@velvetmonkey/flywheel-memory 2.0.60 → 2.0.62

package/dist/index.js

@@ -1816,7 +1816,23 @@ init_levenshtein();
 import * as crypto from "crypto";
 import * as fs3 from "fs";
 import * as path2 from "path";
-var MODEL_ID = "Xenova/all-MiniLM-L6-v2";
+var MODEL_REGISTRY = {
+  "Xenova/all-MiniLM-L6-v2": { id: "Xenova/all-MiniLM-L6-v2", dims: 384 },
+  "Xenova/bge-small-en-v1.5": { id: "Xenova/bge-small-en-v1.5", dims: 384 },
+  "Xenova/all-MiniLM-L12-v2": { id: "Xenova/all-MiniLM-L12-v2", dims: 384 },
+  "nomic-ai/nomic-embed-text-v1": { id: "nomic-ai/nomic-embed-text-v1", dims: 768 }
+};
+var DEFAULT_MODEL = "Xenova/all-MiniLM-L6-v2";
+function getModelConfig() {
+  const envModel = process.env.EMBEDDING_MODEL?.trim();
+  if (!envModel) return MODEL_REGISTRY[DEFAULT_MODEL];
+  if (MODEL_REGISTRY[envModel]) return MODEL_REGISTRY[envModel];
+  return { id: envModel, dims: 0 };
+}
+var activeModelConfig = getModelConfig();
+function getActiveModelId() {
+  return activeModelConfig.id;
+}
 var EXCLUDED_DIRS2 = /* @__PURE__ */ new Set([
   ".obsidian",
   ".trash",
@@ -1852,9 +1868,14 @@ async function initEmbeddings() {
   initPromise = (async () => {
     try {
       const transformers = await Function("specifier", "return import(specifier)")("@huggingface/transformers");
-      pipeline = await transformers.pipeline("feature-extraction", MODEL_ID, {
+      pipeline = await transformers.pipeline("feature-extraction", activeModelConfig.id, {
         dtype: "fp32"
       });
+      if (activeModelConfig.dims === 0) {
+        const probe = await pipeline("test", { pooling: "mean", normalize: true });
+        activeModelConfig.dims = probe.data.length;
+        console.error(`[Semantic] Probed model ${activeModelConfig.id}: ${activeModelConfig.dims} dims`);
+      }
     } catch (err) {
       initPromise = null;
       if (err instanceof Error && (err.message.includes("Cannot find package") || err.message.includes("MODULE_NOT_FOUND") || err.message.includes("Cannot find module") || err.message.includes("ERR_MODULE_NOT_FOUND"))) {
@@ -1927,7 +1948,7 @@ async function buildEmbeddingsIndex(vaultPath2, onProgress) {
       }
       const embedding = await embedText(content);
       const buf = Buffer.from(embedding.buffer, embedding.byteOffset, embedding.byteLength);
-      upsert.run(file.path, buf, hash, MODEL_ID, Date.now());
+      upsert.run(file.path, buf, hash, activeModelConfig.id, Date.now());
     } catch {
       progress.skipped++;
     }
@@ -1956,7 +1977,7 @@ async function updateEmbedding(notePath, absolutePath) {
     db.prepare(`
       INSERT OR REPLACE INTO note_embeddings (path, embedding, content_hash, model, updated_at)
       VALUES (?, ?, ?, ?, ?)
-    `).run(notePath, buf, hash, MODEL_ID, Date.now());
+    `).run(notePath, buf, hash, activeModelConfig.id, Date.now());
   } catch {
   }
 }
@@ -2058,6 +2079,20 @@ function hasEmbeddingsIndex() {
     return false;
   }
 }
+function getStoredEmbeddingModel() {
+  if (!db) return null;
+  try {
+    const row = db.prepare("SELECT model FROM note_embeddings LIMIT 1").get();
+    return row?.model ?? null;
+  } catch {
+    return null;
+  }
+}
+function needsEmbeddingRebuild() {
+  const stored = getStoredEmbeddingModel();
+  if (stored === null) return false;
+  return stored !== activeModelConfig.id;
+}
 function getEmbeddingsCount() {
   if (!db) return 0;
   try {
@@ -2129,7 +2164,7 @@ async function buildEntityEmbeddingsIndex(vaultPath2, entities, onProgress) {
       }
       const embedding = await embedTextCached(text);
       const buf = Buffer.from(embedding.buffer, embedding.byteOffset, embedding.byteLength);
-      upsert.run(name, buf, hash, MODEL_ID, Date.now());
+      upsert.run(name, buf, hash, activeModelConfig.id, Date.now());
       updated++;
     } catch {
     }
@@ -2156,7 +2191,7 @@ async function updateEntityEmbedding(entityName, entity, vaultPath2) {
     db.prepare(`
       INSERT OR REPLACE INTO entity_embeddings (entity_name, embedding, source_hash, model, updated_at)
       VALUES (?, ?, ?, ?, ?)
-    `).run(entityName, buf, hash, MODEL_ID, Date.now());
+    `).run(entityName, buf, hash, activeModelConfig.id, Date.now());
     entityEmbeddingsMap.set(entityName, embedding);
   } catch {
   }
@@ -9112,6 +9147,7 @@ function registerHealthTools(server2, getIndex, getVaultPath, getConfig = () =>
     embeddings_building: z3.boolean().describe("Whether semantic embeddings are currently building"),
     embeddings_ready: z3.boolean().describe("Whether semantic embeddings have been built (enables hybrid keyword+semantic search)"),
     embeddings_count: z3.coerce.number().describe("Number of notes with semantic embeddings"),
+    embedding_model: z3.string().optional().describe("Active embedding model ID (when embeddings are built)"),
     tasks_ready: z3.boolean().describe("Whether the task cache is ready to serve queries"),
     tasks_building: z3.boolean().describe("Whether the task cache is currently rebuilding"),
     watcher_state: z3.enum(["starting", "ready", "rebuilding", "dirty", "error"]).optional().describe("Current file watcher state"),
@@ -9297,6 +9333,7 @@ function registerHealthTools(server2, getIndex, getVaultPath, getConfig = () =>
         embeddings_building: isEmbeddingsBuilding(),
         embeddings_ready: hasEmbeddingsIndex(),
         embeddings_count: getEmbeddingsCount(),
+        embedding_model: hasEmbeddingsIndex() ? getActiveModelId() : void 0,
         tasks_ready: isTaskCacheReady(),
         tasks_building: isTaskCacheBuilding(),
         watcher_state: getWatcherStatus2()?.state,
@@ -19547,7 +19584,18 @@ async function runPostIndexWork(index) {
     serverLog("config", `Vault: ${flywheelConfig.vault_name}`);
   }
   if (process.env.FLYWHEEL_SKIP_EMBEDDINGS !== "true") {
-    if (hasEmbeddingsIndex()) {
+    let modelChanged = false;
+    if (hasEmbeddingsIndex() && needsEmbeddingRebuild()) {
+      const oldModel = getStoredEmbeddingModel();
+      serverLog("semantic", `Embedding model changed from ${oldModel} to ${getActiveModelId()}, rebuilding`);
+      if (stateDb) {
+        stateDb.db.exec("DELETE FROM note_embeddings");
+        stateDb.db.exec("DELETE FROM entity_embeddings");
+      }
+      setEmbeddingsBuildState("none");
+      modelChanged = true;
+    }
+    if (hasEmbeddingsIndex() && !modelChanged) {
       serverLog("semantic", "Embeddings already built, skipping full scan");
     } else {
       setEmbeddingsBuilding(true);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@velvetmonkey/flywheel-memory",
-  "version": "2.0.60",
+  "version": "2.0.62",
   "description": "MCP server that gives Claude full read/write access to your Obsidian vault. Select from 42 tools for search, backlinks, graph queries, mutations, and hybrid semantic search.",
   "type": "module",
   "main": "dist/index.js",
@@ -52,7 +52,7 @@
   },
   "dependencies": {
     "@modelcontextprotocol/sdk": "^1.25.1",
-    "@velvetmonkey/vault-core": "^2.0.60",
+    "@velvetmonkey/vault-core": "^2.0.62",
     "better-sqlite3": "^11.0.0",
     "chokidar": "^4.0.0",
     "gray-matter": "^4.0.3",