@voidwire/lore 1.3.0 → 1.4.0

package/lib/semantic.ts

@@ -13,6 +13,44 @@ import { getDatabasePath, openDatabase } from "./db.js";
 import { search as keywordSearch, type SearchResult } from "./search.js";
 import { getConfig } from "./config.js";
 
+// ─── Embedding Server (server-first, in-process fallback) ────────────────────
+
+const EMBED_SERVER = process.env.EMBED_SERVER_URL || "http://localhost:8090";
+
+/**
+ * Try the persistent embedding server first (warm: ~9ms vs 244ms in-process).
+ * Returns null on any failure — caller falls back to in-process.
+ */
+async function serverEmbed(
+  text: string,
+  prefix: string,
+): Promise<number[] | null> {
+  try {
+    const resp = await fetch(`${EMBED_SERVER}/embed`, {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify({ text, prefix }),
+      signal: AbortSignal.timeout(500),
+    });
+    if (!resp.ok) return null;
+    const data = (await resp.json()) as {
+      embedding?: number[];
+      dims?: number;
+    };
+    if (
+      !Array.isArray(data.embedding) ||
+      data.embedding.length !== EMBEDDING_DIM
+    ) {
+      return null;
+    }
+    return data.embedding;
+  } catch {
+    return null; // Server not running or timed out — fall back silently
+  }
+}
+
+// ─── Types ───────────────────────────────────────────────────────────────────
+
 export interface SemanticResult {
   rowid: number;
   source: string;
@@ -76,6 +114,11 @@ async function getEmbeddingPipeline(): Promise<EmbeddingPipeline> {
  * @returns 768-dimensional embedding vector
  */
 export async function embedQuery(query: string): Promise<number[]> {
+  // Try persistent server first (~9ms warm vs 244ms in-process)
+  const serverResult = await serverEmbed(query, "search_query");
+  if (serverResult) return serverResult;
+
+  // Fall back to in-process model loading
   const embedder = await getEmbeddingPipeline();
 
   // nomic model requires "search_query: " prefix for queries
@@ -104,6 +147,11 @@ export async function embedQuery(query: string): Promise<number[]> {
  * @returns 768-dimensional embedding vector
  */
 export async function embedDocument(text: string): Promise<number[]> {
+  // Try persistent server first (~9ms warm vs 244ms in-process)
+  const serverResult = await serverEmbed(text, "search_document");
+  if (serverResult) return serverResult;
+
+  // Fall back to in-process model loading
   const embedder = await getEmbeddingPipeline();
 
   const prefixedText = `search_document: ${text}`;
@@ -131,12 +179,23 @@ export async function embedDocument(text: string): Promise<number[]> {
 export async function embedDocuments(texts: string[]): Promise<number[][]> {
   if (texts.length === 0) return [];
 
-  const embedder = await getEmbeddingPipeline();
   const results: number[][] = [];
 
-  // Process one at a time (transformers.js doesn't batch well)
-  // But we benefit from cached pipeline
+  // Try persistent server first for each document
+  let serverAvailable = true;
   for (const text of texts) {
+    if (serverAvailable) {
+      const serverResult = await serverEmbed(text, "search_document");
+      if (serverResult) {
+        results.push(serverResult);
+        continue;
+      }
+      // Server failed — stop trying and fall back for remaining
+      serverAvailable = false;
+    }
+
+    // Fall back to in-process
+    const embedder = await getEmbeddingPipeline();
     const prefixedText = `search_document: ${text}`;
     const output = await embedder(prefixedText, {
       pooling: "mean",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@voidwire/lore",
-  "version": "1.3.0",
+  "version": "1.4.0",
   "description": "Unified knowledge CLI - Search, list, and capture your indexed knowledge",
   "type": "module",
   "main": "./index.ts",