weflow-cli 1.5.0 → 1.6.0

package/scripts/semantic_search.py

@@ -1,431 +1,435 @@
-#!/usr/bin/env python3
-"""
-全局语义搜索 — 基于阿里云百炼 Embedding API (text-embedding-v4) + NumPy。
-
-用法:
-  # 构建索引（首次或增量）
-  python scripts/semantic_search.py build
-
-  # 搜索（有索引用向量，否则关键词 fallback）
-  python scripts/semantic_search.py search "有人推荐过遥感的工具吗" --top-k 10
-
-  # 增量更新（只处理新数据）
-  python scripts/semantic_search.py update
-
-输出: JSON 格式
-"""
-
-import sys, os, json, hashlib, argparse
-from datetime import datetime, timezone, timedelta
-from pathlib import Path
-
-try:
-    import numpy as np
-except ImportError:
-    print(json.dumps({"error": "请安装: pip install numpy"}))
-    sys.exit(1)
-
-try:
-    from sqlcipher3 import dbapi2 as sqlcipher
-except ImportError:
-    print(json.dumps({"error": "请安装: pip install sqlcipher3"}))
-    sys.exit(1)
-
-try:
-    import urllib.request
-    import urllib.parse
-except:
-    pass
-
-sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
-from _utils import load_config, decrypt_lock
-
-OUTPUT_ROOT = 'output'
-INDEX_DIR = Path(OUTPUT_ROOT) / '.semantic_index'
-VECTORS_FILE = INDEX_DIR / 'vectors.npy'
-META_FILE = INDEX_DIR / 'meta.json'
-EMBEDDING_DIM = 1024  # 阿里云 text-embedding-v4
-BATCH_SIZE = 10  # 阿里云 text-embedding-v4 单次最多 10 条
-TZ = timezone(timedelta(hours=8))
-
-
-def json_output(data):
-    print(json.dumps(data, ensure_ascii=False, indent=2))
-
-
-# ====== Embedding API ======
-
-def get_embeddings(texts: list[str], api_key: str) -> list[list[float]]:
-    """Call 阿里云百炼 embedding API (OpenAI-compatible)."""
-    if not texts:
-        return []
-
-    url = "https://dashscope.aliyuncs.com/compatible-mode/v1/embeddings"
-    headers = {
-        "Authorization": f"Bearer {api_key}",
-        "Content-Type": "application/json",
-    }
-
-    all_embeddings = []
-    for i in range(0, len(texts), BATCH_SIZE):
-        batch = texts[i:i + BATCH_SIZE]
-        data = {
-            "model": "text-embedding-v4",
-            "input": batch,
-        }
-
-        req = urllib.request.Request(
-            url,
-            data=json.dumps(data).encode('utf-8'),
-            headers=headers,
-        )
-
-        try:
-            with urllib.request.urlopen(req, timeout=120) as resp:
-                result = json.loads(resp.read().decode('utf-8'))
-                embeddings = [item['embedding'] for item in result['data']]
-                all_embeddings.extend(embeddings)
-        except Exception as e:
-            err_body = ''
-            if hasattr(e, 'read'):
-                try: err_body = e.read().decode()[:300]
-                except: pass
-            print(f"[WARN] Embedding API batch {i//BATCH_SIZE + 1} 失败: {e} {err_body}", file=sys.stderr)
-            all_embeddings.extend([[0.0] * EMBEDDING_DIM for _ in batch])
-
-    return all_embeddings
-
-
-# ====== Data Collection ======
-
-def open_db(db_path, key_hex, salt_hex):
-    raw_key = f"x'{key_hex}{salt_hex}'"
-    conn = sqlcipher.connect(db_path)
-    c = conn.cursor()
-    c.execute(f'PRAGMA key = "{raw_key}";')
-    c.execute("SELECT count(*) FROM sqlite_master")
-    return conn
-
-
-def get_name_map(contact_db, contact_key, contact_salt):
-    name_map = {}
-    if not contact_db or not contact_key or not os.path.exists(contact_db):
-        return name_map
-    try:
-        conn = open_db(contact_db, contact_key, contact_salt)
-        c = conn.cursor()
-        c.execute("SELECT username, COALESCE(NULLIF(remark,''), NULLIF(nick_name,''), username) FROM contact")
-        for r in c.fetchall():
-            name_map[r[0]] = r[1]
-        conn.close()
-    except:
-        pass
-    return name_map
-
-
-def collect_chat_messages(conn, name_map, days=90):
-    """Collect chat messages for indexing."""
-    c = conn.cursor()
-    now = datetime.now(TZ)
-    start_ts = int((now - timedelta(days=days)).timestamp())
-
-    try:
-        c.execute("SELECT user_name FROM Name2Id WHERE is_session = 1")
-        sessions = [r[0] for r in c.fetchall()]
-    except:
-        c.execute("SELECT name FROM sqlite_master WHERE type='table' AND name LIKE 'Msg_%'")
-        sessions = [r[0] for r in c.fetchall()]
-
-    items = []
-    for talker in sessions[:100]:  # Limit to avoid too many
-        tbl = f"Msg_{hashlib.md5(talker.encode()).hexdigest()}"
-        try:
-            c.execute(f'SELECT COUNT(*) FROM sqlite_master WHERE name="{tbl}"')
-            if c.fetchone()[0] == 0:
-                continue
-
-            c.execute(f'''
-                SELECT create_time, real_sender_id, message_content
-                FROM "{tbl}"
-                WHERE create_time >= ?
-                ORDER BY create_time DESC
-                LIMIT 500
-            ''', (start_ts,))
-
-            for ts, sender, content in c.fetchall():
-                if not content or not isinstance(content, str) or len(content) < 10:
-                    continue
-                sender_name = name_map.get(sender, sender) if sender else name_map.get(talker, talker)
-                dt = datetime.fromtimestamp(ts, tz=TZ)
-                items.append({
-                    "id": f"chat:{talker}:{ts}",
-                    "type": "chat",
-                    "talker": name_map.get(talker, talker),
-                    "sender": sender_name,
-                    "time": dt.strftime('%Y-%m-%d %H:%M'),
-                    "text": content[:500],
-                })
-        except:
-            pass
-
-    return items
-
-
-def collect_articles():
-    """Collect articles from biz-daily."""
-    items = []
-    daily_dir = Path(OUTPUT_ROOT) / 'biz-daily'
-    if not daily_dir.exists():
-        return items
-
-    for date_dir in sorted(daily_dir.iterdir(), reverse=True)[:30]:  # Last 30 days
-        if not date_dir.is_dir():
-            continue
-        for topic_dir in date_dir.iterdir():
-            if not topic_dir.is_dir():
-                continue
-            for f in topic_dir.glob(' marriage*.md'):
-                if f.name == 'README.md':
-                    continue
-                try:
-                    content = f.read_text(encoding='utf-8')
-                    # Extract title and body
-                    lines = content.split('\n')
-                    title = ''
-                    body_start = 0
-                    for i, line in enumerate(lines):
-                        if line.startswith('title:'):
-                            title = line.split(':', 1)[1].strip().strip('"')
-                        if line.startswith('## 正文'):
-                            body_start = i + 1
-                            break
-                    body = '\n'.join(lines[body_start:body_start + 50]) if body_start else content[:1000]
-                    if title:
-                        items.append({
-                            "id": f"article:{f.relative_to(daily_dir)}",
-                            "type": "article",
-                            "title": title,
-                            "date": date_dir.name,
-                            "topic": topic_dir.name,
-                            "text": f"{title}\n{body[:500]}",
-                        })
-                except:
-                    pass
-
-    return items
-
-
-# ====== Index Operations ======
-
-def build_index(api_key: str, full: bool = False):
-    """Build or update the semantic index."""
-    INDEX_DIR.mkdir(parents=True, exist_ok=True)
-
-    # Load existing index
-    existing_ids = set()
-    if not full and META_FILE.exists():
-        meta = json.loads(META_FILE.read_text(encoding='utf-8'))
-        existing_ids = {item['id'] for item in meta}
-        print(f"现有索引: {len(existing_ids)} 条", file=sys.stderr)
-
-    # Collect data
-    print("收集数据...", file=sys.stderr)
-
-    # Load DB
-    config = load_config()
-    nt_db = config.get('ntDbPath', '')
-    if not nt_db:
-        return {"error": "未初始化，请先运行 weflow-cli init"}
-
-    nt_key = decrypt_lock(config.get('ntKey', ''))
-    nt_salt = config.get('ntSalt', '')
-
-    # Get name map
-    msg_dir = os.path.dirname(nt_db.replace('\\', '/'))
-    wxid_dir = os.path.dirname(os.path.dirname(msg_dir))
-    contact_db = os.path.join(wxid_dir, 'db_storage', 'contact', 'contact.db')
-    contact_key_enc = config.get('contactKey', '')
-    contact_salt = config.get('contactSalt', '')
-    contact_key = decrypt_lock(contact_key_enc) if contact_key_enc else ''
-    name_map = get_name_map(contact_db, contact_key, contact_salt)
-
-    # Collect items
-    items = []
-    try:
-        conn = open_db(nt_db, nt_key, nt_salt)
-        chat_items = collect_chat_messages(conn, name_map, days=90)
-        items.extend(chat_items)
-        conn.close()
-    except Exception as e:
-        print(f"[WARN] 聊天消息收集失败: {e}", file=sys.stderr)
-
-    article_items = collect_articles()
-    items.extend(article_items)
-
-    # Filter new items
-    new_items = [item for item in items if item['id'] not in existing_ids]
-    print(f"总数据: {len(items)} 条, 新数据: {len(new_items)} 条", file=sys.stderr)
-
-    if not new_items:
-        return {"status": "up_to_date", "total": len(items)}
-
-    # Generate embeddings
-    print("生成 embeddings...", file=sys.stderr)
-    texts = [item['text'] for item in new_items]
-    embeddings = get_embeddings(texts, api_key)
-
-    if not embeddings or all(e == [0.0] * EMBEDDING_DIM for e in embeddings):
-        return {"error": "Embedding 生成失败，请检查 API key"}
-
-    # Load existing vectors
-    if VECTORS_FILE.exists() and not full:
-        vectors = np.load(VECTORS_FILE)
-        meta = json.loads(META_FILE.read_text(encoding='utf-8'))
-    else:
-        vectors = np.empty((0, EMBEDDING_DIM), dtype=np.float32)
-        meta = []
-
-    # Append new vectors
-    new_vectors = np.array(embeddings, dtype=np.float32)
-    vectors = np.vstack([vectors, new_vectors]) if vectors.size else new_vectors
-    meta.extend(new_items)
-
-    # Normalize vectors for cosine similarity
-    norms = np.linalg.norm(vectors, axis=1, keepdims=True)
-    norms[norms == 0] = 1  # Avoid division by zero
-    vectors = vectors / norms
-
-    # Save
-    np.save(VECTORS_FILE, vectors)
-    META_FILE.write_text(json.dumps(meta, ensure_ascii=False), encoding='utf-8')
-
-    return {
-        "status": "success",
-        "total": len(meta),
-        "new": len(new_items),
-    }
-
-
-def keyword_search(query: str, top_k: int = 10):
-    """Simple keyword fallback: scan articles + messages for keyword matches."""
-    results = []
-    keywords = query.lower().split()
-
-    # Scan biz-daily articles
-    biz_dir = Path('output/biz-daily')
-    if biz_dir.exists():
-        for md_file in sorted(biz_dir.rglob('*.md'), reverse=True):
-            if md_file.name == 'README.md' or md_file.name.startswith('.'):
-                continue
-            try:
-                content = md_file.read_text(encoding='utf-8')[:5000]
-            except:
-                continue
-            score = sum(content.lower().count(kw) for kw in keywords)
-            if score > 0:
-                title = content.split('\n')[0].lstrip('# ').strip() if content.startswith('#') else md_file.stem
-                results.append({
-                    'title': title,
-                    'source': str(md_file.relative_to(biz_dir)),
-                    'score': score,
-                    'text': content[:300].strip(),
-                })
-
-    # Scan chat messages from mcp_bridge
-    try:
-        sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
-        from mcp_bridge import search_messages as _search_msg
-        msg_results = _search_msg(query, 50)
-        for r in msg_results.get('results', [])[:top_k]:
-            results.append({
-                'title': f"[{r.get('time','')}] {r.get('talker','')} > {r.get('sender','')}",
-                'source': '微信聊天',
-                'score': len(query),
-                'text': r.get('content', '')[:200],
-            })
-    except:
-        pass
-
-    results.sort(key=lambda x: -x['score'])
-    return results[:top_k]
-
-
-def search(query: str, api_key: str, top_k: int = 10):
-    """Semantic search with keyword fallback."""
-    # Try embedding-based search first
-    if VECTORS_FILE.exists() and META_FILE.exists():
-        try:
-            vectors = np.load(VECTORS_FILE)
-            meta = json.loads(META_FILE.read_text(encoding='utf-8'))
-            query_embeddings = get_embeddings([query], api_key)
-            if query_embeddings and not all(v == 0 for v in query_embeddings[0]):
-                query_vec = np.array(query_embeddings[0], dtype=np.float32)
-                query_vec = query_vec / np.linalg.norm(query_vec)
-                similarities = np.dot(vectors, query_vec)
-                top_indices = np.argsort(similarities)[::-1][:top_k]
-                results = []
-                for idx in top_indices:
-                    item = meta[idx]
-                    results.append({**item, 'score': float(similarities[idx])})
-                return results
-        except:
-            pass
-
-    # Fallback: keyword search
-    return keyword_search(query, top_k)
-
-    results = []
-    for idx in top_indices:
-        item = meta[idx]
-        results.append({
-            **item,
-            "score": float(similarities[idx]),
-        })
-
-    return {
-        "query": query,
-        "total": len(meta),
-        "results": results,
-    }
-
-
-# ====== Main ======
-
-def main():
-    sys.stdout.reconfigure(encoding='utf-8', errors='replace')
-    parser = argparse.ArgumentParser()
-    subparsers = parser.add_subparsers(dest='command')
-
-    # build
-    p = subparsers.add_parser('build')
-    p.add_argument('--api-key', help='DeepSeek API key（优先从 config 读取）')
-    p.add_argument('--full', action='store_true', help='全量重建')
-
-    # update
-    p = subparsers.add_parser('update')
-    p.add_argument('--api-key', help='DeepSeek API key（优先从 config 读取）')
-
-    # search
-    p = subparsers.add_parser('search')
-    p.add_argument('query')
-    p.add_argument('--api-key', help='DeepSeek API key（向量搜索时需要，关键词 fallback 不需要）')
-    p.add_argument('--top-k', type=int, default=10)
-
-    args = parser.parse_args()
-    config = load_config()
-    api_key = args.api_key or os.environ.get('DASHSCOPE_API_KEY', '') or config.get('dashscopeApiKey', '')
-
-    if args.command == 'build':
-        result = build_index(api_key, full=args.full)
-    elif args.command == 'update':
-        result = build_index(api_key, full=False)
-    elif args.command == 'search':
-        result = search(args.query, api_key, top_k=args.top_k)
-    else:
-        result = {"error": f"未知命令: {args.command}"}
-
-    json_output(result)
-
-
-if __name__ == '__main__':
-    main()
+#!/usr/bin/env python3
+"""
+全局语义搜索 — 基于阿里云百炼 Embedding API (text-embedding-v4) + NumPy。
+
+用法:
+  # 构建索引（首次或增量）
+  python scripts/semantic_search.py build
+
+  # 搜索（有索引用向量，否则关键词 fallback）
+  python scripts/semantic_search.py search "有人推荐过遥感的工具吗" --top-k 10
+
+  # 增量更新（只处理新数据）
+  python scripts/semantic_search.py update
+
+输出: JSON 格式
+"""
+
+import sys, os, json, hashlib, argparse
+from datetime import datetime, timezone, timedelta
+from pathlib import Path
+
+try:
+    import numpy as np
+except ImportError:
+    print(json.dumps({"error": "请安装: pip install numpy"}))
+    sys.exit(1)
+
+try:
+    from sqlcipher3 import dbapi2 as sqlcipher
+except ImportError:
+    print(json.dumps({"error": "请安装: pip install sqlcipher3"}))
+    sys.exit(1)
+
+try:
+    import urllib.request
+    import urllib.parse
+except:
+    pass
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+from _utils import load_config, decrypt_lock
+
+OUTPUT_ROOT = 'output'
+INDEX_DIR = Path(OUTPUT_ROOT) / '.semantic_index'
+VECTORS_FILE = INDEX_DIR / 'vectors.npy'
+META_FILE = INDEX_DIR / 'meta.json'
+EMBEDDING_DIM = 1024  # 阿里云 text-embedding-v4
+BATCH_SIZE = 10  # 阿里云 text-embedding-v4 单次最多 10 条
+TZ = timezone(timedelta(hours=8))
+
+
+def json_output(data):
+    print(json.dumps(data, ensure_ascii=False, indent=2))
+
+
+# ====== Embedding API ======
+
+def get_embeddings(texts: list[str], api_key: str) -> list[list[float]]:
+    """Call 阿里云百炼 embedding API (OpenAI-compatible)."""
+    if not texts:
+        return []
+
+    url = "https://dashscope.aliyuncs.com/compatible-mode/v1/embeddings"
+    headers = {
+        "Authorization": f"Bearer {api_key}",
+        "Content-Type": "application/json",
+    }
+
+    all_embeddings = []
+    for i in range(0, len(texts), BATCH_SIZE):
+        batch = texts[i:i + BATCH_SIZE]
+        data = {
+            "model": "text-embedding-v4",
+            "input": batch,
+        }
+
+        req = urllib.request.Request(
+            url,
+            data=json.dumps(data).encode('utf-8'),
+            headers=headers,
+        )
+
+        try:
+            with urllib.request.urlopen(req, timeout=120) as resp:
+                result = json.loads(resp.read().decode('utf-8'))
+                embeddings = [item['embedding'] for item in result['data']]
+                all_embeddings.extend(embeddings)
+        except Exception as e:
+            err_body = ''
+            if hasattr(e, 'read'):
+                try: err_body = e.read().decode()[:300]
+                except: pass
+            print(f"[WARN] Embedding API batch {i//BATCH_SIZE + 1} 失败: {e} {err_body}", file=sys.stderr)
+            all_embeddings.extend([[0.0] * EMBEDDING_DIM for _ in batch])
+
+    return all_embeddings
+
+
+# ====== Data Collection ======
+
+def open_db(db_path, key_hex, salt_hex):
+    raw_key = f"x'{key_hex}{salt_hex}'"
+    conn = sqlcipher.connect(db_path)
+    c = conn.cursor()
+    c.execute(f'PRAGMA key = "{raw_key}";')
+    c.execute("SELECT count(*) FROM sqlite_master")
+    return conn
+
+
+def get_name_map(contact_db, contact_key, contact_salt):
+    name_map = {}
+    if not contact_db or not contact_key or not os.path.exists(contact_db):
+        return name_map
+    try:
+        conn = open_db(contact_db, contact_key, contact_salt)
+        c = conn.cursor()
+        c.execute("SELECT username, COALESCE(NULLIF(remark,''), NULLIF(nick_name,''), username) FROM contact")
+        for r in c.fetchall():
+            name_map[r[0]] = r[1]
+        conn.close()
+    except:
+        pass
+    return name_map
+
+
+def collect_chat_messages(conn, name_map, days=90):
+    """Collect chat messages for indexing."""
+    c = conn.cursor()
+    now = datetime.now(TZ)
+    start_ts = int((now - timedelta(days=days)).timestamp())
+
+    try:
+        c.execute("SELECT user_name FROM Name2Id WHERE is_session = 1")
+        sessions = [r[0] for r in c.fetchall()]
+    except:
+        c.execute("SELECT name FROM sqlite_master WHERE type='table' AND name LIKE 'Msg_%'")
+        sessions = [r[0] for r in c.fetchall()]
+
+    items = []
+    for talker in sessions[:100]:  # Limit to avoid too many
+        tbl = f"Msg_{hashlib.md5(talker.encode()).hexdigest()}"
+        try:
+            c.execute(f'SELECT COUNT(*) FROM sqlite_master WHERE name="{tbl}"')
+            if c.fetchone()[0] == 0:
+                continue
+
+            c.execute(f'''
+                SELECT create_time, real_sender_id, message_content
+                FROM "{tbl}"
+                WHERE create_time >= ?
+                ORDER BY create_time DESC
+                LIMIT 500
+            ''', (start_ts,))
+
+            for ts, sender, content in c.fetchall():
+                if not content or not isinstance(content, str) or len(content) < 10:
+                    continue
+                sender_name = name_map.get(sender, sender) if sender else name_map.get(talker, talker)
+                dt = datetime.fromtimestamp(ts, tz=TZ)
+                items.append({
+                    "id": f"chat:{talker}:{ts}",
+                    "type": "chat",
+                    "talker": name_map.get(talker, talker),
+                    "sender": sender_name,
+                    "time": dt.strftime('%Y-%m-%d %H:%M'),
+                    "text": content[:500],
+                })
+        except:
+            pass
+
+    return items
+
+
+def collect_articles():
+    """Collect articles from biz-daily."""
+    items = []
+    daily_dir = Path(OUTPUT_ROOT) / 'biz-daily'
+    if not daily_dir.exists():
+        return items
+
+    for date_dir in sorted(daily_dir.iterdir(), reverse=True)[:30]:  # Last 30 days
+        if not date_dir.is_dir():
+            continue
+        for topic_dir in date_dir.iterdir():
+            if not topic_dir.is_dir():
+                continue
+            for f in topic_dir.glob(' marriage*.md'):
+                if f.name == 'README.md':
+                    continue
+                try:
+                    content = f.read_text(encoding='utf-8')
+                    # Extract title and body
+                    lines = content.split('\n')
+                    title = ''
+                    body_start = 0
+                    for i, line in enumerate(lines):
+                        if line.startswith('title:'):
+                            title = line.split(':', 1)[1].strip().strip('"')
+                        if line.startswith('## 正文'):
+                            body_start = i + 1
+                            break
+                    body = '\n'.join(lines[body_start:body_start + 50]) if body_start else content[:1000]
+                    if title:
+                        items.append({
+                            "id": f"article:{f.relative_to(daily_dir)}",
+                            "type": "article",
+                            "title": title,
+                            "date": date_dir.name,
+                            "topic": topic_dir.name,
+                            "text": f"{title}\n{body[:500]}",
+                        })
+                except:
+                    pass
+
+    return items
+
+
+# ====== Index Operations ======
+
+def build_index(api_key: str, full: bool = False):
+    """Build or update the semantic index."""
+    INDEX_DIR.mkdir(parents=True, exist_ok=True)
+
+    # Load existing index
+    existing_ids = set()
+    if not full and META_FILE.exists():
+        meta = json.loads(META_FILE.read_text(encoding='utf-8'))
+        existing_ids = {item['id'] for item in meta}
+        print(f"现有索引: {len(existing_ids)} 条", file=sys.stderr)
+
+    # Collect data
+    print("收集数据...", file=sys.stderr)
+
+    # Load DB
+    config = load_config()
+    nt_db = config.get('ntDbPath', '')
+    if not nt_db:
+        return {"error": "未初始化，请先运行 weflow-cli init"}
+
+    nt_key = decrypt_lock(config.get('ntKey', ''))
+    nt_salt = config.get('ntSalt', '')
+
+    # Get name map
+    msg_dir = os.path.dirname(nt_db.replace('\\', '/'))
+    wxid_dir = os.path.dirname(os.path.dirname(msg_dir))
+    contact_db = os.path.join(wxid_dir, 'db_storage', 'contact', 'contact.db')
+    contact_key_enc = config.get('contactKey', '')
+    contact_salt = config.get('contactSalt', '')
+    contact_key = decrypt_lock(contact_key_enc) if contact_key_enc else ''
+    name_map = get_name_map(contact_db, contact_key, contact_salt)
+
+    # Collect items
+    items = []
+    try:
+        conn = open_db(nt_db, nt_key, nt_salt)
+        chat_items = collect_chat_messages(conn, name_map, days=90)
+        items.extend(chat_items)
+        conn.close()
+    except Exception as e:
+        print(f"[WARN] 聊天消息收集失败: {e}", file=sys.stderr)
+
+    article_items = collect_articles()
+    items.extend(article_items)
+
+    # Filter new items
+    new_items = [item for item in items if item['id'] not in existing_ids]
+    print(f"总数据: {len(items)} 条, 新数据: {len(new_items)} 条", file=sys.stderr)
+
+    if not new_items:
+        return {"status": "up_to_date", "total": len(items)}
+
+    # Generate embeddings
+    print("生成 embeddings...", file=sys.stderr)
+    texts = [item['text'] for item in new_items]
+    embeddings = get_embeddings(texts, api_key)
+
+    if not embeddings or all(e == [0.0] * EMBEDDING_DIM for e in embeddings):
+        return {"error": "Embedding 生成失败，请检查 API key"}
+
+    # Load existing vectors
+    if VECTORS_FILE.exists() and not full:
+        vectors = np.load(VECTORS_FILE)
+        meta = json.loads(META_FILE.read_text(encoding='utf-8'))
+    else:
+        vectors = np.empty((0, EMBEDDING_DIM), dtype=np.float32)
+        meta = []
+
+    # Append new vectors
+    new_vectors = np.array(embeddings, dtype=np.float32)
+    vectors = np.vstack([vectors, new_vectors]) if vectors.size else new_vectors
+    meta.extend(new_items)
+
+    # Normalize vectors for cosine similarity
+    norms = np.linalg.norm(vectors, axis=1, keepdims=True)
+    norms[norms == 0] = 1  # Avoid division by zero
+    vectors = vectors / norms
+
+    # Save
+    np.save(VECTORS_FILE, vectors)
+    META_FILE.write_text(json.dumps(meta, ensure_ascii=False), encoding='utf-8')
+
+    return {
+        "status": "success",
+        "total": len(meta),
+        "new": len(new_items),
+    }
+
+
+def keyword_search(query: str, top_k: int = 10):
+    """Simple keyword fallback: scan articles + messages for keyword matches."""
+    results = []
+    keywords = query.lower().split()
+
+    # Scan biz-daily articles
+    biz_dir = Path('output/biz-daily')
+    if biz_dir.exists():
+        for md_file in sorted(biz_dir.rglob('*.md'), reverse=True):
+            if md_file.name == 'README.md' or md_file.name.startswith('.'):
+                continue
+            try:
+                content = md_file.read_text(encoding='utf-8')[:5000]
+            except:
+                continue
+            score = sum(content.lower().count(kw) for kw in keywords)
+            if score > 0:
+                title = content.split('\n')[0].lstrip('# ').strip() if content.startswith('#') else md_file.stem
+                results.append({
+                    'title': title,
+                    'source': str(md_file.relative_to(biz_dir)),
+                    'score': score,
+                    'text': content[:300].strip(),
+                })
+
+    # Scan chat messages from mcp_bridge
+    try:
+        sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+        from mcp_bridge import search_messages as _search_msg
+        msg_results = _search_msg(query, 50)
+        for r in msg_results.get('results', [])[:top_k]:
+            results.append({
+                'title': f"[{r.get('time','')}] {r.get('talker','')} > {r.get('sender','')}",
+                'source': '微信聊天',
+                'score': len(query),
+                'text': r.get('content', '')[:200],
+            })
+    except:
+        pass
+
+    results.sort(key=lambda x: -x['score'])
+    return results[:top_k]
+
+
+def search(query: str, api_key: str, top_k: int = 10):
+    """Semantic search with keyword fallback."""
+    # Try embedding-based search first
+    if VECTORS_FILE.exists() and META_FILE.exists():
+        try:
+            vectors = np.load(VECTORS_FILE)
+            meta = json.loads(META_FILE.read_text(encoding='utf-8'))
+            query_embeddings = get_embeddings([query], api_key)
+            if query_embeddings and not all(v == 0 for v in query_embeddings[0]):
+                query_vec = np.array(query_embeddings[0], dtype=np.float32)
+                query_vec = query_vec / np.linalg.norm(query_vec)
+                similarities = np.dot(vectors, query_vec)
+                top_indices = np.argsort(similarities)[::-1][:top_k]
+                results = []
+                for idx in top_indices:
+                    item = meta[idx]
+                    results.append({**item, 'score': float(similarities[idx])})
+                return results
+        except:
+            pass
+
+    # Fallback: keyword search
+    return keyword_search(query, top_k)
+
+    results = []
+    for idx in top_indices:
+        item = meta[idx]
+        results.append({
+            **item,
+            "score": float(similarities[idx]),
+        })
+
+    return {
+        "query": query,
+        "total": len(meta),
+        "results": results,
+    }
+
+
+# ====== Main ======
+
+def main():
+    sys.stdout.reconfigure(encoding='utf-8', errors='replace')
+    parser = argparse.ArgumentParser()
+    subparsers = parser.add_subparsers(dest='command')
+
+    # build
+    p = subparsers.add_parser('build')
+    p.add_argument('--api-key', help='DeepSeek API key（优先从 config 读取）')
+    p.add_argument('--full', action='store_true', help='全量重建')
+
+    # update
+    p = subparsers.add_parser('update')
+    p.add_argument('--api-key', help='DeepSeek API key（优先从 config 读取）')
+
+    # search
+    p = subparsers.add_parser('search')
+    p.add_argument('query', nargs='?')
+    p.add_argument('--api-key', help='DeepSeek API key（向量搜索时需要，关键词 fallback 不需要）')
+    p.add_argument('--top-k', type=int, default=10)
+
+    args = parser.parse_args()
+    config = load_config()
+    api_key = args.api_key or os.environ.get('DASHSCOPE_API_KEY', '') or config.get('dashscopeApiKey', '')
+
+    if args.command == 'build':
+        result = build_index(api_key, full=args.full)
+    elif args.command == 'update':
+        result = build_index(api_key, full=False)
+    elif args.command == 'search':
+        query = args.query or os.environ.get('WEFLOW_SEARCH_QUERY', '')
+        if not query:
+            result = {"error": "missing search query"}
+        else:
+            result = search(query, api_key, top_k=args.top_k)
+    else:
+        result = {"error": f"未知命令: {args.command}"}
+
+    json_output(result)
+
+
+if __name__ == '__main__':
+    main()