fix: Update DiskANN submodule with LAPACKE library linking fix

- Points to commit 19f9603 with explicit LAPACKE library discovery and linking
- Resolves 'undefined symbol: LAPACKE_sgesdd' runtime error on ARM64 Linux
- Completes ARM64 Linux wheel build compatibility for Docker deployments

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

.github/workflows/build-reusable.yml

@@ -54,6 +54,17 @@ jobs:
             python: '3.12'
           - os: ubuntu-22.04
             python: '3.13'
+          # ARM64 Linux builds
+          - os: ubuntu-24.04-arm
+            python: '3.9'
+          - os: ubuntu-24.04-arm
+            python: '3.10'
+          - os: ubuntu-24.04-arm
+            python: '3.11'
+          - os: ubuntu-24.04-arm
+            python: '3.12'
+          - os: ubuntu-24.04-arm
+            python: '3.13'
           - os: macos-14
             python: '3.9'
           - os: macos-14
@@ -108,13 +119,46 @@ jobs:
             pkg-config libabsl-dev libaio-dev libprotobuf-dev \
             patchelf
 
-          # Install Intel MKL for DiskANN
-          wget -q https://registrationcenter-download.intel.com/akdlm/IRC_NAS/79153e0f-74d7-45af-b8c2-258941adf58a/intel-onemkl-2025.0.0.940.sh
-          sudo sh intel-onemkl-2025.0.0.940.sh -a --components intel.oneapi.lin.mkl.devel --action install --eula accept -s
-          source /opt/intel/oneapi/setvars.sh
-          echo "MKLROOT=/opt/intel/oneapi/mkl/latest" >> $GITHUB_ENV
-          echo "LD_LIBRARY_PATH=/opt/intel/oneapi/compiler/latest/linux/compiler/lib/intel64_lin" >> $GITHUB_ENV
-          echo "LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/intel/oneapi/mkl/latest/lib/intel64" >> $GITHUB_ENV
+          # Debug: Show system information
+          echo "🔍 System Information:"
+          echo "Architecture: $(uname -m)"
+          echo "OS: $(uname -a)"
+          echo "CPU info: $(lscpu | head -5)"
+
+          # Install math library based on architecture
+          ARCH=$(uname -m)
+          echo "🔍 Setting up math library for architecture: $ARCH"
+
+          if [[ "$ARCH" == "x86_64" ]]; then
+            # Install Intel MKL for DiskANN on x86_64
+            echo "📦 Installing Intel MKL for x86_64..."
+            wget -q https://registrationcenter-download.intel.com/akdlm/IRC_NAS/79153e0f-74d7-45af-b8c2-258941adf58a/intel-onemkl-2025.0.0.940.sh
+            sudo sh intel-onemkl-2025.0.0.940.sh -a --components intel.oneapi.lin.mkl.devel --action install --eula accept -s
+            source /opt/intel/oneapi/setvars.sh
+            echo "MKLROOT=/opt/intel/oneapi/mkl/latest" >> $GITHUB_ENV
+            echo "LD_LIBRARY_PATH=/opt/intel/oneapi/compiler/latest/linux/compiler/lib/intel64_lin" >> $GITHUB_ENV
+            echo "LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/intel/oneapi/mkl/latest/lib/intel64" >> $GITHUB_ENV
+            echo "✅ Intel MKL installed for x86_64"
+
+            # Debug: Check MKL installation
+            echo "🔍 MKL Installation Check:"
+            ls -la /opt/intel/oneapi/mkl/latest/ || echo "MKL directory not found"
+            ls -la /opt/intel/oneapi/mkl/latest/lib/ || echo "MKL lib directory not found"
+
+          elif [[ "$ARCH" == "aarch64" ]]; then
+            # Use OpenBLAS for ARM64 (MKL installer not compatible with ARM64)
+            echo "📦 Installing OpenBLAS for ARM64..."
+            sudo apt-get install -y libopenblas-dev liblapack-dev liblapacke-dev
+            echo "✅ OpenBLAS installed for ARM64"
+
+            # Debug: Check OpenBLAS installation
+            echo "🔍 OpenBLAS Installation Check:"
+            dpkg -l | grep openblas || echo "OpenBLAS package not found"
+            ls -la /usr/lib/aarch64-linux-gnu/openblas/ || echo "OpenBLAS directory not found"
+          fi
+
+          # Debug: Show final library paths
+          echo "🔍 Final LD_LIBRARY_PATH: $LD_LIBRARY_PATH"
 
       - name: Install system dependencies (macOS)
         if: runner.os == 'macOS'

.gitignore

@@ -93,3 +93,10 @@ packages/leann-backend-diskann/third_party/DiskANN/_deps/
 batchtest.py
 tests/__pytest_cache__/
 tests/__pycache__/
+paru-bin/
+
+CLAUDE.md
+CLAUDE.local.md
+.claude/*.local.*
+.claude/local/*
+benchmarks/data/

.pre-commit-config.yaml

@@ -13,4 +13,5 @@ repos:
     rev: v0.12.7  # Fixed version to match pyproject.toml
     hooks:
       - id: ruff
+        args: [--fix, --exit-non-zero-on-fix]
       - id: ruff-format

README.md

@@ -8,6 +8,8 @@
   <img src="https://img.shields.io/badge/Platform-Ubuntu%20%26%20Arch%20%26%20WSL%20%7C%20macOS%20(ARM64%2FIntel)-lightgrey" alt="Platform">
   <img src="https://img.shields.io/badge/License-MIT-green.svg" alt="MIT License">
   <img src="https://img.shields.io/badge/MCP-Native%20Integration-blue" alt="MCP Integration">
+  <a href="https://join.slack.com/t/leann-e2u9779/shared_invite/zt-3ckd2f6w1-OX08~NN4gkWhh10PRVBj1Q"><img src="https://img.shields.io/badge/Slack-Join-4A154B?logo=slack&logoColor=white" alt="Join Slack">
+  <a href="assets/wechat_user_group.JPG" title="Join WeChat group"><img src="https://img.shields.io/badge/WeChat-Join-2DC100?logo=wechat&logoColor=white" alt="Join WeChat group"></a>
 </p>
 
 <h2 align="center" tabindex="-1" class="heading-element" dir="auto">
@@ -176,6 +178,8 @@ response = chat.ask("How much storage does LEANN save?", top_k=1)
 
 LEANN supports RAG on various data sources including documents (`.pdf`, `.txt`, `.md`), Apple Mail, Google Search History, WeChat, and more.
 
+
+
 ### Generation Model Setup
 
 LEANN supports multiple LLM providers for text generation (OpenAI API, HuggingFace, Ollama).
@@ -218,7 +222,8 @@ ollama pull llama3.2:1b
 
 </details>
 
-### ⭐ Flexible Configuration
+
+## ⭐ Flexible Configuration
 
 LEANN provides flexible parameters for embedding models, search strategies, and data processing to fit your specific needs.
 
@@ -294,6 +299,12 @@ python -m apps.document_rag --data-dir "~/Documents/Papers" --chunk-size 1024
 
 # Filter only markdown and Python files with smaller chunks
 python -m apps.document_rag --data-dir "./docs" --chunk-size 256 --file-types .md .py
+
+# Enable AST-aware chunking for code files
+python -m apps.document_rag --enable-code-chunking --data-dir "./my_project"
+
+# Or use the specialized code RAG for better code understanding
+python -m apps.code_rag --repo-dir "./my_codebase" --query "How does authentication work?"
 ```
 
 </details>
@@ -468,10 +479,20 @@ Once the index is built, you can ask questions like:
 
 ### 🚀 Claude Code Integration: Transform Your Development Workflow!
 
+<details>
+<summary><strong>NEW!! AST‑Aware Code Chunking</strong></summary>
+
+LEANN features intelligent code chunking that preserves semantic boundaries (functions, classes, methods) for Python, Java, C#, and TypeScript, improving code understanding compared to text-based chunking.
+
+📖 Read the [AST Chunking Guide →](docs/ast_chunking_guide.md)
+
+</details>
+
 **The future of code assistance is here.** Transform your development workflow with LEANN's native MCP integration for Claude Code. Index your entire codebase and get intelligent code assistance directly in your IDE.
 
 **Key features:**
 - 🔍 **Semantic code search** across your entire project, fully local index and lightweight
+- 🧠 **AST-aware chunking** preserves code structure (functions, classes)
 - 📚 **Context-aware assistance** for debugging and development
 - 🚀 **Zero-config setup** with automatic language detection
 
@@ -534,7 +555,8 @@ leann remove my-docs
 
 **Key CLI features:**
 - Auto-detects document formats (PDF, TXT, MD, DOCX, PPTX + code files)
-- Smart text chunking with overlap
+- **🧠 AST-aware chunking** for Python, Java, C#, TypeScript files
+- Smart text chunking with overlap for all other content
 - Multiple LLM providers (Ollama, OpenAI, HuggingFace)
 - Organized index storage in `.leann/indexes/` (project-local)
 - Support for advanced search parameters
@@ -607,6 +629,33 @@ Options:
 
 </details>
 
+## 🚀 Advanced Features
+
+### 🎯 Metadata Filtering
+
+LEANN supports a simple metadata filtering system to enable sophisticated use cases like document filtering by date/type, code search by file extension, and content management based on custom criteria.
+
+```python
+# Add metadata during indexing
+builder.add_text(
+    "def authenticate_user(token): ...",
+    metadata={"file_extension": ".py", "lines_of_code": 25}
+)
+
+# Search with filters
+results = searcher.search(
+    query="authentication function",
+    metadata_filters={
+        "file_extension": {"==": ".py"},
+        "lines_of_code": {"<": 100}
+    }
+)
+```
+
+**Supported operators**: `==`, `!=`, `<`, `<=`, `>`, `>=`, `in`, `not_in`, `contains`, `starts_with`, `ends_with`, `is_true`, `is_false`
+
+📖 **[Complete Metadata filtering guide →](docs/metadata_filtering.md)**
+
 ## 🏗️ Architecture & How It Works
 
 <p align="center">
@@ -646,6 +695,7 @@ Options:
 ```bash
 uv pip install -e ".[dev]"  # Install dev dependencies
 python benchmarks/run_evaluation.py    # Will auto-download evaluation data and run benchmarks
+python benchmarks/run_evaluation.py benchmarks/data/indices/rpj_wiki/rpj_wiki --num-queries 2000    # After downloading data, you can run the benchmark with our biggest index
 ```
 
 The evaluation script downloads data automatically on first run. The last three results were tested with partial personal data, and you can reproduce them with your own data!
@@ -685,6 +735,9 @@ MIT License - see [LICENSE](LICENSE) for details.
 
 Core Contributors: [Yichuan Wang](https://yichuan-w.github.io/) & [Zhifei Li](https://github.com/andylizf).
 
+Active Contributors: [Gabriel Dehan](https://github.com/gabriel-dehan)
+
+
 We welcome more contributors! Feel free to open issues or submit PRs.
 
 This work is done at [**Berkeley Sky Computing Lab**](https://sky.cs.berkeley.edu/).

apps/base_rag_example.py

@@ -11,7 +11,6 @@ from typing import Any
 import dotenv
 from leann.api import LeannBuilder, LeannChat
 from leann.registry import register_project_directory
-from llama_index.core.node_parser import SentenceSplitter
 
 dotenv.load_dotenv()
 
@@ -109,6 +108,38 @@ class BaseRAGExample(ABC):
             help="Thinking budget for reasoning models (low/medium/high). Supported by GPT-Oss:20b and other reasoning models.",
         )
 
+        # AST Chunking parameters
+        ast_group = parser.add_argument_group("AST Chunking Parameters")
+        ast_group.add_argument(
+            "--use-ast-chunking",
+            action="store_true",
+            help="Enable AST-aware chunking for code files (requires astchunk)",
+        )
+        ast_group.add_argument(
+            "--ast-chunk-size",
+            type=int,
+            default=512,
+            help="Maximum characters per AST chunk (default: 512)",
+        )
+        ast_group.add_argument(
+            "--ast-chunk-overlap",
+            type=int,
+            default=64,
+            help="Overlap between AST chunks (default: 64)",
+        )
+        ast_group.add_argument(
+            "--code-file-extensions",
+            nargs="+",
+            default=None,
+            help="Additional code file extensions to process with AST chunking (e.g., .py .java .cs .ts)",
+        )
+        ast_group.add_argument(
+            "--ast-fallback-traditional",
+            action="store_true",
+            default=True,
+            help="Fall back to traditional chunking if AST chunking fails (default: True)",
+        )
+
         # Search parameters
         search_group = parser.add_argument_group("Search Parameters")
         search_group.add_argument(
@@ -268,7 +299,6 @@ class BaseRAGExample(ABC):
         chat = LeannChat(
             index_path,
             llm_config=self.get_llm_config(args),
-            system_prompt=f"You are a helpful assistant that answers questions about {self.name} data.",
             complexity=args.search_complexity,
         )
 
@@ -310,21 +340,3 @@ class BaseRAGExample(ABC):
             await self.run_single_query(args, index_path, args.query)
         else:
             await self.run_interactive_chat(args, index_path)
-
-
-def create_text_chunks(documents, chunk_size=256, chunk_overlap=25) -> list[str]:
-    """Helper function to create text chunks from documents."""
-    node_parser = SentenceSplitter(
-        chunk_size=chunk_size,
-        chunk_overlap=chunk_overlap,
-        separator=" ",
-        paragraph_separator="\n\n",
-    )
-
-    all_texts = []
-    for doc in documents:
-        nodes = node_parser.get_nodes_from_documents([doc])
-        if nodes:
-            all_texts.extend(node.get_content() for node in nodes)
-
-    return all_texts

apps/browser_rag.py

@@ -10,7 +10,8 @@ from pathlib import Path
 # Add parent directory to path for imports
 sys.path.insert(0, str(Path(__file__).parent))
 
-from base_rag_example import BaseRAGExample, create_text_chunks
+from base_rag_example import BaseRAGExample
+from chunking import create_text_chunks
 
 from .history_data.history import ChromeHistoryReader
 

apps/chunking/__init__.py

@@ -0,0 +1,22 @@
+"""
+Chunking utilities for LEANN RAG applications.
+Provides AST-aware and traditional text chunking functionality.
+"""
+
+from .utils import (
+    CODE_EXTENSIONS,
+    create_ast_chunks,
+    create_text_chunks,
+    create_traditional_chunks,
+    detect_code_files,
+    get_language_from_extension,
+)
+
+__all__ = [
+    "CODE_EXTENSIONS",
+    "create_ast_chunks",
+    "create_text_chunks",
+    "create_traditional_chunks",
+    "detect_code_files",
+    "get_language_from_extension",
+]

apps/chunking/utils.py

@@ -0,0 +1,320 @@
+"""
+Enhanced chunking utilities with AST-aware code chunking support.
+Provides unified interface for both traditional and AST-based text chunking.
+"""
+
+import logging
+from pathlib import Path
+from typing import Optional
+
+from llama_index.core.node_parser import SentenceSplitter
+
+logger = logging.getLogger(__name__)
+
+# Code file extensions supported by astchunk
+CODE_EXTENSIONS = {
+    ".py": "python",
+    ".java": "java",
+    ".cs": "csharp",
+    ".ts": "typescript",
+    ".tsx": "typescript",
+    ".js": "typescript",
+    ".jsx": "typescript",
+}
+
+# Default chunk parameters for different content types
+DEFAULT_CHUNK_PARAMS = {
+    "code": {
+        "max_chunk_size": 512,
+        "chunk_overlap": 64,
+    },
+    "text": {
+        "chunk_size": 256,
+        "chunk_overlap": 128,
+    },
+}
+
+
+def detect_code_files(documents, code_extensions=None) -> tuple[list, list]:
+    """
+    Separate documents into code files and regular text files.
+
+    Args:
+        documents: List of LlamaIndex Document objects
+        code_extensions: Dict mapping file extensions to languages (defaults to CODE_EXTENSIONS)
+
+    Returns:
+        Tuple of (code_documents, text_documents)
+    """
+    if code_extensions is None:
+        code_extensions = CODE_EXTENSIONS
+
+    code_docs = []
+    text_docs = []
+
+    for doc in documents:
+        # Get file path from metadata
+        file_path = doc.metadata.get("file_path", "")
+        if not file_path:
+            # Fallback to file_name
+            file_path = doc.metadata.get("file_name", "")
+
+        if file_path:
+            file_ext = Path(file_path).suffix.lower()
+            if file_ext in code_extensions:
+                # Add language info to metadata
+                doc.metadata["language"] = code_extensions[file_ext]
+                doc.metadata["is_code"] = True
+                code_docs.append(doc)
+            else:
+                doc.metadata["is_code"] = False
+                text_docs.append(doc)
+        else:
+            # If no file path, treat as text
+            doc.metadata["is_code"] = False
+            text_docs.append(doc)
+
+    logger.info(f"Detected {len(code_docs)} code files and {len(text_docs)} text files")
+    return code_docs, text_docs
+
+
+def get_language_from_extension(file_path: str) -> Optional[str]:
+    """Get the programming language from file extension."""
+    ext = Path(file_path).suffix.lower()
+    return CODE_EXTENSIONS.get(ext)
+
+
+def create_ast_chunks(
+    documents,
+    max_chunk_size: int = 512,
+    chunk_overlap: int = 64,
+    metadata_template: str = "default",
+) -> list[str]:
+    """
+    Create AST-aware chunks from code documents using astchunk.
+
+    Args:
+        documents: List of code documents
+        max_chunk_size: Maximum characters per chunk
+        chunk_overlap: Number of AST nodes to overlap between chunks
+        metadata_template: Template for chunk metadata
+
+    Returns:
+        List of text chunks with preserved code structure
+    """
+    try:
+        from astchunk import ASTChunkBuilder
+    except ImportError as e:
+        logger.error(f"astchunk not available: {e}")
+        logger.info("Falling back to traditional chunking for code files")
+        return create_traditional_chunks(documents, max_chunk_size, chunk_overlap)
+
+    all_chunks = []
+
+    for doc in documents:
+        # Get language from metadata (set by detect_code_files)
+        language = doc.metadata.get("language")
+        if not language:
+            logger.warning(
+                "No language detected for document, falling back to traditional chunking"
+            )
+            traditional_chunks = create_traditional_chunks([doc], max_chunk_size, chunk_overlap)
+            all_chunks.extend(traditional_chunks)
+            continue
+
+        try:
+            # Configure astchunk
+            configs = {
+                "max_chunk_size": max_chunk_size,
+                "language": language,
+                "metadata_template": metadata_template,
+                "chunk_overlap": chunk_overlap if chunk_overlap > 0 else 0,
+            }
+
+            # Add repository-level metadata if available
+            repo_metadata = {
+                "file_path": doc.metadata.get("file_path", ""),
+                "file_name": doc.metadata.get("file_name", ""),
+                "creation_date": doc.metadata.get("creation_date", ""),
+                "last_modified_date": doc.metadata.get("last_modified_date", ""),
+            }
+            configs["repo_level_metadata"] = repo_metadata
+
+            # Create chunk builder and process
+            chunk_builder = ASTChunkBuilder(**configs)
+            code_content = doc.get_content()
+
+            if not code_content or not code_content.strip():
+                logger.warning("Empty code content, skipping")
+                continue
+
+            chunks = chunk_builder.chunkify(code_content)
+
+            # Extract text content from chunks
+            for chunk in chunks:
+                if hasattr(chunk, "text"):
+                    chunk_text = chunk.text
+                elif isinstance(chunk, dict) and "text" in chunk:
+                    chunk_text = chunk["text"]
+                elif isinstance(chunk, str):
+                    chunk_text = chunk
+                else:
+                    # Try to convert to string
+                    chunk_text = str(chunk)
+
+                if chunk_text and chunk_text.strip():
+                    all_chunks.append(chunk_text.strip())
+
+            logger.info(
+                f"Created {len(chunks)} AST chunks from {language} file: {doc.metadata.get('file_name', 'unknown')}"
+            )
+
+        except Exception as e:
+            logger.warning(f"AST chunking failed for {language} file: {e}")
+            logger.info("Falling back to traditional chunking")
+            traditional_chunks = create_traditional_chunks([doc], max_chunk_size, chunk_overlap)
+            all_chunks.extend(traditional_chunks)
+
+    return all_chunks
+
+
+def create_traditional_chunks(
+    documents, chunk_size: int = 256, chunk_overlap: int = 128
+) -> list[str]:
+    """
+    Create traditional text chunks using LlamaIndex SentenceSplitter.
+
+    Args:
+        documents: List of documents to chunk
+        chunk_size: Size of each chunk in characters
+        chunk_overlap: Overlap between chunks
+
+    Returns:
+        List of text chunks
+    """
+    # Handle invalid chunk_size values
+    if chunk_size <= 0:
+        logger.warning(f"Invalid chunk_size={chunk_size}, using default value of 256")
+        chunk_size = 256
+
+    # Ensure chunk_overlap is not negative and not larger than chunk_size
+    if chunk_overlap < 0:
+        chunk_overlap = 0
+    if chunk_overlap >= chunk_size:
+        chunk_overlap = chunk_size // 2
+
+    node_parser = SentenceSplitter(
+        chunk_size=chunk_size,
+        chunk_overlap=chunk_overlap,
+        separator=" ",
+        paragraph_separator="\n\n",
+    )
+
+    all_texts = []
+    for doc in documents:
+        try:
+            nodes = node_parser.get_nodes_from_documents([doc])
+            if nodes:
+                chunk_texts = [node.get_content() for node in nodes]
+                all_texts.extend(chunk_texts)
+                logger.debug(f"Created {len(chunk_texts)} traditional chunks from document")
+        except Exception as e:
+            logger.error(f"Traditional chunking failed for document: {e}")
+            # As last resort, add the raw content
+            content = doc.get_content()
+            if content and content.strip():
+                all_texts.append(content.strip())
+
+    return all_texts
+
+
+def create_text_chunks(
+    documents,
+    chunk_size: int = 256,
+    chunk_overlap: int = 128,
+    use_ast_chunking: bool = False,
+    ast_chunk_size: int = 512,
+    ast_chunk_overlap: int = 64,
+    code_file_extensions: Optional[list[str]] = None,
+    ast_fallback_traditional: bool = True,
+) -> list[str]:
+    """
+    Create text chunks from documents with optional AST support for code files.
+
+    Args:
+        documents: List of LlamaIndex Document objects
+        chunk_size: Size for traditional text chunks
+        chunk_overlap: Overlap for traditional text chunks
+        use_ast_chunking: Whether to use AST chunking for code files
+        ast_chunk_size: Size for AST chunks
+        ast_chunk_overlap: Overlap for AST chunks
+        code_file_extensions: Custom list of code file extensions
+        ast_fallback_traditional: Fall back to traditional chunking on AST errors
+
+    Returns:
+        List of text chunks
+    """
+    if not documents:
+        logger.warning("No documents provided for chunking")
+        return []
+
+    # Create a local copy of supported extensions for this function call
+    local_code_extensions = CODE_EXTENSIONS.copy()
+
+    # Update supported extensions if provided
+    if code_file_extensions:
+        # Map extensions to languages (simplified mapping)
+        ext_mapping = {
+            ".py": "python",
+            ".java": "java",
+            ".cs": "c_sharp",
+            ".ts": "typescript",
+            ".tsx": "typescript",
+        }
+        for ext in code_file_extensions:
+            if ext.lower() not in local_code_extensions:
+                # Try to guess language from extension
+                if ext.lower() in ext_mapping:
+                    local_code_extensions[ext.lower()] = ext_mapping[ext.lower()]
+                else:
+                    logger.warning(f"Unsupported extension {ext}, will use traditional chunking")
+
+    all_chunks = []
+
+    if use_ast_chunking:
+        # Separate code and text documents using local extensions
+        code_docs, text_docs = detect_code_files(documents, local_code_extensions)
+
+        # Process code files with AST chunking
+        if code_docs:
+            logger.info(f"Processing {len(code_docs)} code files with AST chunking")
+            try:
+                ast_chunks = create_ast_chunks(
+                    code_docs, max_chunk_size=ast_chunk_size, chunk_overlap=ast_chunk_overlap
+                )
+                all_chunks.extend(ast_chunks)
+                logger.info(f"Created {len(ast_chunks)} AST chunks from code files")
+            except Exception as e:
+                logger.error(f"AST chunking failed: {e}")
+                if ast_fallback_traditional:
+                    logger.info("Falling back to traditional chunking for code files")
+                    traditional_code_chunks = create_traditional_chunks(
+                        code_docs, chunk_size, chunk_overlap
+                    )
+                    all_chunks.extend(traditional_code_chunks)
+                else:
+                    raise
+
+        # Process text files with traditional chunking
+        if text_docs:
+            logger.info(f"Processing {len(text_docs)} text files with traditional chunking")
+            text_chunks = create_traditional_chunks(text_docs, chunk_size, chunk_overlap)
+            all_chunks.extend(text_chunks)
+            logger.info(f"Created {len(text_chunks)} traditional chunks from text files")
+    else:
+        # Use traditional chunking for all files
+        logger.info(f"Processing {len(documents)} documents with traditional chunking")
+        all_chunks = create_traditional_chunks(documents, chunk_size, chunk_overlap)
+
+    logger.info(f"Total chunks created: {len(all_chunks)}")
+    return all_chunks

apps/code_rag.py

@@ -0,0 +1,211 @@
+"""
+Code RAG example using AST-aware chunking for optimal code understanding.
+Specialized for code repositories with automatic language detection and
+optimized chunking parameters.
+"""
+
+import sys
+from pathlib import Path
+
+# Add parent directory to path for imports
+sys.path.insert(0, str(Path(__file__).parent))
+
+from base_rag_example import BaseRAGExample
+from chunking import CODE_EXTENSIONS, create_text_chunks
+from llama_index.core import SimpleDirectoryReader
+
+
+class CodeRAG(BaseRAGExample):
+    """Specialized RAG example for code repositories with AST-aware chunking."""
+
+    def __init__(self):
+        super().__init__(
+            name="Code",
+            description="Process and query code repositories with AST-aware chunking",
+            default_index_name="code_index",
+        )
+        # Override defaults for code-specific usage
+        self.embedding_model_default = "facebook/contriever"  # Good for code
+        self.max_items_default = -1  # Process all code files by default
+
+    def _add_specific_arguments(self, parser):
+        """Add code-specific arguments."""
+        code_group = parser.add_argument_group("Code Repository Parameters")
+
+        code_group.add_argument(
+            "--repo-dir",
+            type=str,
+            default=".",
+            help="Code repository directory to index (default: current directory)",
+        )
+        code_group.add_argument(
+            "--include-extensions",
+            nargs="+",
+            default=list(CODE_EXTENSIONS.keys()),
+            help="File extensions to include (default: supported code extensions)",
+        )
+        code_group.add_argument(
+            "--exclude-dirs",
+            nargs="+",
+            default=[
+                ".git",
+                "__pycache__",
+                "node_modules",
+                "venv",
+                ".venv",
+                "build",
+                "dist",
+                "target",
+            ],
+            help="Directories to exclude from indexing",
+        )
+        code_group.add_argument(
+            "--max-file-size",
+            type=int,
+            default=1000000,  # 1MB
+            help="Maximum file size in bytes to process (default: 1MB)",
+        )
+        code_group.add_argument(
+            "--include-comments",
+            action="store_true",
+            help="Include comments in chunking (useful for documentation)",
+        )
+        code_group.add_argument(
+            "--preserve-imports",
+            action="store_true",
+            default=True,
+            help="Try to preserve import statements in chunks (default: True)",
+        )
+
+    async def load_data(self, args) -> list[str]:
+        """Load code files and convert to AST-aware chunks."""
+        print(f"🔍 Scanning code repository: {args.repo_dir}")
+        print(f"📁 Including extensions: {args.include_extensions}")
+        print(f"🚫 Excluding directories: {args.exclude_dirs}")
+
+        # Check if repository directory exists
+        repo_path = Path(args.repo_dir)
+        if not repo_path.exists():
+            raise ValueError(f"Repository directory not found: {args.repo_dir}")
+
+        # Load code files with filtering
+        reader_kwargs = {
+            "recursive": True,
+            "encoding": "utf-8",
+            "required_exts": args.include_extensions,
+            "exclude_hidden": True,
+        }
+
+        # Create exclusion filter
+        def file_filter(file_path: str) -> bool:
+            """Filter out unwanted files and directories."""
+            path = Path(file_path)
+
+            # Check file size
+            try:
+                if path.stat().st_size > args.max_file_size:
+                    print(f"⚠️ Skipping large file: {path.name} ({path.stat().st_size} bytes)")
+                    return False
+            except Exception:
+                return False
+
+            # Check if in excluded directory
+            for exclude_dir in args.exclude_dirs:
+                if exclude_dir in path.parts:
+                    return False
+
+            return True
+
+        try:
+            # Load documents with file filtering
+            documents = SimpleDirectoryReader(
+                args.repo_dir,
+                file_extractor=None,  # Use default extractors
+                **reader_kwargs,
+            ).load_data(show_progress=True)
+
+            # Apply custom filtering
+            filtered_docs = []
+            for doc in documents:
+                file_path = doc.metadata.get("file_path", "")
+                if file_filter(file_path):
+                    filtered_docs.append(doc)
+
+            documents = filtered_docs
+
+        except Exception as e:
+            print(f"❌ Error loading code files: {e}")
+            return []
+
+        if not documents:
+            print(
+                f"❌ No code files found in {args.repo_dir} with extensions {args.include_extensions}"
+            )
+            return []
+
+        print(f"✅ Loaded {len(documents)} code files")
+
+        # Show breakdown by language/extension
+        ext_counts = {}
+        for doc in documents:
+            file_path = doc.metadata.get("file_path", "")
+            if file_path:
+                ext = Path(file_path).suffix.lower()
+                ext_counts[ext] = ext_counts.get(ext, 0) + 1
+
+        print("📊 Files by extension:")
+        for ext, count in sorted(ext_counts.items()):
+            print(f"   {ext}: {count} files")
+
+        # Use AST-aware chunking by default for code
+        print(
+            f"🧠 Using AST-aware chunking (chunk_size: {args.ast_chunk_size}, overlap: {args.ast_chunk_overlap})"
+        )
+
+        all_texts = create_text_chunks(
+            documents,
+            chunk_size=256,  # Fallback for non-code files
+            chunk_overlap=64,
+            use_ast_chunking=True,  # Always use AST for code RAG
+            ast_chunk_size=args.ast_chunk_size,
+            ast_chunk_overlap=args.ast_chunk_overlap,
+            code_file_extensions=args.include_extensions,
+            ast_fallback_traditional=True,
+        )
+
+        # Apply max_items limit if specified
+        if args.max_items > 0 and len(all_texts) > args.max_items:
+            print(f"⏳ Limiting to {args.max_items} chunks (from {len(all_texts)})")
+            all_texts = all_texts[: args.max_items]
+
+        print(f"✅ Generated {len(all_texts)} code chunks")
+        return all_texts
+
+
+if __name__ == "__main__":
+    import asyncio
+
+    # Example queries for code RAG
+    print("\n💻 Code RAG Example")
+    print("=" * 50)
+    print("\nExample queries you can try:")
+    print("- 'How does the embedding computation work?'")
+    print("- 'What are the main classes in this codebase?'")
+    print("- 'Show me the search implementation'")
+    print("- 'How is error handling implemented?'")
+    print("- 'What design patterns are used?'")
+    print("- 'Explain the chunking logic'")
+    print("\n🚀 Features:")
+    print("- ✅ AST-aware chunking preserves code structure")
+    print("- ✅ Automatic language detection")
+    print("- ✅ Smart filtering of large files and common excludes")
+    print("- ✅ Optimized for code understanding")
+    print("\nUsage examples:")
+    print("  python -m apps.code_rag --repo-dir ./my_project")
+    print(
+        "  python -m apps.code_rag --include-extensions .py .js --query 'How does authentication work?'"
+    )
+    print("\nOr run without --query for interactive mode\n")
+
+    rag = CodeRAG()
+    asyncio.run(rag.run())

apps/document_rag.py

@@ -9,7 +9,8 @@ from pathlib import Path
 # Add parent directory to path for imports
 sys.path.insert(0, str(Path(__file__).parent))
 
-from base_rag_example import BaseRAGExample, create_text_chunks
+from base_rag_example import BaseRAGExample
+from chunking import create_text_chunks
 from llama_index.core import SimpleDirectoryReader
 
 
@@ -44,6 +45,11 @@ class DocumentRAG(BaseRAGExample):
         doc_group.add_argument(
             "--chunk-overlap", type=int, default=128, help="Text chunk overlap (default: 128)"
         )
+        doc_group.add_argument(
+            "--enable-code-chunking",
+            action="store_true",
+            help="Enable AST-aware chunking for code files in the data directory",
+        )
 
     async def load_data(self, args) -> list[str]:
         """Load documents and convert to text chunks."""
@@ -76,9 +82,22 @@ class DocumentRAG(BaseRAGExample):
 
         print(f"Loaded {len(documents)} documents")
 
-        # Convert to text chunks
+        # Determine chunking strategy
+        use_ast = args.enable_code_chunking or getattr(args, "use_ast_chunking", False)
+
+        if use_ast:
+            print("Using AST-aware chunking for code files")
+
+        # Convert to text chunks with optional AST support
         all_texts = create_text_chunks(
-            documents, chunk_size=args.chunk_size, chunk_overlap=args.chunk_overlap
+            documents,
+            chunk_size=args.chunk_size,
+            chunk_overlap=args.chunk_overlap,
+            use_ast_chunking=use_ast,
+            ast_chunk_size=getattr(args, "ast_chunk_size", 512),
+            ast_chunk_overlap=getattr(args, "ast_chunk_overlap", 64),
+            code_file_extensions=getattr(args, "code_file_extensions", None),
+            ast_fallback_traditional=getattr(args, "ast_fallback_traditional", True),
         )
 
         # Apply max_items limit if specified
@@ -102,6 +121,10 @@ if __name__ == "__main__":
     print(
         "- 'What is the problem of developing pan gu model Huawei meets? (盘古大模型开发中遇到什么问题?)'"
     )
+    print("\n🚀 NEW: Code-aware chunking available!")
+    print("- Use --enable-code-chunking to enable AST-aware chunking for code files")
+    print("- Supports Python, Java, C#, TypeScript files")
+    print("- Better semantic understanding of code structure")
     print("\nOr run without --query for interactive mode\n")
 
     rag = DocumentRAG()

apps/email_rag.py

@@ -9,7 +9,8 @@ from pathlib import Path
 # Add parent directory to path for imports
 sys.path.insert(0, str(Path(__file__).parent))
 
-from base_rag_example import BaseRAGExample, create_text_chunks
+from base_rag_example import BaseRAGExample
+from chunking import create_text_chunks
 
 from .email_data.LEANN_email_reader import EmlxReader
 

benchmarks/data/.gitattributes

@@ -1,82 +0,0 @@
-*.7z filter=lfs diff=lfs merge=lfs -text
-*.arrow filter=lfs diff=lfs merge=lfs -text
-*.bin filter=lfs diff=lfs merge=lfs -text
-*.bz2 filter=lfs diff=lfs merge=lfs -text
-*.ckpt filter=lfs diff=lfs merge=lfs -text
-*.ftz filter=lfs diff=lfs merge=lfs -text
-*.gz filter=lfs diff=lfs merge=lfs -text
-*.h5 filter=lfs diff=lfs merge=lfs -text
-*.joblib filter=lfs diff=lfs merge=lfs -text
-*.lfs.* filter=lfs diff=lfs merge=lfs -text
-*.lz4 filter=lfs diff=lfs merge=lfs -text
-*.mds filter=lfs diff=lfs merge=lfs -text
-*.mlmodel filter=lfs diff=lfs merge=lfs -text
-*.model filter=lfs diff=lfs merge=lfs -text
-*.msgpack filter=lfs diff=lfs merge=lfs -text
-*.npy filter=lfs diff=lfs merge=lfs -text
-*.npz filter=lfs diff=lfs merge=lfs -text
-*.onnx filter=lfs diff=lfs merge=lfs -text
-*.ot filter=lfs diff=lfs merge=lfs -text
-*.parquet filter=lfs diff=lfs merge=lfs -text
-*.pb filter=lfs diff=lfs merge=lfs -text
-*.pickle filter=lfs diff=lfs merge=lfs -text
-*.pkl filter=lfs diff=lfs merge=lfs -text
-*.pt filter=lfs diff=lfs merge=lfs -text
-*.pth filter=lfs diff=lfs merge=lfs -text
-*.rar filter=lfs diff=lfs merge=lfs -text
-*.safetensors filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
-*.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
-*.tflite filter=lfs diff=lfs merge=lfs -text
-*.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
-*.xz filter=lfs diff=lfs merge=lfs -text
-*.zip filter=lfs diff=lfs merge=lfs -text
-*.zst filter=lfs diff=lfs merge=lfs -text
-*tfevents* filter=lfs diff=lfs merge=lfs -text
-# Audio files - uncompressed
-*.pcm filter=lfs diff=lfs merge=lfs -text
-*.sam filter=lfs diff=lfs merge=lfs -text
-*.raw filter=lfs diff=lfs merge=lfs -text
-# Audio files - compressed
-*.aac filter=lfs diff=lfs merge=lfs -text
-*.flac filter=lfs diff=lfs merge=lfs -text
-*.mp3 filter=lfs diff=lfs merge=lfs -text
-*.ogg filter=lfs diff=lfs merge=lfs -text
-*.wav filter=lfs diff=lfs merge=lfs -text
-# Image files - uncompressed
-*.bmp filter=lfs diff=lfs merge=lfs -text
-*.gif filter=lfs diff=lfs merge=lfs -text
-*.png filter=lfs diff=lfs merge=lfs -text
-*.tiff filter=lfs diff=lfs merge=lfs -text
-# Image files - compressed
-*.jpg filter=lfs diff=lfs merge=lfs -text
-*.jpeg filter=lfs diff=lfs merge=lfs -text
-*.webp filter=lfs diff=lfs merge=lfs -text
-# Video files - compressed
-*.mp4 filter=lfs diff=lfs merge=lfs -text
-*.webm filter=lfs diff=lfs merge=lfs -text
-ground_truth/dpr/id_map.json filter=lfs diff=lfs merge=lfs -text
-indices/dpr/dpr_diskann.passages.idx filter=lfs diff=lfs merge=lfs -text
-indices/dpr/dpr_diskann.passages.jsonl filter=lfs diff=lfs merge=lfs -text
-indices/dpr/dpr_diskann_disk.index filter=lfs diff=lfs merge=lfs -text
-indices/dpr/leann.labels.map filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/leann.labels.map filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/rpj_wiki.index filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/rpj_wiki.passages.0.idx filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/rpj_wiki.passages.0.jsonl filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/rpj_wiki.passages.1.idx filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/rpj_wiki.passages.1.jsonl filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/rpj_wiki.passages.2.idx filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/rpj_wiki.passages.2.jsonl filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/rpj_wiki.passages.3.idx filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/rpj_wiki.passages.3.jsonl filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/rpj_wiki.passages.4.idx filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/rpj_wiki.passages.4.jsonl filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/rpj_wiki.passages.5.idx filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/rpj_wiki.passages.5.jsonl filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/rpj_wiki.passages.6.idx filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/rpj_wiki.passages.6.jsonl filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/rpj_wiki.passages.7.idx filter=lfs diff=lfs merge=lfs -text
-indices/rpj_wiki/rpj_wiki.passages.7.jsonl filter=lfs diff=lfs merge=lfs -text

benchmarks/data/README.md

@@ -0,0 +1,44 @@
+---
+license: mit
+---
+
+# LEANN-RAG Evaluation Data
+
+This repository contains the necessary data to run the recall evaluation scripts for the [LEANN-RAG](https://huggingface.co/LEANN-RAG) project.
+
+## Dataset Components
+
+This dataset is structured into three main parts:
+
+1.  **Pre-built LEANN Indices**:
+    *   `dpr/`: A pre-built index for the DPR dataset.
+    *   `rpj_wiki/`: A pre-built index for the RPJ-Wiki dataset.
+    These indices were created using the `leann-core` library and are required by the `LeannSearcher`.
+
+2.  **Ground Truth Data**:
+    *   `ground_truth/`: Contains the ground truth files (`flat_results_nq_k3.json`) for both the DPR and RPJ-Wiki datasets. These files map queries to the original passage IDs from the Natural Questions benchmark, evaluated using the Contriever model.
+
+3.  **Queries**:
+    *   `queries/`: Contains the `nq_open.jsonl` file with the Natural Questions queries used for the evaluation.
+
+## Usage
+
+To use this data, you can download it locally using the `huggingface-hub` library. First, install the library:
+
+```bash
+pip install huggingface-hub
+```
+
+Then, you can download the entire dataset to a local directory (e.g., `data/`) with the following Python script:
+
+```python
+from huggingface_hub import snapshot_download
+
+snapshot_download(
+    repo_id="LEANN-RAG/leann-rag-evaluation-data",
+    repo_type="dataset",
+    local_dir="data"
+)
+```
+
+This will download all the necessary files into a local `data` folder, preserving the repository structure. The evaluation scripts in the main [LEANN-RAG Space](https://huggingface.co/LEANN-RAG) are configured to work with this data structure.

benchmarks/run_evaluation.py

@@ -12,7 +12,7 @@ import time
 from pathlib import Path
 
 import numpy as np
-from leann.api import LeannBuilder, LeannSearcher
+from leann.api import LeannBuilder, LeannChat, LeannSearcher
 
 
 def download_data_if_needed(data_root: Path, download_embeddings: bool = False):
@@ -197,6 +197,25 @@ def main():
     parser.add_argument(
         "--ef-search", type=int, default=120, help="The 'efSearch' parameter for HNSW."
     )
+    parser.add_argument(
+        "--batch-size",
+        type=int,
+        default=0,
+        help="Batch size for HNSW batched search (0 disables batching)",
+    )
+    parser.add_argument(
+        "--llm-type",
+        type=str,
+        choices=["ollama", "hf", "openai", "gemini", "simulated"],
+        default="ollama",
+        help="LLM backend type to optionally query during evaluation (default: ollama)",
+    )
+    parser.add_argument(
+        "--llm-model",
+        type=str,
+        default="qwen3:1.7b",
+        help="LLM model identifier for the chosen backend (default: qwen3:1.7b)",
+    )
     args = parser.parse_args()
 
     # --- Path Configuration ---
@@ -318,9 +337,24 @@ def main():
 
         for i in range(num_eval_queries):
             start_time = time.time()
-            new_results = searcher.search(queries[i], top_k=args.top_k, ef=args.ef_search)
+            new_results = searcher.search(
+                queries[i],
+                top_k=args.top_k,
+                complexity=args.ef_search,
+                batch_size=args.batch_size,
+            )
             search_times.append(time.time() - start_time)
 
+            # Optional: also call the LLM with configurable backend/model (does not affect recall)
+            llm_config = {"type": args.llm_type, "model": args.llm_model}
+            chat = LeannChat(args.index_path, llm_config=llm_config, searcher=searcher)
+            answer = chat.ask(
+                queries[i],
+                top_k=args.top_k,
+                complexity=args.ef_search,
+                batch_size=args.batch_size,
+            )
+            print(f"Answer: {answer}")
             # Correct Recall Calculation: Based on TEXT content
             new_texts = {result.text for result in new_results}
 

benchmarks/simple_mac_tpt_test.py

@@ -20,7 +20,7 @@ except ImportError:
 
 @dataclass
 class BenchmarkConfig:
-    model_path: str = "facebook/contriever"
+    model_path: str = "facebook/contriever-msmarco"
     batch_sizes: list[int] = None
     seq_length: int = 256
     num_runs: int = 5
@@ -34,7 +34,7 @@ class BenchmarkConfig:
 
     def __post_init__(self):
         if self.batch_sizes is None:
-            self.batch_sizes = [1, 2, 4, 8, 16, 32, 64]
+            self.batch_sizes = [1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024]
 
 
 class MLXBenchmark:
@@ -179,11 +179,14 @@ class Benchmark:
 
     def _run_inference(self, input_ids: torch.Tensor) -> float:
         attention_mask = torch.ones_like(input_ids)
-
+        # print shape of input_ids and attention_mask
+        print(f"input_ids shape: {input_ids.shape}")
+        print(f"attention_mask shape: {attention_mask.shape}")
         start_time = time.time()
         with torch.no_grad():
             self.model(input_ids=input_ids, attention_mask=attention_mask)
-        # mps sync
+        if torch.cuda.is_available():
+            torch.cuda.synchronize()
         if torch.backends.mps.is_available():
             torch.mps.synchronize()
         end_time = time.time()

docs/ast_chunking_guide.md

@@ -0,0 +1,128 @@
+# AST-Aware Code chunking guide
+
+## Overview
+
+This guide covers best practices for using AST-aware code chunking in LEANN. AST chunking provides better semantic understanding of code structure compared to traditional text-based chunking.
+
+## Quick Start
+
+### Basic Usage
+
+```bash
+# Enable AST chunking for mixed content (code + docs)
+python -m apps.document_rag --enable-code-chunking --data-dir ./my_project
+
+# Specialized code repository indexing
+python -m apps.code_rag --repo-dir ./my_codebase
+
+# Global CLI with AST support
+leann build my-code-index --docs ./src --use-ast-chunking
+```
+
+### Installation
+
+```bash
+# Install LEANN with AST chunking support
+uv pip install -e "."
+```
+
+## Best Practices
+
+### When to Use AST Chunking
+
+✅ **Recommended for:**
+- Code repositories with multiple languages
+- Mixed documentation and code content
+- Complex codebases with deep function/class hierarchies
+- When working with Claude Code for code assistance
+
+❌ **Not recommended for:**
+- Pure text documents
+- Very large files (>1MB)
+- Languages not supported by tree-sitter
+
+### Optimal Configuration
+
+```bash
+# Recommended settings for most codebases
+python -m apps.code_rag \
+    --repo-dir ./src \
+    --ast-chunk-size 768 \
+    --ast-chunk-overlap 96 \
+    --exclude-dirs .git __pycache__ node_modules build dist
+```
+
+### Supported Languages
+
+| Extension | Language | Status |
+|-----------|----------|--------|
+| `.py` | Python | ✅ Full support |
+| `.java` | Java | ✅ Full support |
+| `.cs` | C# | ✅ Full support |
+| `.ts`, `.tsx` | TypeScript | ✅ Full support |
+| `.js`, `.jsx` | JavaScript | ✅ Via TypeScript parser |
+
+## Integration Examples
+
+### Document RAG with Code Support
+
+```python
+# Enable code chunking in document RAG
+python -m apps.document_rag \
+    --enable-code-chunking \
+    --data-dir ./project \
+    --query "How does authentication work in the codebase?"
+```
+
+### Claude Code Integration
+
+When using with Claude Code MCP server, AST chunking provides better context for:
+- Code completion and suggestions
+- Bug analysis and debugging
+- Architecture understanding
+- Refactoring assistance
+
+## Troubleshooting
+
+### Common Issues
+
+1. **Fallback to Traditional Chunking**
+   - Normal behavior for unsupported languages
+   - Check logs for specific language support
+
+2. **Performance with Large Files**
+   - Adjust `--max-file-size` parameter
+   - Use `--exclude-dirs` to skip unnecessary directories
+
+3. **Quality Issues**
+   - Try different `--ast-chunk-size` values (512, 768, 1024)
+   - Adjust overlap for better context preservation
+
+### Debug Mode
+
+```bash
+export LEANN_LOG_LEVEL=DEBUG
+python -m apps.code_rag --repo-dir ./my_code
+```
+
+## Migration from Traditional Chunking
+
+Existing workflows continue to work without changes. To enable AST chunking:
+
+```bash
+# Before
+python -m apps.document_rag --chunk-size 256
+
+# After (maintains traditional chunking for non-code files)
+python -m apps.document_rag --enable-code-chunking --chunk-size 256 --ast-chunk-size 768
+```
+
+## References
+
+- [astchunk GitHub Repository](https://github.com/yilinjz/astchunk)
+- [LEANN MCP Integration](../packages/leann-mcp/README.md)
+- [Research Paper](https://arxiv.org/html/2506.15655v1)
+
+---
+
+**Note**: AST chunking maintains full backward compatibility while enhancing code understanding capabilities.

docs/features.md

@@ -3,6 +3,7 @@
 ## 🔥 Core Features
 
 - **🔄 Real-time Embeddings** - Eliminate heavy embedding storage with dynamic computation using optimized ZMQ servers and highly optimized search paradigm (overlapping and batching) with highly optimized embedding engine
+- **🧠 AST-Aware Code Chunking** - Intelligent code chunking that preserves semantic boundaries (functions, classes, methods) for Python, Java, C#, and TypeScript files
 - **📈 Scalable Architecture** - Handles millions of documents on consumer hardware; the larger your dataset, the more LEANN can save
 - **🎯 Graph Pruning** - Advanced techniques to minimize the storage overhead of vector search to a limited footprint
 - **🏗️ Pluggable Backends** - HNSW/FAISS (default), with optional DiskANN for large-scale deployments

docs/metadata_filtering.md

@@ -0,0 +1,300 @@
+# LEANN Metadata Filtering Usage Guide
+
+## Overview
+
+Leann possesses metadata filtering capabilities that allow you to filter search results based on arbitrary metadata fields set during chunking. This feature enables use cases like spoiler-free book search, document filtering by date/type, code search by file type, and potentially much more.
+
+## Basic Usage
+
+### Adding Metadata to Your Documents
+
+When building your index, add metadata to each text chunk:
+
+```python
+from leann.api import LeannBuilder
+
+builder = LeannBuilder("hnsw")
+
+# Add text with metadata
+builder.add_text(
+    text="Chapter 1: Alice falls down the rabbit hole",
+    metadata={
+        "chapter": 1,
+        "character": "Alice",
+        "themes": ["adventure", "curiosity"],
+        "word_count": 150
+    }
+)
+
+builder.build_index("alice_in_wonderland_index")
+```
+
+### Searching with Metadata Filters
+
+Use the `metadata_filters` parameter in search calls:
+
+```python
+from leann.api import LeannSearcher
+
+searcher = LeannSearcher("alice_in_wonderland_index")
+
+# Search with filters
+results = searcher.search(
+    query="What happens to Alice?",
+    top_k=10,
+    metadata_filters={
+        "chapter": {"<=": 5},           # Only chapters 1-5
+        "spoiler_level": {"!=": "high"} # No high spoilers
+    }
+)
+```
+
+## Filter Syntax
+
+### Basic Structure
+
+```python
+metadata_filters = {
+    "field_name": {"operator": value},
+    "another_field": {"operator": value}
+}
+```
+
+### Supported Operators
+
+#### Comparison Operators
+- `"=="`: Equal to
+- `"!="`: Not equal to
+- `"<"`: Less than
+- `"<="`: Less than or equal
+- `">"`: Greater than
+- `">="`: Greater than or equal
+
+```python
+# Examples
+{"chapter": {"==": 1}}           # Exactly chapter 1
+{"page": {">": 100}}            # Pages after 100
+{"rating": {">=": 4.0}}         # Rating 4.0 or higher
+{"word_count": {"<": 500}}      # Short passages
+```
+
+#### Membership Operators
+- `"in"`: Value is in list
+- `"not_in"`: Value is not in list
+
+```python
+# Examples
+{"character": {"in": ["Alice", "Bob"]}}      # Alice OR Bob
+{"genre": {"not_in": ["horror", "thriller"]}} # Exclude genres
+{"tags": {"in": ["fiction", "adventure"]}}   # Any of these tags
+```
+
+#### String Operators
+- `"contains"`: String contains substring
+- `"starts_with"`: String starts with prefix
+- `"ends_with"`: String ends with suffix
+
+```python
+# Examples
+{"title": {"contains": "alice"}}        # Title contains "alice"
+{"filename": {"ends_with": ".py"}}      # Python files
+{"author": {"starts_with": "Dr."}}      # Authors with "Dr." prefix
+```
+
+#### Boolean Operators
+- `"is_true"`: Field is truthy
+- `"is_false"`: Field is falsy
+
+```python
+# Examples
+{"is_published": {"is_true": True}}     # Published content
+{"is_draft": {"is_false": False}}       # Not drafts
+```
+
+### Multiple Operators on Same Field
+
+You can apply multiple operators to the same field (AND logic):
+
+```python
+metadata_filters = {
+    "word_count": {
+        ">=": 100,    # At least 100 words
+        "<=": 500     # At most 500 words
+    }
+}
+```
+
+### Compound Filters
+
+Multiple fields are combined with AND logic:
+
+```python
+metadata_filters = {
+    "chapter": {"<=": 10},              # Up to chapter 10
+    "character": {"==": "Alice"},       # About Alice
+    "spoiler_level": {"!=": "high"}     # No major spoilers
+}
+```
+
+## Use Case Examples
+
+### 1. Spoiler-Free Book Search
+
+```python
+# Reader has only read up to chapter 5
+def search_spoiler_free(query, max_chapter):
+    return searcher.search(
+        query=query,
+        metadata_filters={
+            "chapter": {"<=": max_chapter},
+            "spoiler_level": {"in": ["none", "low"]}
+        }
+    )
+
+results = search_spoiler_free("What happens to Alice?", max_chapter=5)
+```
+
+### 2. Document Management by Date
+
+```python
+# Find recent documents
+recent_docs = searcher.search(
+    query="project updates",
+    metadata_filters={
+        "date": {">=": "2024-01-01"},
+        "document_type": {"==": "report"}
+    }
+)
+```
+
+### 3. Code Search by File Type
+
+```python
+# Search only Python files
+python_code = searcher.search(
+    query="authentication function",
+    metadata_filters={
+        "file_extension": {"==": ".py"},
+        "lines_of_code": {"<": 100}
+    }
+)
+```
+
+### 4. Content Filtering by Audience
+
+```python
+# Age-appropriate content
+family_content = searcher.search(
+    query="adventure stories",
+    metadata_filters={
+        "age_rating": {"in": ["G", "PG"]},
+        "content_warnings": {"not_in": ["violence", "adult_themes"]}
+    }
+)
+```
+
+### 5. Multi-Book Series Management
+
+```python
+# Search across first 3 books only
+early_series = searcher.search(
+    query="character development",
+    metadata_filters={
+        "series": {"==": "Harry Potter"},
+        "book_number": {"<=": 3}
+    }
+)
+```
+
+## Running the Example
+
+You can see metadata filtering in action with our spoiler-free book RAG example:
+
+```bash
+# Don't forget to set up the environment
+uv venv
+source .venv/bin/activate
+
+# Set your OpenAI API key (required for embeddings, but you can update the example locally and use ollama instead)
+export OPENAI_API_KEY="your-api-key-here"
+
+# Run the spoiler-free book RAG example
+uv run examples/spoiler_free_book_rag.py
+```
+
+This example demonstrates:
+- Building an index with metadata (chapter numbers, characters, themes, locations)
+- Searching with filters to avoid spoilers (e.g., only show results up to chapter 5)
+- Different scenarios for readers at various points in the book
+
+The example uses Alice's Adventures in Wonderland as sample data and shows how you can search for information without revealing plot points from later chapters.
+
+## Advanced Patterns
+
+### Custom Chunking with metadata
+
+```python
+def chunk_book_with_metadata(book_text, book_info):
+    chunks = []
+
+    for chapter_num, chapter_text in parse_chapters(book_text):
+        # Extract entities, themes, etc.
+        characters = extract_characters(chapter_text)
+        themes = classify_themes(chapter_text)
+        spoiler_level = assess_spoiler_level(chapter_text, chapter_num)
+
+        # Create chunks with rich metadata
+        for paragraph in split_paragraphs(chapter_text):
+            chunks.append({
+                "text": paragraph,
+                "metadata": {
+                    "book_title": book_info["title"],
+                    "chapter": chapter_num,
+                    "characters": characters,
+                    "themes": themes,
+                    "spoiler_level": spoiler_level,
+                    "word_count": len(paragraph.split()),
+                    "reading_level": calculate_reading_level(paragraph)
+                }
+            })
+
+    return chunks
+```
+
+## Performance Considerations
+
+### Efficient Filtering Strategies
+
+1. **Post-search filtering**: Applies filters after vector search, which should be efficient for typical result sets (10-100 results).
+
+2. **Metadata design**: Keep metadata fields simple and avoid deeply nested structures.
+
+### Best Practices
+
+1. **Consistent metadata schema**: Use consistent field names and value types across your documents.
+
+2. **Reasonable metadata size**: Keep metadata reasonably sized to avoid storage overhead.
+
+3. **Type consistency**: Use consistent data types for the same fields (e.g., always integers for chapter numbers).
+
+4. **Index multiple granularities**: Consider chunking at different levels (paragraph, section, chapter) with appropriate metadata.
+
+### Adding Metadata to Existing Indices
+
+To add metadata filtering to existing indices, you'll need to rebuild them with metadata:
+
+```python
+# Read existing passages and add metadata
+def add_metadata_to_existing_chunks(chunks):
+    for chunk in chunks:
+        # Extract or assign metadata based on content
+        chunk["metadata"] = extract_metadata(chunk["text"])
+    return chunks
+
+# Rebuild index with metadata
+enhanced_chunks = add_metadata_to_existing_chunks(existing_chunks)
+builder = LeannBuilder("hnsw")
+for chunk in enhanced_chunks:
+    builder.add_text(chunk["text"], chunk["metadata"])
+builder.build_index("enhanced_index")
+```

examples/spoiler_free_book_rag.py

@@ -0,0 +1,250 @@
+#!/usr/bin/env python3
+"""
+Spoiler-Free Book RAG Example using LEANN Metadata Filtering
+
+This example demonstrates how to use LEANN's metadata filtering to create
+a spoiler-free book RAG system where users can search for information
+up to a specific chapter they've read.
+
+Usage:
+    python spoiler_free_book_rag.py
+"""
+
+import os
+import sys
+from typing import Any, Optional
+
+# Add LEANN to path (adjust path as needed)
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "../packages/leann-core/src"))
+
+from leann.api import LeannBuilder, LeannSearcher
+
+
+def chunk_book_with_metadata(book_title: str = "Sample Book") -> list[dict[str, Any]]:
+    """
+    Create sample book chunks with metadata for demonstration.
+
+    In a real implementation, this would parse actual book files (epub, txt, etc.)
+    and extract chapter boundaries, character mentions, etc.
+
+    Args:
+        book_title: Title of the book
+
+    Returns:
+        List of chunk dictionaries with text and metadata
+    """
+    # Sample book chunks with metadata
+    # In practice, you'd use proper text processing libraries
+
+    sample_chunks = [
+        {
+            "text": "Alice was beginning to get very tired of sitting by her sister on the bank, and of having nothing to do.",
+            "metadata": {
+                "book": book_title,
+                "chapter": 1,
+                "page": 1,
+                "characters": ["Alice", "Sister"],
+                "themes": ["boredom", "curiosity"],
+                "location": "riverbank",
+            },
+        },
+        {
+            "text": "So she was considering in her own mind (as well as she could, for the hot day made her feel very sleepy and stupid), whether the pleasure of making a daisy-chain would be worth the trouble of getting up and picking the daisies, when suddenly a White Rabbit with pink eyes ran close by her.",
+            "metadata": {
+                "book": book_title,
+                "chapter": 1,
+                "page": 2,
+                "characters": ["Alice", "White Rabbit"],
+                "themes": ["decision", "surprise", "magic"],
+                "location": "riverbank",
+            },
+        },
+        {
+            "text": "Alice found herself falling down a very deep well. Either the well was very deep, or she fell very slowly, for she had plenty of time as she fell to look about her and to wonder what was going to happen next.",
+            "metadata": {
+                "book": book_title,
+                "chapter": 2,
+                "page": 15,
+                "characters": ["Alice"],
+                "themes": ["falling", "wonder", "transformation"],
+                "location": "rabbit hole",
+            },
+        },
+        {
+            "text": "Alice meets the Cheshire Cat, who tells her that everyone in Wonderland is mad, including Alice herself.",
+            "metadata": {
+                "book": book_title,
+                "chapter": 6,
+                "page": 85,
+                "characters": ["Alice", "Cheshire Cat"],
+                "themes": ["madness", "philosophy", "identity"],
+                "location": "Duchess's house",
+            },
+        },
+        {
+            "text": "At the Queen's croquet ground, Alice witnesses the absurd trial that reveals the arbitrary nature of Wonderland's justice system.",
+            "metadata": {
+                "book": book_title,
+                "chapter": 8,
+                "page": 120,
+                "characters": ["Alice", "Queen of Hearts", "King of Hearts"],
+                "themes": ["justice", "absurdity", "authority"],
+                "location": "Queen's court",
+            },
+        },
+        {
+            "text": "Alice realizes that Wonderland was all a dream, even the Rabbit, as she wakes up on the riverbank next to her sister.",
+            "metadata": {
+                "book": book_title,
+                "chapter": 12,
+                "page": 180,
+                "characters": ["Alice", "Sister", "Rabbit"],
+                "themes": ["revelation", "reality", "growth"],
+                "location": "riverbank",
+            },
+        },
+    ]
+
+    return sample_chunks
+
+
+def build_spoiler_free_index(book_chunks: list[dict[str, Any]], index_name: str) -> str:
+    """
+    Build a LEANN index with book chunks that include spoiler metadata.
+
+    Args:
+        book_chunks: List of book chunks with metadata
+        index_name: Name for the index
+
+    Returns:
+        Path to the built index
+    """
+    print(f"📚 Building spoiler-free book index: {index_name}")
+
+    # Initialize LEANN builder
+    builder = LeannBuilder(
+        backend_name="hnsw", embedding_model="text-embedding-3-small", embedding_mode="openai"
+    )
+
+    # Add each chunk with its metadata
+    for chunk in book_chunks:
+        builder.add_text(text=chunk["text"], metadata=chunk["metadata"])
+
+    # Build the index
+    index_path = f"{index_name}_book_index"
+    builder.build_index(index_path)
+
+    print(f"✅ Index built successfully: {index_path}")
+    return index_path
+
+
+def spoiler_free_search(
+    index_path: str,
+    query: str,
+    max_chapter: int,
+    character_filter: Optional[list[str]] = None,
+) -> list[dict[str, Any]]:
+    """
+    Perform a spoiler-free search on the book index.
+
+    Args:
+        index_path: Path to the LEANN index
+        query: Search query
+        max_chapter: Maximum chapter number to include
+        character_filter: Optional list of characters to focus on
+
+    Returns:
+        List of search results safe for the reader
+    """
+    print(f"🔍 Searching: '{query}' (up to chapter {max_chapter})")
+
+    searcher = LeannSearcher(index_path)
+
+    metadata_filters = {"chapter": {"<=": max_chapter}}
+
+    if character_filter:
+        metadata_filters["characters"] = {"contains": character_filter[0]}
+
+    results = searcher.search(query=query, top_k=10, metadata_filters=metadata_filters)
+
+    return results
+
+
+def demo_spoiler_free_rag():
+    """
+    Demonstrate the spoiler-free book RAG system.
+    """
+    print("🎭 Spoiler-Free Book RAG Demo")
+    print("=" * 40)
+
+    # Step 1: Prepare book data
+    book_title = "Alice's Adventures in Wonderland"
+    book_chunks = chunk_book_with_metadata(book_title)
+
+    print(f"📖 Loaded {len(book_chunks)} chunks from '{book_title}'")
+
+    # Step 2: Build the index (in practice, this would be done once)
+    try:
+        index_path = build_spoiler_free_index(book_chunks, "alice_wonderland")
+    except Exception as e:
+        print(f"❌ Failed to build index (likely missing dependencies): {e}")
+        print(
+            "💡 This demo shows the filtering logic - actual indexing requires LEANN dependencies"
+        )
+        return
+
+    # Step 3: Demonstrate various spoiler-free searches
+    search_scenarios = [
+        {
+            "description": "Reader who has only read Chapter 1",
+            "query": "What can you tell me about the rabbit?",
+            "max_chapter": 1,
+        },
+        {
+            "description": "Reader who has read up to Chapter 5",
+            "query": "Tell me about Alice's adventures",
+            "max_chapter": 5,
+        },
+        {
+            "description": "Reader who has read most of the book",
+            "query": "What does the Cheshire Cat represent?",
+            "max_chapter": 10,
+        },
+        {
+            "description": "Reader who has read the whole book",
+            "query": "What can you tell me about the rabbit?",
+            "max_chapter": 12,
+        },
+    ]
+
+    for scenario in search_scenarios:
+        print(f"\n📚 Scenario: {scenario['description']}")
+        print(f"   Query: {scenario['query']}")
+
+        try:
+            results = spoiler_free_search(
+                index_path=index_path,
+                query=scenario["query"],
+                max_chapter=scenario["max_chapter"],
+            )
+
+            print(f"   📄 Found {len(results)} results:")
+            for i, result in enumerate(results[:3], 1):  # Show top 3
+                chapter = result.metadata.get("chapter", "?")
+                location = result.metadata.get("location", "?")
+                print(f"      {i}. Chapter {chapter} ({location}): {result.text[:80]}...")
+
+        except Exception as e:
+            print(f"   ❌ Search failed: {e}")
+
+
+if __name__ == "__main__":
+    print("📚 LEANN Spoiler-Free Book RAG Example")
+    print("=====================================")
+
+    try:
+        demo_spoiler_free_rag()
+    except ImportError as e:
+        print(f"❌ Cannot run demo due to missing dependencies: {e}")
+    except Exception as e:
+        print(f"❌ Error running demo: {e}")

packages/leann-backend-diskann/leann_backend_diskann/diskann_embedding_server.py

@@ -83,9 +83,7 @@ def create_diskann_embedding_server(
 
     logger.info(f"Loading PassageManager with metadata_file_path: {passages_file}")
     passages = PassageManager(meta["passage_sources"], metadata_file_path=passages_file)
-    logger.info(
-        f"Loaded PassageManager with {len(passages.global_offset_map)} passages from metadata"
-    )
+    logger.info(f"Loaded PassageManager with {len(passages)} passages from metadata")
 
     # Import protobuf after ensuring the path is correct
     try:

packages/leann-backend-diskann/pyproject.toml

@@ -4,8 +4,8 @@ build-backend = "scikit_build_core.build"
 
 [project]
 name = "leann-backend-diskann"
-version = "0.3.1"
-dependencies = ["leann-core==0.3.1", "numpy", "protobuf>=3.19.0"]
+version = "0.3.2"
+dependencies = ["leann-core==0.3.2", "numpy", "protobuf>=3.19.0"]
 
 [tool.scikit-build]
 # Key: simplified CMake path

packages/leann-backend-hnsw/CMakeLists.txt

@@ -49,9 +49,28 @@ set(BUILD_TESTING OFF CACHE BOOL "" FORCE)
 set(FAISS_ENABLE_C_API OFF CACHE BOOL "" FORCE)
 set(FAISS_OPT_LEVEL "generic" CACHE STRING "" FORCE)
 
-# Disable additional SIMD versions to speed up compilation
+# Disable x86-specific SIMD optimizations (important for ARM64 compatibility)
 set(FAISS_ENABLE_AVX2 OFF CACHE BOOL "" FORCE)
 set(FAISS_ENABLE_AVX512 OFF CACHE BOOL "" FORCE)
+set(FAISS_ENABLE_SSE4_1 OFF CACHE BOOL "" FORCE)
+
+# ARM64-specific configuration
+if(CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64|arm64")
+    message(STATUS "Configuring Faiss for ARM64 architecture")
+
+    if(CMAKE_SYSTEM_NAME STREQUAL "Linux")
+        # Use SVE optimization level for ARM64 Linux (as seen in Faiss conda build)
+        set(FAISS_OPT_LEVEL "sve" CACHE STRING "" FORCE)
+        message(STATUS "Setting FAISS_OPT_LEVEL to 'sve' for ARM64 Linux")
+    else()
+        # Use generic optimization for other ARM64 platforms (like macOS)
+        set(FAISS_OPT_LEVEL "generic" CACHE STRING "" FORCE)
+        message(STATUS "Setting FAISS_OPT_LEVEL to 'generic' for ARM64 ${CMAKE_SYSTEM_NAME}")
+    endif()
+
+    # ARM64 compatibility: Faiss submodule has been modified to fix x86 header inclusion
+    message(STATUS "Using ARM64-compatible Faiss submodule")
+endif()
 
 # Additional optimization options from INSTALL.md
 set(CMAKE_BUILD_TYPE "Release" CACHE STRING "" FORCE)

packages/leann-backend-hnsw/leann_backend_hnsw/hnsw_backend.py

@@ -1,6 +1,7 @@
 import logging
 import os
 import shutil
+import time
 from pathlib import Path
 from typing import Any, Literal, Optional
 
@@ -236,6 +237,7 @@ class HNSWSearcher(BaseSearcher):
         distances = np.empty((batch_size_query, top_k), dtype=np.float32)
         labels = np.empty((batch_size_query, top_k), dtype=np.int64)
 
+        search_time = time.time()
         self._index.search(
             query.shape[0],
             faiss.swig_ptr(query),
@@ -244,7 +246,8 @@ class HNSWSearcher(BaseSearcher):
             faiss.swig_ptr(labels),
             params,
         )
-
+        search_time = time.time() - search_time
+        logger.info(f"  Search time in HNSWSearcher.search() backend: {search_time} seconds")
         string_labels = [[str(int_label) for int_label in batch_labels] for batch_labels in labels]
 
         return {"labels": string_labels, "distances": distances}

packages/leann-backend-hnsw/leann_backend_hnsw/hnsw_embedding_server.py

@@ -90,9 +90,7 @@ def create_hnsw_embedding_server(
         embedding_dim: int = int(meta.get("dimensions", 0))
     except Exception:
         embedding_dim = 0
-    logger.info(
-        f"Loaded PassageManager with {len(passages.global_offset_map)} passages from metadata"
-    )
+    logger.info(f"Loaded PassageManager with {len(passages)} passages from metadata")
 
     # (legacy ZMQ thread removed; using shutdown-capable server only)
 

packages/leann-backend-hnsw/pyproject.toml

@@ -6,10 +6,10 @@ build-backend = "scikit_build_core.build"
 
 [project]
 name = "leann-backend-hnsw"
-version = "0.3.1"
+version = "0.3.2"
 description = "Custom-built HNSW (Faiss) backend for the Leann toolkit."
 dependencies = [
-    "leann-core==0.3.1",
+    "leann-core==0.3.2",
     "numpy",
     "pyzmq>=23.0.0",
     "msgpack>=1.0.0",

packages/leann-core/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "leann-core"
-version = "0.3.1"
+version = "0.3.2"
 description = "Core API and plugin system for LEANN"
 readme = "README.md"
 requires-python = ">=3.9"

packages/leann-core/src/leann/api.py

@@ -10,7 +10,7 @@ import time
 import warnings
 from dataclasses import dataclass, field
 from pathlib import Path
-from typing import Any, Literal, Optional
+from typing import Any, Literal, Optional, Union
 
 import numpy as np
 
@@ -18,6 +18,7 @@ from leann.interface import LeannBackendSearcherInterface
 
 from .chat import get_llm
 from .interface import LeannBackendFactoryInterface
+from .metadata_filter import MetadataFilterEngine
 from .registry import BACKEND_REGISTRY
 
 logger = logging.getLogger(__name__)
@@ -119,9 +120,13 @@ class PassageManager:
     def __init__(
         self, passage_sources: list[dict[str, Any]], metadata_file_path: Optional[str] = None
     ):
-        self.offset_maps = {}
-        self.passage_files = {}
-        self.global_offset_map = {}  # Combined map for fast lookup
+        self.offset_maps: dict[str, dict[str, int]] = {}
+        self.passage_files: dict[str, str] = {}
+        # Avoid materializing a single gigantic global map to reduce memory
+        # footprint on very large corpora (e.g., 60M+ passages). Instead, keep
+        # per-shard maps and do a lightweight per-shard lookup on demand.
+        self._total_count: int = 0
+        self.filter_engine = MetadataFilterEngine()  # Initialize filter engine
 
         # Derive index base name for standard sibling fallbacks, e.g., <index_name>.passages.*
         index_name_base = None
@@ -142,12 +147,25 @@ class PassageManager:
                 default_name: Optional[str],
                 source_dict: dict[str, Any],
             ) -> list[Path]:
+                """
+                Build an ordered list of candidate paths. For relative paths specified in
+                metadata, prefer resolution relative to the metadata file directory first,
+                then fall back to CWD-based resolution, and finally to conventional
+                sibling defaults (e.g., <index_base>.passages.idx / .jsonl).
+                """
                 candidates: list[Path] = []
-                # 1) Primary as-is (absolute or relative)
+                # 1) Primary path
                 if primary:
                     p = Path(primary)
-                    candidates.append(p if p.is_absolute() else (Path.cwd() / p))
-                # 2) metadata-relative explicit relative key
+                    if p.is_absolute():
+                        candidates.append(p)
+                    else:
+                        # Prefer metadata-relative resolution for relative paths
+                        if metadata_file_path:
+                            candidates.append(Path(metadata_file_path).parent / p)
+                        # Also consider CWD-relative as a fallback for legacy layouts
+                        candidates.append(Path.cwd() / p)
+                # 2) metadata-relative explicit relative key (if present)
                 if metadata_file_path and source_dict.get(relative_key):
                     candidates.append(Path(metadata_file_path).parent / source_dict[relative_key])
                 # 3) metadata-relative standard sibling filename
@@ -177,23 +195,78 @@ class PassageManager:
                 raise FileNotFoundError(f"Passage index file not found: {index_file}")
 
             with open(index_file, "rb") as f:
-                offset_map = pickle.load(f)
+                offset_map: dict[str, int] = pickle.load(f)
                 self.offset_maps[passage_file] = offset_map
                 self.passage_files[passage_file] = passage_file
-
-                # Build global map for O(1) lookup
-                for passage_id, offset in offset_map.items():
-                    self.global_offset_map[passage_id] = (passage_file, offset)
+                self._total_count += len(offset_map)
 
     def get_passage(self, passage_id: str) -> dict[str, Any]:
-        if passage_id in self.global_offset_map:
-            passage_file, offset = self.global_offset_map[passage_id]
-            # Lazy file opening - only open when needed
-            with open(passage_file, encoding="utf-8") as f:
-                f.seek(offset)
-                return json.loads(f.readline())
+        # Fast path: check each shard map (there are typically few shards).
+        # This avoids building a massive combined dict while keeping lookups
+        # bounded by the number of shards.
+        for passage_file, offset_map in self.offset_maps.items():
+            try:
+                offset = offset_map[passage_id]
+                with open(passage_file, encoding="utf-8") as f:
+                    f.seek(offset)
+                    return json.loads(f.readline())
+            except KeyError:
+                continue
         raise KeyError(f"Passage ID not found: {passage_id}")
 
+    def filter_search_results(
+        self,
+        search_results: list[SearchResult],
+        metadata_filters: Optional[dict[str, dict[str, Union[str, int, float, bool, list]]]],
+    ) -> list[SearchResult]:
+        """
+        Apply metadata filters to search results.
+
+        Args:
+            search_results: List of SearchResult objects
+            metadata_filters: Filter specifications to apply
+
+        Returns:
+            Filtered list of SearchResult objects
+        """
+        if not metadata_filters:
+            return search_results
+
+        logger.debug(f"Applying metadata filters to {len(search_results)} results")
+
+        # Convert SearchResult objects to dictionaries for the filter engine
+        result_dicts = []
+        for result in search_results:
+            result_dicts.append(
+                {
+                    "id": result.id,
+                    "score": result.score,
+                    "text": result.text,
+                    "metadata": result.metadata,
+                }
+            )
+
+        # Apply filters using the filter engine
+        filtered_dicts = self.filter_engine.apply_filters(result_dicts, metadata_filters)
+
+        # Convert back to SearchResult objects
+        filtered_results = []
+        for result_dict in filtered_dicts:
+            filtered_results.append(
+                SearchResult(
+                    id=result_dict["id"],
+                    score=result_dict["score"],
+                    text=result_dict["text"],
+                    metadata=result_dict["metadata"],
+                )
+            )
+
+        logger.debug(f"Filtered results: {len(filtered_results)} remaining")
+        return filtered_results
+
+    def __len__(self) -> int:
+        return self._total_count
+
 
 class LeannBuilder:
     def __init__(
@@ -557,6 +630,8 @@ class LeannSearcher:
         self.passage_manager = PassageManager(
             self.meta_data.get("passage_sources", []), metadata_file_path=self.meta_path_str
         )
+        # Preserve backend name for conditional parameter forwarding
+        self.backend_name = backend_name
         backend_factory = BACKEND_REGISTRY.get(backend_name)
         if backend_factory is None:
             raise ValueError(f"Backend '{backend_name}' not found.")
@@ -576,15 +651,44 @@ class LeannSearcher:
         recompute_embeddings: bool = True,
         pruning_strategy: Literal["global", "local", "proportional"] = "global",
         expected_zmq_port: int = 5557,
+        metadata_filters: Optional[dict[str, dict[str, Union[str, int, float, bool, list]]]] = None,
+        batch_size: int = 0,
         **kwargs,
     ) -> list[SearchResult]:
+        """
+        Search for nearest neighbors with optional metadata filtering.
+
+        Args:
+            query: Text query to search for
+            top_k: Number of nearest neighbors to return
+            complexity: Search complexity/candidate list size, higher = more accurate but slower
+            beam_width: Number of parallel search paths/IO requests per iteration
+            prune_ratio: Ratio of neighbors to prune via approximate distance (0.0-1.0)
+            recompute_embeddings: Whether to fetch fresh embeddings from server vs use stored codes
+            pruning_strategy: Candidate selection strategy - "global" (default), "local", or "proportional"
+            expected_zmq_port: ZMQ port for embedding server communication
+            metadata_filters: Optional filters to apply to search results based on metadata.
+                Format: {"field_name": {"operator": value}}
+                Supported operators:
+                - Comparison: "==", "!=", "<", "<=", ">", ">="
+                - Membership: "in", "not_in"
+                - String: "contains", "starts_with", "ends_with"
+                Example: {"chapter": {"<=": 5}, "tags": {"in": ["fiction", "drama"]}}
+            **kwargs: Backend-specific parameters
+
+        Returns:
+            List of SearchResult objects with text, metadata, and similarity scores
+        """
         logger.info("🔍 LeannSearcher.search() called:")
         logger.info(f"  Query: '{query}'")
         logger.info(f"  Top_k: {top_k}")
+        logger.info(f"  Metadata filters: {metadata_filters}")
         logger.info(f"  Additional kwargs: {kwargs}")
 
         # Smart top_k detection and adjustment
-        total_docs = len(self.passage_manager.global_offset_map)
+        # Use PassageManager length (sum of shard sizes) to avoid
+        # depending on a massive combined map
+        total_docs = len(self.passage_manager)
         original_top_k = top_k
         if top_k > total_docs:
             top_k = total_docs
@@ -613,23 +717,33 @@ class LeannSearcher:
             use_server_if_available=recompute_embeddings,
             zmq_port=zmq_port,
         )
-        # logger.info(f"  Generated embedding shape: {query_embedding.shape}")
-        # time.time() - start_time
-        # logger.info(f"  Embedding time: {embedding_time} seconds")
+        logger.info(f"  Generated embedding shape: {query_embedding.shape}")
+        embedding_time = time.time() - start_time
+        logger.info(f"  Embedding time: {embedding_time} seconds")
 
         start_time = time.time()
+        backend_search_kwargs: dict[str, Any] = {
+            "complexity": complexity,
+            "beam_width": beam_width,
+            "prune_ratio": prune_ratio,
+            "recompute_embeddings": recompute_embeddings,
+            "pruning_strategy": pruning_strategy,
+            "zmq_port": zmq_port,
+        }
+        # Only HNSW supports batching; forward conditionally
+        if self.backend_name == "hnsw":
+            backend_search_kwargs["batch_size"] = batch_size
+
+        # Merge any extra kwargs last
+        backend_search_kwargs.update(kwargs)
+
         results = self.backend_impl.search(
             query_embedding,
             top_k,
-            complexity=complexity,
-            beam_width=beam_width,
-            prune_ratio=prune_ratio,
-            recompute_embeddings=recompute_embeddings,
-            pruning_strategy=pruning_strategy,
-            zmq_port=zmq_port,
-            **kwargs,
+            **backend_search_kwargs,
         )
-        # logger.info(f"  Search time: {search_time} seconds")
+        search_time = time.time() - start_time
+        logger.info(f"  Search time in search() LEANN searcher: {search_time} seconds")
         logger.info(f"  Backend returned: labels={len(results.get('labels', [[]])[0])} results")
 
         enriched_results = []
@@ -668,6 +782,13 @@ class LeannSearcher:
                         f"   {RED}✗{RESET} [{i + 1:2d}] ID: '{string_id}' -> {RED}ERROR: Passage not found!{RESET}"
                     )
 
+        # Apply metadata filters if specified
+        if metadata_filters:
+            logger.info(f"  🔍 Applying metadata filters: {metadata_filters}")
+            enriched_results = self.passage_manager.filter_search_results(
+                enriched_results, metadata_filters
+            )
+
         # Define color codes outside the loop for final message
         GREEN = "\033[92m"
         RESET = "\033[0m"
@@ -708,9 +829,15 @@ class LeannChat:
         index_path: str,
         llm_config: Optional[dict[str, Any]] = None,
         enable_warmup: bool = False,
+        searcher: Optional[LeannSearcher] = None,
         **kwargs,
     ):
-        self.searcher = LeannSearcher(index_path, enable_warmup=enable_warmup, **kwargs)
+        if searcher is None:
+            self.searcher = LeannSearcher(index_path, enable_warmup=enable_warmup, **kwargs)
+            self._owns_searcher = True
+        else:
+            self.searcher = searcher
+            self._owns_searcher = False
         self.llm = get_llm(llm_config)
 
     def ask(
@@ -724,6 +851,8 @@ class LeannChat:
         pruning_strategy: Literal["global", "local", "proportional"] = "global",
         llm_kwargs: Optional[dict[str, Any]] = None,
         expected_zmq_port: int = 5557,
+        metadata_filters: Optional[dict[str, dict[str, Union[str, int, float, bool, list]]]] = None,
+        batch_size: int = 0,
         **search_kwargs,
     ):
         if llm_kwargs is None:
@@ -738,10 +867,12 @@ class LeannChat:
             recompute_embeddings=recompute_embeddings,
             pruning_strategy=pruning_strategy,
             expected_zmq_port=expected_zmq_port,
+            metadata_filters=metadata_filters,
+            batch_size=batch_size,
             **search_kwargs,
         )
         search_time = time.time() - search_time
-        # logger.info(f"  Search time: {search_time} seconds")
+        logger.info(f"  Search time: {search_time} seconds")
         context = "\n\n".join([r.text for r in results])
         prompt = (
             "Here is some retrieved context that might help answer your question:\n\n"
@@ -777,7 +908,9 @@ class LeannChat:
         This method should be called after you're done using the chat interface,
         especially in test environments or batch processing scenarios.
         """
-        if hasattr(self.searcher, "cleanup"):
+        # Only stop the embedding server if this LeannChat instance created the searcher.
+        # When a shared searcher is passed in, avoid shutting down the server to enable reuse.
+        if getattr(self, "_owns_searcher", False) and hasattr(self.searcher, "cleanup"):
             self.searcher.cleanup()
 
     # Enable automatic cleanup patterns

packages/leann-core/src/leann/cli.py

@@ -1,7 +1,8 @@
 import argparse
 import asyncio
+import sys
 from pathlib import Path
-from typing import Optional, Union
+from typing import Any, Optional, Union
 
 from llama_index.core import SimpleDirectoryReader
 from llama_index.core.node_parser import SentenceSplitter
@@ -180,6 +181,29 @@ Examples:
             default=50,
             help="Code chunk overlap (default: 50)",
         )
+        build_parser.add_argument(
+            "--use-ast-chunking",
+            action="store_true",
+            help="Enable AST-aware chunking for code files (requires astchunk)",
+        )
+        build_parser.add_argument(
+            "--ast-chunk-size",
+            type=int,
+            default=768,
+            help="AST chunk size in characters (default: 768)",
+        )
+        build_parser.add_argument(
+            "--ast-chunk-overlap",
+            type=int,
+            default=96,
+            help="AST chunk overlap in characters (default: 96)",
+        )
+        build_parser.add_argument(
+            "--ast-fallback-traditional",
+            action="store_true",
+            default=True,
+            help="Fall back to traditional chunking if AST chunking fails (default: True)",
+        )
 
         # Search command
         search_parser = subparsers.add_parser("search", help="Search documents")
@@ -206,6 +230,11 @@ Examples:
             default="global",
             help="Pruning strategy (default: global)",
         )
+        search_parser.add_argument(
+            "--non-interactive",
+            action="store_true",
+            help="Non-interactive mode: automatically select index without prompting",
+        )
 
         # Ask command
         ask_parser = subparsers.add_parser("ask", help="Ask questions")
@@ -828,6 +857,7 @@ Examples:
         docs_paths: Union[str, list],
         custom_file_types: Union[str, None] = None,
         include_hidden: bool = False,
+        args: Optional[dict[str, Any]] = None,
     ):
         # Handle both single path (string) and multiple paths (list) for backward compatibility
         if isinstance(docs_paths, str):
@@ -1133,18 +1163,50 @@ Examples:
         }
 
         print("start chunking documents")
-        # Add progress bar for document chunking
-        for doc in tqdm(documents, desc="Chunking documents", unit="doc"):
-            # Check if this is a code file based on source path
-            source_path = doc.metadata.get("source", "")
-            is_code_file = any(source_path.endswith(ext) for ext in code_file_exts)
 
-            # Use appropriate parser based on file type
-            parser = self.code_parser if is_code_file else self.node_parser
-            nodes = parser.get_nodes_from_documents([doc])
+        # Check if AST chunking is requested
+        use_ast = getattr(args, "use_ast_chunking", False)
 
-            for node in nodes:
-                all_texts.append(node.get_content())
+        if use_ast:
+            print("🧠 Using AST-aware chunking for code files")
+            try:
+                # Import enhanced chunking utilities
+                # Add apps directory to path to import chunking utilities
+                apps_dir = Path(__file__).parent.parent.parent.parent.parent / "apps"
+                if apps_dir.exists():
+                    sys.path.insert(0, str(apps_dir))
+
+                from chunking import create_text_chunks
+
+                # Use enhanced chunking with AST support
+                all_texts = create_text_chunks(
+                    documents,
+                    chunk_size=self.node_parser.chunk_size,
+                    chunk_overlap=self.node_parser.chunk_overlap,
+                    use_ast_chunking=True,
+                    ast_chunk_size=getattr(args, "ast_chunk_size", 768),
+                    ast_chunk_overlap=getattr(args, "ast_chunk_overlap", 96),
+                    code_file_extensions=None,  # Use defaults
+                    ast_fallback_traditional=getattr(args, "ast_fallback_traditional", True),
+                )
+
+            except ImportError as e:
+                print(f"⚠️  AST chunking not available ({e}), falling back to traditional chunking")
+                use_ast = False
+
+        if not use_ast:
+            # Use traditional chunking logic
+            for doc in tqdm(documents, desc="Chunking documents", unit="doc"):
+                # Check if this is a code file based on source path
+                source_path = doc.metadata.get("source", "")
+                is_code_file = any(source_path.endswith(ext) for ext in code_file_exts)
+
+                # Use appropriate parser based on file type
+                parser = self.code_parser if is_code_file else self.node_parser
+                nodes = parser.get_nodes_from_documents([doc])
+
+                for node in nodes:
+                    all_texts.append(node.get_content())
 
         print(f"Loaded {len(documents)} documents, {len(all_texts)} chunks")
         return all_texts
@@ -1211,7 +1273,7 @@ Examples:
         )
 
         all_texts = self.load_documents(
-            docs_paths, args.file_types, include_hidden=args.include_hidden
+            docs_paths, args.file_types, include_hidden=args.include_hidden, args=args
         )
         if not all_texts:
             print("No documents found")
@@ -1244,13 +1306,101 @@ Examples:
     async def search_documents(self, args):
         index_name = args.index_name
         query = args.query
-        index_path = self.get_index_path(index_name)
 
-        if not self.index_exists(index_name):
-            print(
-                f"Index '{index_name}' not found. Use 'leann build {index_name} --docs <dir> [<dir2> ...]' to create it."
-            )
-            return
+        # First try to find the index in current project
+        index_path = self.get_index_path(index_name)
+        if self.index_exists(index_name):
+            # Found in current project, use it
+            pass
+        else:
+            # Search across all registered projects (like list_indexes does)
+            all_matches = self._find_all_matching_indexes(index_name)
+            if not all_matches:
+                print(
+                    f"Index '{index_name}' not found. Use 'leann build {index_name} --docs <dir> [<dir2> ...]' to create it."
+                )
+                return
+            elif len(all_matches) == 1:
+                # Found exactly one match, use it
+                match = all_matches[0]
+                if match["kind"] == "cli":
+                    index_path = str(match["index_dir"] / "documents.leann")
+                else:
+                    # App format: use the meta file to construct the path
+                    meta_file = match["meta_file"]
+                    file_base = match["file_base"]
+                    index_path = str(meta_file.parent / f"{file_base}.leann")
+
+                project_info = (
+                    "current project"
+                    if match["is_current"]
+                    else f"project '{match['project_path'].name}'"
+                )
+                print(f"Using index '{index_name}' from {project_info}")
+            else:
+                # Multiple matches found
+                if args.non_interactive:
+                    # Non-interactive mode: automatically select the best match
+                    # Priority: current project first, then first available
+                    current_matches = [m for m in all_matches if m["is_current"]]
+                    if current_matches:
+                        match = current_matches[0]
+                        location_desc = "current project"
+                    else:
+                        match = all_matches[0]
+                        location_desc = f"project '{match['project_path'].name}'"
+
+                    if match["kind"] == "cli":
+                        index_path = str(match["index_dir"] / "documents.leann")
+                    else:
+                        meta_file = match["meta_file"]
+                        file_base = match["file_base"]
+                        index_path = str(meta_file.parent / f"{file_base}.leann")
+
+                    print(
+                        f"Found {len(all_matches)} indexes named '{index_name}', using index from {location_desc}"
+                    )
+                else:
+                    # Interactive mode: ask user to choose
+                    print(f"Found {len(all_matches)} indexes named '{index_name}':")
+                    for i, match in enumerate(all_matches, 1):
+                        project_path = match["project_path"]
+                        is_current = match["is_current"]
+                        kind = match.get("kind", "cli")
+
+                        if is_current:
+                            print(
+                                f"   {i}. 🏠 Current project ({'CLI' if kind == 'cli' else 'APP'})"
+                            )
+                        else:
+                            print(
+                                f"   {i}. 📂 {project_path.name} ({'CLI' if kind == 'cli' else 'APP'})"
+                            )
+
+                    try:
+                        choice = input(f"Which index to search? (1-{len(all_matches)}): ").strip()
+                        choice_idx = int(choice) - 1
+                        if 0 <= choice_idx < len(all_matches):
+                            match = all_matches[choice_idx]
+                            if match["kind"] == "cli":
+                                index_path = str(match["index_dir"] / "documents.leann")
+                            else:
+                                meta_file = match["meta_file"]
+                                file_base = match["file_base"]
+                                index_path = str(meta_file.parent / f"{file_base}.leann")
+
+                            project_info = (
+                                "current project"
+                                if match["is_current"]
+                                else f"project '{match['project_path'].name}'"
+                            )
+                            print(f"Using index '{index_name}' from {project_info}")
+                        else:
+                            print("Invalid choice. Aborting search.")
+                            return
+                    except (ValueError, KeyboardInterrupt):
+                        print("Invalid input. Aborting search.")
+                        return
 
         searcher = LeannSearcher(index_path=index_path)
         results = searcher.search(

packages/leann-core/src/leann/embedding_compute.py

@@ -6,6 +6,7 @@ Preserves all optimization parameters to ensure performance
 
 import logging
 import os
+import time
 from typing import Any
 
 import numpy as np
@@ -28,6 +29,8 @@ def compute_embeddings(
     is_build: bool = False,
     batch_size: int = 32,
     adaptive_optimization: bool = True,
+    manual_tokenize: bool = False,
+    max_length: int = 512,
 ) -> np.ndarray:
     """
     Unified embedding computation entry point
@@ -50,6 +53,8 @@ def compute_embeddings(
             is_build=is_build,
             batch_size=batch_size,
             adaptive_optimization=adaptive_optimization,
+            manual_tokenize=manual_tokenize,
+            max_length=max_length,
         )
     elif mode == "openai":
         return compute_embeddings_openai(texts, model_name)
@@ -71,6 +76,8 @@ def compute_embeddings_sentence_transformers(
     batch_size: int = 32,
     is_build: bool = False,
     adaptive_optimization: bool = True,
+    manual_tokenize: bool = False,
+    max_length: int = 512,
 ) -> np.ndarray:
     """
     Compute embeddings using SentenceTransformer with model caching and adaptive optimization
@@ -214,20 +221,130 @@ def compute_embeddings_sentence_transformers(
         logger.info(f"Model cached: {cache_key}")
 
     # Compute embeddings with optimized inference mode
-    logger.info(f"Starting embedding computation... (batch_size: {batch_size})")
+    logger.info(
+        f"Starting embedding computation... (batch_size: {batch_size}, manual_tokenize={manual_tokenize})"
+    )
 
-    # Use torch.inference_mode for optimal performance
-    with torch.inference_mode():
-        embeddings = model.encode(
-            texts,
-            batch_size=batch_size,
-            show_progress_bar=is_build,  # Don't show progress bar in server environment
-            convert_to_numpy=True,
-            normalize_embeddings=False,
-            device=device,
-        )
+    start_time = time.time()
+    if not manual_tokenize:
+        # Use SentenceTransformer's optimized encode path (default)
+        with torch.inference_mode():
+            embeddings = model.encode(
+                texts,
+                batch_size=batch_size,
+                show_progress_bar=is_build,  # Don't show progress bar in server environment
+                convert_to_numpy=True,
+                normalize_embeddings=False,
+                device=device,
+            )
+        # Synchronize if CUDA to measure accurate wall time
+        try:
+            if torch.cuda.is_available():
+                torch.cuda.synchronize()
+        except Exception:
+            pass
+    else:
+        # Manual tokenization + forward pass using HF AutoTokenizer/AutoModel
+        try:
+            from transformers import AutoModel, AutoTokenizer  # type: ignore
+        except Exception as e:
+            raise ImportError(f"transformers is required for manual_tokenize=True: {e}")
 
+        # Cache tokenizer and model
+        tok_cache_key = f"hf_tokenizer_{model_name}"
+        mdl_cache_key = f"hf_model_{model_name}_{device}_{use_fp16}"
+        if tok_cache_key in _model_cache and mdl_cache_key in _model_cache:
+            hf_tokenizer = _model_cache[tok_cache_key]
+            hf_model = _model_cache[mdl_cache_key]
+            logger.info("Using cached HF tokenizer/model for manual path")
+        else:
+            logger.info("Loading HF tokenizer/model for manual tokenization path")
+            hf_tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=True)
+            torch_dtype = torch.float16 if (use_fp16 and device == "cuda") else torch.float32
+            hf_model = AutoModel.from_pretrained(model_name, torch_dtype=torch_dtype)
+            hf_model.to(device)
+            hf_model.eval()
+            # Optional compile on supported devices
+            if device in ["cuda", "mps"]:
+                try:
+                    hf_model = torch.compile(hf_model, mode="reduce-overhead", dynamic=True)  # type: ignore
+                except Exception:
+                    pass
+            _model_cache[tok_cache_key] = hf_tokenizer
+            _model_cache[mdl_cache_key] = hf_model
+
+        all_embeddings: list[np.ndarray] = []
+        # Progress bar when building or for large inputs
+        show_progress = is_build or len(texts) > 32
+        try:
+            if show_progress:
+                from tqdm import tqdm  # type: ignore
+
+                batch_iter = tqdm(
+                    range(0, len(texts), batch_size),
+                    desc="Embedding (manual)",
+                    unit="batch",
+                )
+            else:
+                batch_iter = range(0, len(texts), batch_size)
+        except Exception:
+            batch_iter = range(0, len(texts), batch_size)
+
+        start_time_manual = time.time()
+        with torch.inference_mode():
+            for start_index in batch_iter:
+                end_index = min(start_index + batch_size, len(texts))
+                batch_texts = texts[start_index:end_index]
+                tokenize_start_time = time.time()
+                inputs = hf_tokenizer(
+                    batch_texts,
+                    padding=True,
+                    truncation=True,
+                    max_length=max_length,
+                    return_tensors="pt",
+                )
+                tokenize_end_time = time.time()
+                logger.info(
+                    f"Tokenize time taken: {tokenize_end_time - tokenize_start_time} seconds"
+                )
+                # Print shapes of all input tensors for debugging
+                for k, v in inputs.items():
+                    print(f"inputs[{k!r}] shape: {getattr(v, 'shape', type(v))}")
+                to_device_start_time = time.time()
+                inputs = {k: v.to(device) for k, v in inputs.items()}
+                to_device_end_time = time.time()
+                logger.info(
+                    f"To device time taken: {to_device_end_time - to_device_start_time} seconds"
+                )
+                forward_start_time = time.time()
+                outputs = hf_model(**inputs)
+                forward_end_time = time.time()
+                logger.info(f"Forward time taken: {forward_end_time - forward_start_time} seconds")
+                last_hidden_state = outputs.last_hidden_state  # (B, L, H)
+                attention_mask = inputs.get("attention_mask")
+                if attention_mask is None:
+                    # Fallback: assume all tokens are valid
+                    pooled = last_hidden_state.mean(dim=1)
+                else:
+                    mask = attention_mask.unsqueeze(-1).to(last_hidden_state.dtype)
+                    masked = last_hidden_state * mask
+                    lengths = mask.sum(dim=1).clamp(min=1)
+                    pooled = masked.sum(dim=1) / lengths
+                # Move to CPU float32
+                batch_embeddings = pooled.detach().to("cpu").float().numpy()
+                all_embeddings.append(batch_embeddings)
+
+        embeddings = np.vstack(all_embeddings).astype(np.float32, copy=False)
+        try:
+            if torch.cuda.is_available():
+                torch.cuda.synchronize()
+        except Exception:
+            pass
+        end_time = time.time()
+        logger.info(f"Manual tokenize time taken: {end_time - start_time_manual} seconds")
+    end_time = time.time()
     logger.info(f"Generated {len(embeddings)} embeddings, dimension: {embeddings.shape[1]}")
+    logger.info(f"Time taken: {end_time - start_time} seconds")
 
     # Validate results
     if np.isnan(embeddings).any() or np.isinf(embeddings).any():

packages/leann-core/src/leann/embedding_server_manager.py

@@ -192,6 +192,7 @@ class EmbeddingServerManager:
             stderr_target = None  # Direct to console for visible logs
 
         # Start embedding server subprocess
+        logger.info(f"Starting server process with command: {' '.join(command)}")
         self.server_process = subprocess.Popen(
             command,
             cwd=project_root,

packages/leann-core/src/leann/mcp.py

@@ -94,7 +94,7 @@ def handle_request(request):
                         },
                     }
 
-                # Build simplified command
+                # Build simplified command with non-interactive flag for MCP compatibility
                 cmd = [
                     "leann",
                     "search",
@@ -102,6 +102,7 @@ def handle_request(request):
                     args["query"],
                     f"--top-k={args.get('top_k', 5)}",
                     f"--complexity={args.get('complexity', 32)}",
+                    "--non-interactive",
                 ]
                 result = subprocess.run(cmd, capture_output=True, text=True)
 

packages/leann-core/src/leann/metadata_filter.py

@@ -0,0 +1,240 @@
+"""
+Metadata filtering engine for LEANN search results.
+
+This module provides generic metadata filtering capabilities that can be applied
+to search results from any LEANN backend. The filtering supports various
+operators for different data types including numbers, strings, booleans, and lists.
+"""
+
+import logging
+from typing import Any, Union
+
+logger = logging.getLogger(__name__)
+
+# Type alias for filter specifications
+FilterValue = Union[str, int, float, bool, list]
+FilterSpec = dict[str, FilterValue]
+MetadataFilters = dict[str, FilterSpec]
+
+
+class MetadataFilterEngine:
+    """
+    Engine for evaluating metadata filters against search results.
+
+    Supports various operators for filtering based on metadata fields:
+    - Comparison: ==, !=, <, <=, >, >=
+    - Membership: in, not_in
+    - String operations: contains, starts_with, ends_with
+    - Boolean operations: is_true, is_false
+    """
+
+    def __init__(self):
+        """Initialize the filter engine with supported operators."""
+        self.operators = {
+            "==": self._equals,
+            "!=": self._not_equals,
+            "<": self._less_than,
+            "<=": self._less_than_or_equal,
+            ">": self._greater_than,
+            ">=": self._greater_than_or_equal,
+            "in": self._in,
+            "not_in": self._not_in,
+            "contains": self._contains,
+            "starts_with": self._starts_with,
+            "ends_with": self._ends_with,
+            "is_true": self._is_true,
+            "is_false": self._is_false,
+        }
+
+    def apply_filters(
+        self, search_results: list[dict[str, Any]], metadata_filters: MetadataFilters
+    ) -> list[dict[str, Any]]:
+        """
+        Apply metadata filters to a list of search results.
+
+        Args:
+            search_results: List of result dictionaries, each containing 'metadata' field
+            metadata_filters: Dictionary of filter specifications
+                Format: {"field_name": {"operator": value}}
+
+        Returns:
+            Filtered list of search results
+        """
+        if not metadata_filters:
+            return search_results
+
+        logger.debug(f"Applying filters: {metadata_filters}")
+        logger.debug(f"Input results count: {len(search_results)}")
+
+        filtered_results = []
+        for result in search_results:
+            if self._evaluate_filters(result, metadata_filters):
+                filtered_results.append(result)
+
+        logger.debug(f"Filtered results count: {len(filtered_results)}")
+        return filtered_results
+
+    def _evaluate_filters(self, result: dict[str, Any], filters: MetadataFilters) -> bool:
+        """
+        Evaluate all filters against a single search result.
+
+        All filters must pass (AND logic) for the result to be included.
+
+        Args:
+            result: Full search result dictionary (including metadata, text, etc.)
+            filters: Filter specifications to evaluate
+
+        Returns:
+            True if all filters pass, False otherwise
+        """
+        for field_name, filter_spec in filters.items():
+            if not self._evaluate_field_filter(result, field_name, filter_spec):
+                return False
+        return True
+
+    def _evaluate_field_filter(
+        self, result: dict[str, Any], field_name: str, filter_spec: FilterSpec
+    ) -> bool:
+        """
+        Evaluate a single field filter against a search result.
+
+        Args:
+            result: Full search result dictionary
+            field_name: Name of the field to filter on
+            filter_spec: Filter specification for this field
+
+        Returns:
+            True if the filter passes, False otherwise
+        """
+        # First check top-level fields, then check metadata
+        field_value = result.get(field_name)
+        if field_value is None:
+            # Try to get from metadata if not found at top level
+            metadata = result.get("metadata", {})
+            field_value = metadata.get(field_name)
+
+        # Handle missing fields - they fail all filters except existence checks
+        if field_value is None:
+            logger.debug(f"Field '{field_name}' not found in result or metadata")
+            return False
+
+        # Evaluate each operator in the filter spec
+        for operator, expected_value in filter_spec.items():
+            if operator not in self.operators:
+                logger.warning(f"Unsupported operator: {operator}")
+                return False
+
+            try:
+                if not self.operators[operator](field_value, expected_value):
+                    logger.debug(
+                        f"Filter failed: {field_name} {operator} {expected_value} "
+                        f"(actual: {field_value})"
+                    )
+                    return False
+            except Exception as e:
+                logger.warning(
+                    f"Error evaluating filter {field_name} {operator} {expected_value}: {e}"
+                )
+                return False
+
+        return True
+
+    # Comparison operators
+    def _equals(self, field_value: Any, expected_value: Any) -> bool:
+        """Check if field value equals expected value."""
+        return field_value == expected_value
+
+    def _not_equals(self, field_value: Any, expected_value: Any) -> bool:
+        """Check if field value does not equal expected value."""
+        return field_value != expected_value
+
+    def _less_than(self, field_value: Any, expected_value: Any) -> bool:
+        """Check if field value is less than expected value."""
+        return self._numeric_compare(field_value, expected_value, lambda a, b: a < b)
+
+    def _less_than_or_equal(self, field_value: Any, expected_value: Any) -> bool:
+        """Check if field value is less than or equal to expected value."""
+        return self._numeric_compare(field_value, expected_value, lambda a, b: a <= b)
+
+    def _greater_than(self, field_value: Any, expected_value: Any) -> bool:
+        """Check if field value is greater than expected value."""
+        return self._numeric_compare(field_value, expected_value, lambda a, b: a > b)
+
+    def _greater_than_or_equal(self, field_value: Any, expected_value: Any) -> bool:
+        """Check if field value is greater than or equal to expected value."""
+        return self._numeric_compare(field_value, expected_value, lambda a, b: a >= b)
+
+    # Membership operators
+    def _in(self, field_value: Any, expected_value: Any) -> bool:
+        """Check if field value is in the expected list/collection."""
+        if not isinstance(expected_value, (list, tuple, set)):
+            raise ValueError("'in' operator requires a list, tuple, or set")
+        return field_value in expected_value
+
+    def _not_in(self, field_value: Any, expected_value: Any) -> bool:
+        """Check if field value is not in the expected list/collection."""
+        if not isinstance(expected_value, (list, tuple, set)):
+            raise ValueError("'not_in' operator requires a list, tuple, or set")
+        return field_value not in expected_value
+
+    # String operators
+    def _contains(self, field_value: Any, expected_value: Any) -> bool:
+        """Check if field value contains the expected substring."""
+        field_str = str(field_value)
+        expected_str = str(expected_value)
+        return expected_str in field_str
+
+    def _starts_with(self, field_value: Any, expected_value: Any) -> bool:
+        """Check if field value starts with the expected prefix."""
+        field_str = str(field_value)
+        expected_str = str(expected_value)
+        return field_str.startswith(expected_str)
+
+    def _ends_with(self, field_value: Any, expected_value: Any) -> bool:
+        """Check if field value ends with the expected suffix."""
+        field_str = str(field_value)
+        expected_str = str(expected_value)
+        return field_str.endswith(expected_str)
+
+    # Boolean operators
+    def _is_true(self, field_value: Any, expected_value: Any) -> bool:
+        """Check if field value is truthy."""
+        return bool(field_value)
+
+    def _is_false(self, field_value: Any, expected_value: Any) -> bool:
+        """Check if field value is falsy."""
+        return not bool(field_value)
+
+    # Helper methods
+    def _numeric_compare(self, field_value: Any, expected_value: Any, compare_func) -> bool:
+        """
+        Helper for numeric comparisons with type coercion.
+
+        Args:
+            field_value: Value from metadata
+            expected_value: Value to compare against
+            compare_func: Comparison function to apply
+
+        Returns:
+            Result of comparison
+        """
+        try:
+            # Try to convert both values to numbers for comparison
+            if isinstance(field_value, str) and isinstance(expected_value, str):
+                # String comparison if both are strings
+                return compare_func(field_value, expected_value)
+
+            # Numeric comparison - attempt to convert to float
+            field_num = (
+                float(field_value) if not isinstance(field_value, (int, float)) else field_value
+            )
+            expected_num = (
+                float(expected_value)
+                if not isinstance(expected_value, (int, float))
+                else expected_value
+            )
+
+            return compare_func(field_num, expected_num)
+        except (ValueError, TypeError):
+            # Fall back to string comparison if numeric conversion fails
+            return compare_func(str(field_value), str(expected_value))

packages/leann/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "leann"
-version = "0.3.1"
+version = "0.3.2"
 description = "LEANN - The smallest vector index in the world. RAG Everything with LEANN!"
 readme = "README.md"
 requires-python = ">=3.9"

pyproject.toml

@@ -46,6 +46,13 @@ dependencies = [
     "pathspec>=0.12.1",
     "nbconvert>=7.16.6",
     "gitignore-parser>=0.1.12",
+    # AST-aware code chunking dependencies
+    "astchunk>=0.1.0",
+    "tree-sitter>=0.20.0",
+    "tree-sitter-python>=0.20.0",
+    "tree-sitter-java>=0.20.0",
+    "tree-sitter-c-sharp>=0.20.0",
+    "tree-sitter-typescript>=0.20.0",
 ]
 
 [project.optional-dependencies]

tests/test_astchunk_integration.py

@@ -0,0 +1,397 @@
+"""
+Test suite for astchunk integration with LEANN.
+Tests AST-aware chunking functionality, language detection, and fallback mechanisms.
+"""
+
+import os
+import subprocess
+import sys
+import tempfile
+from pathlib import Path
+from unittest.mock import patch
+
+import pytest
+
+# Add apps directory to path for imports
+sys.path.insert(0, str(Path(__file__).parent.parent / "apps"))
+
+from typing import Optional
+
+from chunking import (
+    create_ast_chunks,
+    create_text_chunks,
+    create_traditional_chunks,
+    detect_code_files,
+    get_language_from_extension,
+)
+
+
+class MockDocument:
+    """Mock LlamaIndex Document for testing."""
+
+    def __init__(self, content: str, file_path: str = "", metadata: Optional[dict] = None):
+        self.content = content
+        self.metadata = metadata or {}
+        if file_path:
+            self.metadata["file_path"] = file_path
+
+    def get_content(self) -> str:
+        return self.content
+
+
+class TestCodeFileDetection:
+    """Test code file detection and language mapping."""
+
+    def test_detect_code_files_python(self):
+        """Test detection of Python files."""
+        docs = [
+            MockDocument("print('hello')", "/path/to/file.py"),
+            MockDocument("This is text", "/path/to/file.txt"),
+        ]
+
+        code_docs, text_docs = detect_code_files(docs)
+
+        assert len(code_docs) == 1
+        assert len(text_docs) == 1
+        assert code_docs[0].metadata["language"] == "python"
+        assert code_docs[0].metadata["is_code"] is True
+        assert text_docs[0].metadata["is_code"] is False
+
+    def test_detect_code_files_multiple_languages(self):
+        """Test detection of multiple programming languages."""
+        docs = [
+            MockDocument("def func():", "/path/to/script.py"),
+            MockDocument("public class Test {}", "/path/to/Test.java"),
+            MockDocument("interface ITest {}", "/path/to/test.ts"),
+            MockDocument("using System;", "/path/to/Program.cs"),
+            MockDocument("Regular text content", "/path/to/document.txt"),
+        ]
+
+        code_docs, text_docs = detect_code_files(docs)
+
+        assert len(code_docs) == 4
+        assert len(text_docs) == 1
+
+        languages = [doc.metadata["language"] for doc in code_docs]
+        assert "python" in languages
+        assert "java" in languages
+        assert "typescript" in languages
+        assert "csharp" in languages
+
+    def test_detect_code_files_no_file_path(self):
+        """Test handling of documents without file paths."""
+        docs = [
+            MockDocument("some content"),
+            MockDocument("other content", metadata={"some_key": "value"}),
+        ]
+
+        code_docs, text_docs = detect_code_files(docs)
+
+        assert len(code_docs) == 0
+        assert len(text_docs) == 2
+        for doc in text_docs:
+            assert doc.metadata["is_code"] is False
+
+    def test_get_language_from_extension(self):
+        """Test language detection from file extensions."""
+        assert get_language_from_extension("test.py") == "python"
+        assert get_language_from_extension("Test.java") == "java"
+        assert get_language_from_extension("component.tsx") == "typescript"
+        assert get_language_from_extension("Program.cs") == "csharp"
+        assert get_language_from_extension("document.txt") is None
+        assert get_language_from_extension("") is None
+
+
+class TestChunkingFunctions:
+    """Test various chunking functionality."""
+
+    def test_create_traditional_chunks(self):
+        """Test traditional text chunking."""
+        docs = [
+            MockDocument(
+                "This is a test document. It has multiple sentences. We want to test chunking."
+            )
+        ]
+
+        chunks = create_traditional_chunks(docs, chunk_size=50, chunk_overlap=10)
+
+        assert len(chunks) > 0
+        assert all(isinstance(chunk, str) for chunk in chunks)
+        assert all(len(chunk.strip()) > 0 for chunk in chunks)
+
+    def test_create_traditional_chunks_empty_docs(self):
+        """Test traditional chunking with empty documents."""
+        chunks = create_traditional_chunks([], chunk_size=50, chunk_overlap=10)
+        assert chunks == []
+
+    @pytest.mark.skipif(
+        os.environ.get("CI") == "true",
+        reason="Skip astchunk tests in CI - dependency may not be available",
+    )
+    def test_create_ast_chunks_with_astchunk_available(self):
+        """Test AST chunking when astchunk is available."""
+        python_code = '''
+def hello_world():
+    """Print hello world message."""
+    print("Hello, World!")
+
+def add_numbers(a, b):
+    """Add two numbers and return the result."""
+    return a + b
+
+class Calculator:
+    """A simple calculator class."""
+
+    def __init__(self):
+        self.history = []
+
+    def add(self, a, b):
+        result = a + b
+        self.history.append(f"{a} + {b} = {result}")
+        return result
+'''
+
+        docs = [MockDocument(python_code, "/test/calculator.py", {"language": "python"})]
+
+        try:
+            chunks = create_ast_chunks(docs, max_chunk_size=200, chunk_overlap=50)
+
+            # Should have multiple chunks due to different functions/classes
+            assert len(chunks) > 0
+            assert all(isinstance(chunk, str) for chunk in chunks)
+            assert all(len(chunk.strip()) > 0 for chunk in chunks)
+
+            # Check that code structure is somewhat preserved
+            combined_content = " ".join(chunks)
+            assert "def hello_world" in combined_content
+            assert "class Calculator" in combined_content
+
+        except ImportError:
+            # astchunk not available, should fall back to traditional chunking
+            chunks = create_ast_chunks(docs, max_chunk_size=200, chunk_overlap=50)
+            assert len(chunks) > 0  # Should still get chunks from fallback
+
+    def test_create_ast_chunks_fallback_to_traditional(self):
+        """Test AST chunking falls back to traditional when astchunk is not available."""
+        docs = [MockDocument("def test(): pass", "/test/script.py", {"language": "python"})]
+
+        # Mock astchunk import to fail
+        with patch("chunking.create_ast_chunks"):
+            # First call (actual test) should import astchunk and potentially fail
+            # Let's call the actual function to test the import error handling
+            chunks = create_ast_chunks(docs)
+
+            # Should return some chunks (either from astchunk or fallback)
+            assert isinstance(chunks, list)
+
+    def test_create_text_chunks_traditional_mode(self):
+        """Test text chunking in traditional mode."""
+        docs = [
+            MockDocument("def test(): pass", "/test/script.py"),
+            MockDocument("This is regular text.", "/test/doc.txt"),
+        ]
+
+        chunks = create_text_chunks(docs, use_ast_chunking=False, chunk_size=50, chunk_overlap=10)
+
+        assert len(chunks) > 0
+        assert all(isinstance(chunk, str) for chunk in chunks)
+
+    def test_create_text_chunks_ast_mode(self):
+        """Test text chunking in AST mode."""
+        docs = [
+            MockDocument("def test(): pass", "/test/script.py"),
+            MockDocument("This is regular text.", "/test/doc.txt"),
+        ]
+
+        chunks = create_text_chunks(
+            docs,
+            use_ast_chunking=True,
+            ast_chunk_size=100,
+            ast_chunk_overlap=20,
+            chunk_size=50,
+            chunk_overlap=10,
+        )
+
+        assert len(chunks) > 0
+        assert all(isinstance(chunk, str) for chunk in chunks)
+
+    def test_create_text_chunks_custom_extensions(self):
+        """Test text chunking with custom code file extensions."""
+        docs = [
+            MockDocument("function test() {}", "/test/script.js"),  # Not in default extensions
+            MockDocument("Regular text", "/test/doc.txt"),
+        ]
+
+        # First without custom extensions - should treat .js as text
+        chunks_without = create_text_chunks(docs, use_ast_chunking=True, code_file_extensions=None)
+
+        # Then with custom extensions - should treat .js as code
+        chunks_with = create_text_chunks(
+            docs, use_ast_chunking=True, code_file_extensions=[".js", ".jsx"]
+        )
+
+        # Both should return chunks
+        assert len(chunks_without) > 0
+        assert len(chunks_with) > 0
+
+
+class TestIntegrationWithDocumentRAG:
+    """Integration tests with the document RAG system."""
+
+    @pytest.fixture
+    def temp_code_dir(self):
+        """Create a temporary directory with sample code files."""
+        with tempfile.TemporaryDirectory() as temp_dir:
+            temp_path = Path(temp_dir)
+
+            # Create sample Python file
+            python_file = temp_path / "example.py"
+            python_file.write_text('''
+def fibonacci(n):
+    """Calculate fibonacci number."""
+    if n <= 1:
+        return n
+    return fibonacci(n-1) + fibonacci(n-2)
+
+class MathUtils:
+    @staticmethod
+    def factorial(n):
+        if n <= 1:
+            return 1
+        return n * MathUtils.factorial(n-1)
+''')
+
+            # Create sample text file
+            text_file = temp_path / "readme.txt"
+            text_file.write_text("This is a sample text file for testing purposes.")
+
+            yield temp_path
+
+    @pytest.mark.skipif(
+        os.environ.get("CI") == "true",
+        reason="Skip integration tests in CI to avoid dependency issues",
+    )
+    def test_document_rag_with_ast_chunking(self, temp_code_dir):
+        """Test document RAG with AST chunking enabled."""
+        with tempfile.TemporaryDirectory() as index_dir:
+            cmd = [
+                sys.executable,
+                "apps/document_rag.py",
+                "--llm",
+                "simulated",
+                "--embedding-model",
+                "facebook/contriever",
+                "--embedding-mode",
+                "sentence-transformers",
+                "--index-dir",
+                index_dir,
+                "--data-dir",
+                str(temp_code_dir),
+                "--enable-code-chunking",
+                "--query",
+                "How does the fibonacci function work?",
+            ]
+
+            env = os.environ.copy()
+            env["HF_HUB_DISABLE_SYMLINKS"] = "1"
+            env["TOKENIZERS_PARALLELISM"] = "false"
+
+            try:
+                result = subprocess.run(
+                    cmd,
+                    capture_output=True,
+                    text=True,
+                    timeout=300,  # 5 minutes
+                    env=env,
+                )
+
+                # Should succeed even if astchunk is not available (fallback)
+                assert result.returncode == 0, f"Command failed: {result.stderr}"
+
+                output = result.stdout + result.stderr
+                assert "Index saved to" in output or "Using existing index" in output
+
+            except subprocess.TimeoutExpired:
+                pytest.skip("Test timed out - likely due to model download in CI")
+
+    @pytest.mark.skipif(
+        os.environ.get("CI") == "true",
+        reason="Skip integration tests in CI to avoid dependency issues",
+    )
+    def test_code_rag_application(self, temp_code_dir):
+        """Test the specialized code RAG application."""
+        with tempfile.TemporaryDirectory() as index_dir:
+            cmd = [
+                sys.executable,
+                "apps/code_rag.py",
+                "--llm",
+                "simulated",
+                "--embedding-model",
+                "facebook/contriever",
+                "--index-dir",
+                index_dir,
+                "--repo-dir",
+                str(temp_code_dir),
+                "--query",
+                "What classes are defined in this code?",
+            ]
+
+            env = os.environ.copy()
+            env["HF_HUB_DISABLE_SYMLINKS"] = "1"
+            env["TOKENIZERS_PARALLELISM"] = "false"
+
+            try:
+                result = subprocess.run(cmd, capture_output=True, text=True, timeout=300, env=env)
+
+                # Should succeed
+                assert result.returncode == 0, f"Command failed: {result.stderr}"
+
+                output = result.stdout + result.stderr
+                assert "Using AST-aware chunking" in output or "traditional chunking" in output
+
+            except subprocess.TimeoutExpired:
+                pytest.skip("Test timed out - likely due to model download in CI")
+
+
+class TestErrorHandling:
+    """Test error handling and edge cases."""
+
+    def test_text_chunking_empty_documents(self):
+        """Test text chunking with empty document list."""
+        chunks = create_text_chunks([])
+        assert chunks == []
+
+    def test_text_chunking_invalid_parameters(self):
+        """Test text chunking with invalid parameters."""
+        docs = [MockDocument("test content")]
+
+        # Should handle negative chunk sizes gracefully
+        chunks = create_text_chunks(
+            docs, chunk_size=0, chunk_overlap=0, ast_chunk_size=0, ast_chunk_overlap=0
+        )
+
+        # Should still return some result
+        assert isinstance(chunks, list)
+
+    def test_create_ast_chunks_no_language(self):
+        """Test AST chunking with documents missing language metadata."""
+        docs = [MockDocument("def test(): pass", "/test/script.py")]  # No language set
+
+        chunks = create_ast_chunks(docs)
+
+        # Should fall back to traditional chunking
+        assert isinstance(chunks, list)
+        assert len(chunks) >= 0  # May be empty if fallback also fails
+
+    def test_create_ast_chunks_empty_content(self):
+        """Test AST chunking with empty content."""
+        docs = [MockDocument("", "/test/script.py", {"language": "python"})]
+
+        chunks = create_ast_chunks(docs)
+
+        # Should handle empty content gracefully
+        assert isinstance(chunks, list)
+
+
+if __name__ == "__main__":
+    pytest.main([__file__, "-v"])

tests/test_document_rag.py

@@ -57,6 +57,51 @@ def test_document_rag_simulated(test_data_dir):
         assert "This is a simulated answer" in output
 
 
+@pytest.mark.skipif(
+    os.environ.get("CI") == "true",
+    reason="Skip AST chunking tests in CI to avoid dependency issues",
+)
+def test_document_rag_with_ast_chunking(test_data_dir):
+    """Test document_rag with AST-aware chunking enabled."""
+    with tempfile.TemporaryDirectory() as temp_dir:
+        # Use a subdirectory that doesn't exist yet to force index creation
+        index_dir = Path(temp_dir) / "test_ast_index"
+        cmd = [
+            sys.executable,
+            "apps/document_rag.py",
+            "--llm",
+            "simulated",
+            "--embedding-model",
+            "facebook/contriever",
+            "--embedding-mode",
+            "sentence-transformers",
+            "--index-dir",
+            str(index_dir),
+            "--data-dir",
+            str(test_data_dir),
+            "--enable-code-chunking",  # Enable AST chunking
+            "--query",
+            "What is Pride and Prejudice about?",
+        ]
+
+        env = os.environ.copy()
+        env["HF_HUB_DISABLE_SYMLINKS"] = "1"
+        env["TOKENIZERS_PARALLELISM"] = "false"
+
+        result = subprocess.run(cmd, capture_output=True, text=True, timeout=600, env=env)
+
+        # Check return code
+        assert result.returncode == 0, f"Command failed: {result.stderr}"
+
+        # Verify output
+        output = result.stdout + result.stderr
+        assert "Index saved to" in output or "Using existing index" in output
+        assert "This is a simulated answer" in output
+
+        # Should mention AST chunking if code files are present
+        # (might not be relevant for the test data, but command should succeed)
+
+
 @pytest.mark.skipif(not os.environ.get("OPENAI_API_KEY"), reason="OpenAI API key not available")
 @pytest.mark.skipif(
     os.environ.get("CI") == "true", reason="Skip OpenAI tests in CI to avoid API costs"

tests/test_metadata_filtering.py

@@ -0,0 +1,365 @@
+"""
+Comprehensive tests for metadata filtering functionality.
+
+This module tests the MetadataFilterEngine class and its integration
+with the LEANN search system.
+"""
+
+import os
+
+# Import the modules we're testing
+import sys
+from unittest.mock import Mock, patch
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "../packages/leann-core/src"))
+
+from leann.api import PassageManager, SearchResult
+from leann.metadata_filter import MetadataFilterEngine
+
+
+class TestMetadataFilterEngine:
+    """Test suite for the MetadataFilterEngine class."""
+
+    def setup_method(self):
+        """Setup test fixtures."""
+        self.engine = MetadataFilterEngine()
+
+        # Sample search results for testing
+        self.sample_results = [
+            {
+                "id": "doc1",
+                "score": 0.95,
+                "text": "This is chapter 1 content",
+                "metadata": {
+                    "chapter": 1,
+                    "character": "Alice",
+                    "tags": ["adventure", "fantasy"],
+                    "word_count": 150,
+                    "is_published": True,
+                    "genre": "fiction",
+                },
+            },
+            {
+                "id": "doc2",
+                "score": 0.87,
+                "text": "This is chapter 3 content",
+                "metadata": {
+                    "chapter": 3,
+                    "character": "Bob",
+                    "tags": ["mystery", "thriller"],
+                    "word_count": 250,
+                    "is_published": True,
+                    "genre": "fiction",
+                },
+            },
+            {
+                "id": "doc3",
+                "score": 0.82,
+                "text": "This is chapter 5 content",
+                "metadata": {
+                    "chapter": 5,
+                    "character": "Alice",
+                    "tags": ["romance", "drama"],
+                    "word_count": 300,
+                    "is_published": False,
+                    "genre": "non-fiction",
+                },
+            },
+            {
+                "id": "doc4",
+                "score": 0.78,
+                "text": "This is chapter 10 content",
+                "metadata": {
+                    "chapter": 10,
+                    "character": "Charlie",
+                    "tags": ["action", "adventure"],
+                    "word_count": 400,
+                    "is_published": True,
+                    "genre": "fiction",
+                },
+            },
+        ]
+
+    def test_engine_initialization(self):
+        """Test that the filter engine initializes correctly."""
+        assert self.engine is not None
+        assert len(self.engine.operators) > 0
+        assert "==" in self.engine.operators
+        assert "contains" in self.engine.operators
+        assert "in" in self.engine.operators
+
+    def test_direct_instantiation(self):
+        """Test direct instantiation of the engine."""
+        engine = MetadataFilterEngine()
+        assert isinstance(engine, MetadataFilterEngine)
+
+    def test_no_filters_returns_all_results(self):
+        """Test that passing None or empty filters returns all results."""
+        # Test with None
+        result = self.engine.apply_filters(self.sample_results, None)
+        assert len(result) == len(self.sample_results)
+
+        # Test with empty dict
+        result = self.engine.apply_filters(self.sample_results, {})
+        assert len(result) == len(self.sample_results)
+
+    # Test comparison operators
+    def test_equals_filter(self):
+        """Test equals (==) filter."""
+        filters = {"chapter": {"==": 1}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 1
+        assert result[0]["id"] == "doc1"
+
+    def test_not_equals_filter(self):
+        """Test not equals (!=) filter."""
+        filters = {"genre": {"!=": "fiction"}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 1
+        assert result[0]["metadata"]["genre"] == "non-fiction"
+
+    def test_less_than_filter(self):
+        """Test less than (<) filter."""
+        filters = {"chapter": {"<": 5}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 2
+        chapters = [r["metadata"]["chapter"] for r in result]
+        assert all(ch < 5 for ch in chapters)
+
+    def test_less_than_or_equal_filter(self):
+        """Test less than or equal (<=) filter."""
+        filters = {"chapter": {"<=": 5}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 3
+        chapters = [r["metadata"]["chapter"] for r in result]
+        assert all(ch <= 5 for ch in chapters)
+
+    def test_greater_than_filter(self):
+        """Test greater than (>) filter."""
+        filters = {"word_count": {">": 200}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 3  # Documents with word_count 250, 300, 400
+        word_counts = [r["metadata"]["word_count"] for r in result]
+        assert all(wc > 200 for wc in word_counts)
+
+    def test_greater_than_or_equal_filter(self):
+        """Test greater than or equal (>=) filter."""
+        filters = {"word_count": {">=": 250}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 3
+        word_counts = [r["metadata"]["word_count"] for r in result]
+        assert all(wc >= 250 for wc in word_counts)
+
+    # Test membership operators
+    def test_in_filter(self):
+        """Test in filter."""
+        filters = {"character": {"in": ["Alice", "Bob"]}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 3
+        characters = [r["metadata"]["character"] for r in result]
+        assert all(ch in ["Alice", "Bob"] for ch in characters)
+
+    def test_not_in_filter(self):
+        """Test not_in filter."""
+        filters = {"character": {"not_in": ["Alice", "Bob"]}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 1
+        assert result[0]["metadata"]["character"] == "Charlie"
+
+    # Test string operators
+    def test_contains_filter(self):
+        """Test contains filter."""
+        filters = {"genre": {"contains": "fiction"}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 4  # Both "fiction" and "non-fiction"
+
+    def test_starts_with_filter(self):
+        """Test starts_with filter."""
+        filters = {"genre": {"starts_with": "non"}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 1
+        assert result[0]["metadata"]["genre"] == "non-fiction"
+
+    def test_ends_with_filter(self):
+        """Test ends_with filter."""
+        filters = {"text": {"ends_with": "content"}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 4  # All sample texts end with "content"
+
+    # Test boolean operators
+    def test_is_true_filter(self):
+        """Test is_true filter."""
+        filters = {"is_published": {"is_true": True}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 3
+        assert all(r["metadata"]["is_published"] for r in result)
+
+    def test_is_false_filter(self):
+        """Test is_false filter."""
+        filters = {"is_published": {"is_false": False}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 1
+        assert not result[0]["metadata"]["is_published"]
+
+    # Test compound filters (AND logic)
+    def test_compound_filters(self):
+        """Test multiple filters applied together (AND logic)."""
+        filters = {"genre": {"==": "fiction"}, "chapter": {"<=": 5}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 2
+        for r in result:
+            assert r["metadata"]["genre"] == "fiction"
+            assert r["metadata"]["chapter"] <= 5
+
+    def test_multiple_operators_same_field(self):
+        """Test multiple operators on the same field."""
+        filters = {"word_count": {">=": 200, "<=": 350}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 2
+        for r in result:
+            wc = r["metadata"]["word_count"]
+            assert 200 <= wc <= 350
+
+    # Test edge cases
+    def test_missing_field_fails_filter(self):
+        """Test that missing metadata fields fail filters."""
+        filters = {"nonexistent_field": {"==": "value"}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 0
+
+    def test_invalid_operator(self):
+        """Test that invalid operators are handled gracefully."""
+        filters = {"chapter": {"invalid_op": 1}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 0  # Should filter out all results
+
+    def test_type_coercion_numeric(self):
+        """Test numeric type coercion in comparisons."""
+        # Add a result with string chapter number
+        test_results = [
+            *self.sample_results,
+            {
+                "id": "doc5",
+                "score": 0.75,
+                "text": "String chapter test",
+                "metadata": {"chapter": "2", "genre": "test"},
+            },
+        ]
+
+        filters = {"chapter": {"<": 3}}
+        result = self.engine.apply_filters(test_results, filters)
+        # Should include doc1 (chapter=1) and doc5 (chapter="2")
+        assert len(result) == 2
+        ids = [r["id"] for r in result]
+        assert "doc1" in ids
+        assert "doc5" in ids
+
+    def test_list_membership_with_nested_tags(self):
+        """Test membership operations with list metadata."""
+        # Note: This tests the metadata structure, not list field filtering
+        # For list field filtering, we'd need to modify the test data
+        filters = {"character": {"in": ["Alice"]}}
+        result = self.engine.apply_filters(self.sample_results, filters)
+        assert len(result) == 2
+        assert all(r["metadata"]["character"] == "Alice" for r in result)
+
+    def test_empty_results_list(self):
+        """Test filtering on empty results list."""
+        filters = {"chapter": {"==": 1}}
+        result = self.engine.apply_filters([], filters)
+        assert len(result) == 0
+
+
+class TestPassageManagerFiltering:
+    """Test suite for PassageManager filtering integration."""
+
+    def setup_method(self):
+        """Setup test fixtures."""
+        # Mock the passage manager without actual file I/O
+        self.passage_manager = Mock(spec=PassageManager)
+        self.passage_manager.filter_engine = MetadataFilterEngine()
+
+        # Sample SearchResult objects
+        self.search_results = [
+            SearchResult(
+                id="doc1",
+                score=0.95,
+                text="Chapter 1 content",
+                metadata={"chapter": 1, "character": "Alice"},
+            ),
+            SearchResult(
+                id="doc2",
+                score=0.87,
+                text="Chapter 5 content",
+                metadata={"chapter": 5, "character": "Bob"},
+            ),
+            SearchResult(
+                id="doc3",
+                score=0.82,
+                text="Chapter 10 content",
+                metadata={"chapter": 10, "character": "Alice"},
+            ),
+        ]
+
+    def test_search_result_filtering(self):
+        """Test filtering SearchResult objects."""
+        # Create a real PassageManager instance just for the filtering method
+        # We'll mock the file operations
+        with patch("builtins.open"), patch("json.loads"), patch("pickle.load"):
+            pm = PassageManager([{"type": "jsonl", "path": "test.jsonl"}])
+
+            filters = {"chapter": {"<=": 5}}
+            result = pm.filter_search_results(self.search_results, filters)
+
+            assert len(result) == 2
+            chapters = [r.metadata["chapter"] for r in result]
+            assert all(ch <= 5 for ch in chapters)
+
+    def test_filter_search_results_no_filters(self):
+        """Test that None filters return all results."""
+        with patch("builtins.open"), patch("json.loads"), patch("pickle.load"):
+            pm = PassageManager([{"type": "jsonl", "path": "test.jsonl"}])
+
+            result = pm.filter_search_results(self.search_results, None)
+            assert len(result) == len(self.search_results)
+
+    def test_filter_maintains_search_result_type(self):
+        """Test that filtering returns SearchResult objects."""
+        with patch("builtins.open"), patch("json.loads"), patch("pickle.load"):
+            pm = PassageManager([{"type": "jsonl", "path": "test.jsonl"}])
+
+            filters = {"character": {"==": "Alice"}}
+            result = pm.filter_search_results(self.search_results, filters)
+
+            assert len(result) == 2
+            for r in result:
+                assert isinstance(r, SearchResult)
+                assert r.metadata["character"] == "Alice"
+
+
+# Integration tests would go here, but they require actual LEANN backend setup
+# These would test the full pipeline from LeannSearcher.search() with metadata_filters
+
+if __name__ == "__main__":
+    # Run basic smoke tests
+    engine = MetadataFilterEngine()
+
+    sample_data = [
+        {
+            "id": "test1",
+            "score": 0.9,
+            "text": "Test content",
+            "metadata": {"chapter": 1, "published": True},
+        }
+    ]
+
+    # Test basic filtering
+    result = engine.apply_filters(sample_data, {"chapter": {"==": 1}})
+    assert len(result) == 1
+    print("✅ Basic filtering test passed")
+
+    result = engine.apply_filters(sample_data, {"chapter": {"==": 2}})
+    assert len(result) == 0
+    print("✅ No match filtering test passed")
+
+    print("🎉 All smoke tests passed!")