@ez-corp/ez-search 1.0.8 → 1.0.10

package/dist/cli/commands/index-cmd.js

@@ -18,7 +18,7 @@
  * Model routing:
  *   code  -> jinaai/jina-embeddings-v2-base-code, col-768
  *   text  -> nomic-ai/nomic-embed-text-v1.5, col-768  (prefix: "search_document: ")
- *   image -> Xenova/clip-vit-base-patch32, col-512     (one vector per file)
+ *   image -> Xenova/clip-vit-base-patch16, col-512     (one vector per file)
  */
 import * as path from 'path';
 import * as fsp from 'fs/promises';
@@ -30,7 +30,7 @@ const BATCH_SIZE = 32;
  * Used by both code and text pipelines (they differ only in chunker, model, prefix, tokenizer).
  */
 async function runTextEmbeddingPipeline(opts) {
-    const { type, files, col768, manifest, hashContent, hashText, makeChunkId } = opts;
+    const { type, files, col768, manifest, hashContent, hashText, makeChunkId, progress } = opts;
     let filesIndexed = 0;
     let filesSkipped = 0;
     let chunksCreated = 0;
@@ -38,7 +38,9 @@ async function runTextEmbeddingPipeline(opts) {
     let chunksRemoved = 0;
     // Determine which files need processing (mtime+size fast path, hash confirmation)
     const filesToProcess = [];
-    for (const file of files) {
+    for (let fi = 0; fi < files.length; fi++) {
+        const file = files[fi];
+        progress.update(`${type}: checking files`, fi + 1, files.length);
         const existing = manifest.files[file.relativePath];
         if (existing && existing.mtime === file.mtimeMs && existing.size === file.sizeBytes) {
             filesSkipped++;
@@ -190,11 +192,15 @@ async function runTextEmbeddingPipeline(opts) {
     }
     // Embed all pending chunks
     if (allPendingChunks.length > 0) {
+        progress.update(`${type}: loading model...`);
         const { createEmbeddingPipeline } = await import('../../services/model-router.js');
         pipe = await createEmbeddingPipeline(type);
         // Nomic requires "search_document: " prefix on indexed documents
         const prefix = type === 'text' ? 'search_document: ' : '';
+        const totalBatches = Math.ceil(allPendingChunks.length / BATCH_SIZE);
         for (let batchStart = 0; batchStart < allPendingChunks.length; batchStart += BATCH_SIZE) {
+            const batchNum = Math.floor(batchStart / BATCH_SIZE) + 1;
+            progress.update(`${type}: embedding`, batchNum, totalBatches);
             const batch = allPendingChunks.slice(batchStart, batchStart + BATCH_SIZE);
             const texts = batch.map((c) => prefix + c.text);
             const embeddings = await pipe.embed(texts);
@@ -226,6 +232,11 @@ async function runTextEmbeddingPipeline(opts) {
 }
 export async function runIndex(targetPath, options) {
     const startTime = Date.now();
+    const { ProgressReporter } = await import('../progress.js');
+    const progress = new ProgressReporter({
+        quiet: options.quiet,
+        json: options.format !== 'text',
+    });
     try {
         // 1. Resolve path
         const absPath = path.resolve(targetPath);
@@ -267,6 +278,7 @@ export async function runIndex(targetPath, options) {
             const scannedFiles = [];
             for await (const file of scanFiles(absPath, { useIgnoreFiles: options.ignore, typeFilter: fileType })) {
                 scannedFiles.push(file);
+                progress.update(`scanning ${fileType} files... ${scannedFiles.length} found`);
             }
             totalFilesScanned += scannedFiles.length;
             if (scannedFiles.length === 0) {
@@ -305,6 +317,7 @@ export async function runIndex(targetPath, options) {
                     hashContent,
                     hashText,
                     makeChunkId,
+                    progress,
                 });
                 totalFilesIndexed += result.filesIndexed;
                 totalFilesSkipped += result.filesSkipped;
@@ -356,9 +369,12 @@ export async function runIndex(targetPath, options) {
                 }
                 if (filesToProcess.length > 0) {
                     // Load CLIP pipeline once for the batch
+                    progress.update('image: loading model...');
                     const { createImageEmbeddingPipeline } = await import('../../services/image-embedder.js');
                     const imagePipeline = await createImageEmbeddingPipeline();
-                    for (const file of filesToProcess) {
+                    for (let imgIdx = 0; imgIdx < filesToProcess.length; imgIdx++) {
+                        const file = filesToProcess[imgIdx];
+                        progress.update('image: embedding', imgIdx + 1, filesToProcess.length);
                         const buf = await fsp.readFile(file.absolutePath);
                         const fileHash = hashContent(buf);
                         const embedding = await imagePipeline.embedImage(buf);
@@ -393,6 +409,7 @@ export async function runIndex(targetPath, options) {
             emitError({ code: 'EMPTY_DIR', message: 'No supported files found in directory', suggestion: 'Ensure the directory contains supported file types (.ts, .js, .py, .go, .rs, .c, .cpp, .md, .txt, .jpg, .png, .webp)' }, format);
         }
         // 6. Optimize, close collections, THEN save manifest
+        progress.update('optimizing index...');
         col768.optimize();
         col768.close();
         if (imageFilesProcessed) {
@@ -400,6 +417,7 @@ export async function runIndex(targetPath, options) {
         }
         col512.close();
         saveManifest(absPath, manifest);
+        progress.done();
         // 7. Output results
         const durationMs = Date.now() - startTime;
         const hasChanges = totalFilesIndexed > 0 || allDeletedPaths.length > 0;
@@ -447,6 +465,7 @@ export async function runIndex(targetPath, options) {
         return output;
     }
     catch (err) {
+        progress.done();
         const { emitError } = await import('../errors.js');
         const message = err instanceof Error ? err.message : String(err);
         return emitError({ code: 'GENERAL_ERROR', message, suggestion: 'Check the error above and retry' }, options.format === 'text' ? 'text' : 'json');

package/dist/cli/progress.js

@@ -0,0 +1,33 @@
+/**
+ * Live progress reporter for CLI indexing.
+ *
+ * Writes a single updating line to stderr using \r + ANSI clear.
+ * Only active when stderr is a TTY and output isn't suppressed.
+ * Does not interfere with JSON/text output on stdout.
+ */
+const BAR_WIDTH = 20;
+export class ProgressReporter {
+    enabled;
+    constructor(opts) {
+        this.enabled = !opts.quiet && !opts.json && !!process.stderr.isTTY;
+    }
+    /** Overwrite the current line with a status message + optional progress bar. */
+    update(message, current, total) {
+        if (!this.enabled)
+            return;
+        let line = message;
+        if (total != null && current != null && total > 0) {
+            const pct = Math.min(current / total, 1);
+            const filled = Math.round(pct * BAR_WIDTH);
+            const bar = '\u2588'.repeat(filled) + '\u2591'.repeat(BAR_WIDTH - filled);
+            line = `${message} [${bar}] ${current}/${total}`;
+        }
+        process.stderr.write(`\r\x1b[K${line}`);
+    }
+    /** Clear the progress line. Call when indexing is complete. */
+    done() {
+        if (!this.enabled)
+            return;
+        process.stderr.write('\r\x1b[K');
+    }
+}

package/dist/services/image-embedder.js

@@ -14,7 +14,7 @@
 import { CLIPVisionModelWithProjection, CLIPTextModelWithProjection, AutoProcessor, AutoTokenizer, RawImage, env } from '@huggingface/transformers';
 import { resolveModelCachePath } from '../config/paths.js';
 // ── Constants ─────────────────────────────────────────────────────────────────
-const CLIP_MODEL_ID = 'Xenova/clip-vit-base-patch32';
+const CLIP_MODEL_ID = 'Xenova/clip-vit-base-patch16';
 const CLIP_DIM = 512;
 // ── Helpers ──────────────────────────────────────────────────────────────────
 /**
@@ -36,7 +36,7 @@ function l2Normalize(vec) {
 }
 // ── Public API ────────────────────────────────────────────────────────────────
 /**
- * Create an ImageEmbeddingPipeline backed by CLIP ViT-B/32 (fp32).
+ * Create an ImageEmbeddingPipeline backed by CLIP ViT-B/16 (fp32).
  *
  * Loads the AutoProcessor and CLIPVisionModelWithProjection in parallel.
  * Model weights are cached in ~/.ez-search/models/.
@@ -80,7 +80,7 @@ export async function createImageEmbeddingPipeline() {
     };
 }
 /**
- * Create a ClipTextPipeline backed by CLIP ViT-B/32 (fp32).
+ * Create a ClipTextPipeline backed by CLIP ViT-B/16 (fp32).
  *
  * Loads AutoTokenizer and CLIPTextModelWithProjection in parallel.
  * Used for text-to-image search: encode query text into CLIP's 512-dim space,

package/dist/services/manifest-cache.js

@@ -13,7 +13,7 @@ import { readFileSync, writeFileSync, renameSync, existsSync, mkdirSync } from '
 import * as path from 'path';
 import { resolveProjectStoragePath } from '../config/paths.js';
 // ── Constants ─────────────────────────────────────────────────────────────────
-export const MANIFEST_VERSION = 3;
+export const MANIFEST_VERSION = 4;
 export const MANIFEST_FILENAME = 'manifest.json';
 // ── Helpers ───────────────────────────────────────────────────────────────────
 function manifestPath(projectDir) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ez-corp/ez-search",
-  "version": "1.0.8",
+  "version": "1.0.10",
   "description": "Semantic codebase search with zero cloud dependencies",
   "type": "module",
   "main": "dist/cli/index.js",