max 800 tokens, so 800 + 200 + query ≈ 1100 tokens typical.
-    // Use 2048 for safety margin. Still 17× less than auto (40960).
-    static RERANK_CONTEXT_SIZE = 2048;
+    // Default 2048 was too small for longer documents (e.g. session transcripts,
+    // CJK text, or large markdown files) — callers hit "input lengths exceed
+    // context size" errors even after truncation because the overhead estimate
+    // was insufficient.  4096 comfortably fits the largest real-world chunks
+    // while staying well below the 40 960-token auto size.
+    // Override with QMD_RERANK_CONTEXT_SIZE env var if you need more headroom.
+    static RERANK_CONTEXT_SIZE = (() => {
+        const v = parseInt(process.env.QMD_RERANK_CONTEXT_SIZE ?? "", 10);
+        return Number.isFinite(v) && v > 0 ? v : 4096;
+    })();
+    static EMBED_CONTEXT_SIZE = (() => {
+        const v = parseInt(process.env.QMD_EMBED_CONTEXT_SIZE ?? "", 10);
+        return Number.isFinite(v) && v > 0 ? v : 2048;
+    })();
     async ensureRerankContexts() {
         if (this.rerankContexts.length === 0) {
             const model = await this.ensureRerankModel();
@@ -555,15 +590,41 @@ export class LlamaCpp {
     // ==========================================================================
     // Core API methods
     // ==========================================================================
+    /**
+     * Truncate text to fit within the embedding model's context window.
+     * Uses the model's own tokenizer for accurate token counting, then
+     * detokenizes back to text if truncation is needed.
+     * Returns the (possibly truncated) text and whether truncation occurred.
+     */
+    async truncateToContextSize(text) {
+        if (!this.embedModel)
+            return { text, truncated: false };
+        const maxTokens = this.embedModel.trainContextSize;
+        if (maxTokens <= 0)
+            return { text, truncated: false };
+        const tokens = this.embedModel.tokenize(text);
+        if (tokens.length <= maxTokens)
+            return { text, truncated: false };
+        // Leave a small margin (4 tokens) for BOS/EOS overhead
+        const safeLimit = Math.max(1, maxTokens - 4);
+        const truncatedTokens = tokens.slice(0, safeLimit);
+        const truncatedText = this.embedModel.detokenize(truncatedTokens);
+        return { text: truncatedText, truncated: true };
+    }
     async embed(text, options = {}) {
         // Ping activity at start to keep models alive during this operation
         this.touchActivity();
         try {
             const context = await this.ensureEmbedContext();
-            const embedding = await context.getEmbeddingFor(text);
+            // Guard: truncate text that exceeds model context window to prevent GGML crash
+            const { text: safeText, truncated } = await this.truncateToContextSize(text);
+            if (truncated) {
+                console.warn(`⚠ Text truncated to fit embedding context (${this.embedModel?.trainContextSize} tokens)`);
+            }
+            const embedding = await context.getEmbeddingFor(safeText);
             return {
                 embedding: Array.from(embedding.vector),
-                model: this.embedModelUri,
+                model: options.model ?? this.embedModelUri,
             };
         }
         catch (error) {
@@ -575,7 +636,7 @@ export class LlamaCpp {
      * Batch embed multiple texts efficiently
      * Uses Promise.all for parallel embedding - node-llama-cpp handles batching internally
      */
-    async embedBatch(texts) {
+    async embedBatch(texts, options = {}) {
         if (this._ciMode)
             throw new Error("LLM operations are disabled in CI (set CI=true)");
         // Ping activity at start to keep models alive during this operation
@@ -591,9 +652,13 @@ export class LlamaCpp {
                 const embeddings = [];
                 for (const text of texts) {
                     try {
-                        const embedding = await context.getEmbeddingFor(text);
+                        const { text: safeText, truncated } = await this.truncateToContextSize(text);
+                        if (truncated) {
+                            console.warn(`⚠ Batch text truncated to fit embedding context (${this.embedModel?.trainContextSize} tokens)`);
+                        }
+                        const embedding = await context.getEmbeddingFor(safeText);
                         this.touchActivity();
-                        embeddings.push({ embedding: Array.from(embedding.vector), model: this.embedModelUri });
+                        embeddings.push({ embedding: Array.from(embedding.vector), model: options.model ?? this.embedModelUri });
                     }
                     catch (err) {
                         console.error("Embedding error for text:", err);
@@ -610,9 +675,13 @@ export class LlamaCpp {
                 const results = [];
                 for (const text of chunk) {
                     try {
-                        const embedding = await ctx.getEmbeddingFor(text);
+                        const { text: safeText, truncated } = await this.truncateToContextSize(text);
+                        if (truncated) {
+                            console.warn(`⚠ Batch text truncated to fit embedding context (${this.embedModel?.trainContextSize} tokens)`);
+                        }
+                        const embedding = await ctx.getEmbeddingFor(safeText);
                         this.touchActivity();
-                        results.push({ embedding: Array.from(embedding.vector), model: this.embedModelUri });
+                        results.push({ embedding: Array.from(embedding.vector), model: options.model ?? this.embedModelUri });
                     }
                     catch (err) {
                         console.error("Embedding error for text:", err);
@@ -767,8 +836,10 @@ export class LlamaCpp {
             await genContext.dispose();
         }
     }
-    // Qwen3 reranker chat template overhead (system prompt, tags, separators)
-    static RERANK_TEMPLATE_OVERHEAD = 200;
+    // Qwen3 reranker chat template overhead (system prompt, tags, separators).
+    // Measured at ~350 tokens on real queries; use 512 as a safe upper bound so
+    // the truncation budget never lets a document slip past the context limit.
+    static RERANK_TEMPLATE_OVERHEAD = 512;
     static RERANK_TARGET_DOCS_PER_CONTEXT = 10;
     async rerank(query, documents, options = {}) {
         if (this._ciMode)
@@ -1028,8 +1099,8 @@ class LLMSession {
     async embed(text, options) {
         return this.withOperation(() => this.manager.getLlamaCpp().embed(text, options));
     }
-    async embedBatch(texts) {
-        return this.withOperation(() => this.manager.getLlamaCpp().embedBatch(texts));
+    async embedBatch(texts, options) {
+        return this.withOperation(() => this.manager.getLlamaCpp().embedBatch(texts, options));
     }
     async expandQuery(query, options) {
         return this.withOperation(() => this.manager.getLlamaCpp().expandQuery(query, options));
@@ -1106,8 +1177,7 @@ let defaultLlamaCpp = null;
  */
 export function getDefaultLlamaCpp() {
     if (!defaultLlamaCpp) {
-        const embedModel = process.env.QMD_EMBED_MODEL;
-        defaultLlamaCpp = new LlamaCpp(embedModel ? { embedModel } : {});
+        defaultLlamaCpp = new LlamaCpp();
     }
     return defaultLlamaCpp;
 }

package/dist/mcp/server.js

@@ -8,13 +8,17 @@
  */
 import { createServer } from "node:http";
 import { randomUUID } from "node:crypto";
+import { readFileSync } from "node:fs";
+import { join, dirname } from "node:path";
 import { fileURLToPath } from "url";
 import { McpServer, ResourceTemplate } from "@modelcontextprotocol/sdk/server/mcp.js";
 import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
 import { WebStandardStreamableHTTPServerTransport } from "@modelcontextprotocol/sdk/server/webStandardStreamableHttp.js";
 import { isInitializeRequest } from "@modelcontextprotocol/sdk/types.js";
 import { z } from "zod";
+import { existsSync } from "fs";
 import { createStore, extractSnippet, addLineNumbers, getDefaultDbPath, DEFAULT_MULTI_GET_MAX_BYTES, } from "../index.js";
+import { getConfigPath } from "../collections.js";
 // =============================================================================
 // Helper functions
 // =============================================================================
@@ -39,6 +43,16 @@ function formatSearchSummary(results, query) {
     }
     return lines.join('\n');
 }
+function getPackageVersion() {
+    try {
+        const pkgPath = join(dirname(fileURLToPath(import.meta.url)), "../../package.json");
+        const pkg = JSON.parse(readFileSync(pkgPath, "utf-8"));
+        return pkg.version ?? "unknown";
+    }
+    catch {
+        return "unknown";
+    }
+}
 // =============================================================================
 // MCP Server
 // =============================================================================
@@ -108,7 +122,7 @@ async function buildInstructions(store) {
  * Shared by both stdio and HTTP transports.
  */
 async function createMcpServer(store) {
-    const server = new McpServer({ name: "qmd", version: "0.9.9" }, { instructions: await buildInstructions(store) });
+    const server = new McpServer({ name: "qmd", version: getPackageVersion() }, { instructions: await buildInstructions(store) });
     // Pre-fetch default collection names for search tools
     const defaultCollectionNames = await store.getDefaultCollectionNames();
     // ---------------------------------------------------------------------------
@@ -218,8 +232,9 @@ Intent-aware lex (C++ performance, not sports):
             candidateLimit: z.number().optional().describe("Maximum candidates to rerank (default: 40, lower = faster but may miss results)"),
             collections: z.array(z.string()).optional().describe("Filter to collections (OR match)"),
             intent: z.string().optional().describe("Background context to disambiguate the query. Example: query='performance', intent='web page load times and Core Web Vitals'. Does not search on its own."),
+            rerank: z.boolean().optional().default(true).describe("Rerank results using LLM (default: true). Set to false for faster results on CPU-only machines."),
         },
-    }, async ({ searches, limit, minScore, candidateLimit, collections, intent }) => {
+    }, async ({ searches, limit, minScore, candidateLimit, collections, intent, rerank }) => {
         // Map to internal format
         const queries = searches.map(s => ({
             type: s.type,
@@ -232,6 +247,7 @@ Intent-aware lex (C++ performance, not sports):
             collections: effectiveCollections.length > 0 ? effectiveCollections : undefined,
             limit,
             minScore,
+            rerank,
             intent,
         });
         // Use first lex or vec query for snippet extraction
@@ -387,7 +403,7 @@ Intent-aware lex (C++ performance, not sports):
             `  Collections: ${status.collections.length}`,
         ];
         for (const col of status.collections) {
-            summary.push(`    - ${col.path} (${col.documents} docs)`);
+            summary.push(`    - ${col.name}: ${col.path} (${col.documents} docs)`);
         }
         return {
             content: [{ type: "text", text: summary.join('\n') }],
@@ -400,7 +416,11 @@ Intent-aware lex (C++ performance, not sports):
 // Transport: stdio (default)
 // =============================================================================
 export async function startMcpServer() {
-    const store = await createStore({ dbPath: getDefaultDbPath() });
+    const configPath = getConfigPath();
+    const store = await createStore({
+        dbPath: getDefaultDbPath(),
+        ...(existsSync(configPath) ? { configPath } : {}),
+    });
     const server = await createMcpServer(store);
     const transport = new StdioServerTransport();
     await server.connect(transport);
@@ -410,7 +430,11 @@ export async function startMcpServer() {
  * Binds to localhost only. Returns a handle for shutdown and port discovery.
  */
 export async function startMcpHttpServer(port, options) {
-    const store = await createStore({ dbPath: getDefaultDbPath() });
+    const configPath = getConfigPath();
+    const store = await createStore({
+        dbPath: getDefaultDbPath(),
+        ...(existsSync(configPath) ? { configPath } : {}),
+    });
     // Pre-fetch default collection names for REST endpoint
     const defaultCollectionNames = await store.getDefaultCollectionNames();
     // Session map: each client gets its own McpServer + Transport pair (MCP spec requirement).