selfevolveagent / evoagentx /tools /database_faiss.py
iLOVE2D's picture
Upload 2846 files
5374a2d verified
"""
FAISS Database Toolkit for EvoAgentX
This module provides a comprehensive toolkit for interacting with FAISS vector databases
through the existing RAG engine and storage infrastructure. It includes tools for querying,
inserting, deleting, and managing vector data with semantic search capabilities.
The toolkit wraps the existing RAGEngine and StorageHandler to provide a unified interface
for vector database operations that can be easily used by agents.
Key Features:
- Automatic database path validation and creation
- Support for existing database detection and reuse
- Robust error handling for invalid paths
- Default configuration with sensible defaults
- Seamless integration with the RAG pipeline
- Automatic file path detection and content processing
- Support for multiple file formats (PDF, text, Markdown, etc.)
Usage:
# Using default configuration with automatic path handling
toolkit = FaissToolkit(db_path="./my_database.db")
# Using custom configuration
toolkit = FaissToolkit(
storage_config=custom_storage_config,
rag_config=custom_rag_config
)
# Insert documents including file paths
toolkit.get_tool("faiss_insert")(
documents=[
"This is some text content",
"./documents/report.pdf", # Will be automatically read and processed
"./data/notes.txt" # Will be automatically read and processed
]
)
"""
import os
import asyncio
import concurrent.futures
from typing import Dict, Any, List, Optional
from uuid import uuid4
from datetime import datetime
from pathlib import Path
from .tool import Tool, Toolkit
from .storage_handler import LocalStorageHandler
from ..core.module import BaseModule
from ..core.logging import logger
from ..rag.rag import RAGEngine
from ..rag.rag_config import RAGConfig
from ..rag.schema import Query, Document, Corpus, DocumentMetadata
from ..storages.base import StorageHandler
from ..storages.storages_config import StoreConfig
from .storage_handler import FileStorageHandler
def _ensure_database_path(db_path: str) -> str:
"""
Ensure the database path exists and is properly configured.
Args:
db_path (str): The database file path
Returns:
str: The validated and prepared database path
Raises:
ValueError: If the path is invalid or cannot be created
"""
if not db_path:
raise ValueError("Database path cannot be empty")
# Convert to Path object for easier manipulation and make it absolute
path = Path(db_path).resolve()
# Validate the path is not a directory
if path.exists() and path.is_dir():
raise ValueError(f"Database path points to a directory: {db_path}")
# Ensure the parent directory exists
try:
path.parent.mkdir(parents=True, exist_ok=True)
except Exception as e:
raise ValueError(f"Cannot create directory for database path {db_path}: {e}")
# Check if database file exists
if path.exists():
logger.info(f"Found existing database at: {db_path}")
# Validate it's a valid SQLite database by trying to connect
try:
import sqlite3
conn = sqlite3.connect(str(path))
conn.execute("SELECT name FROM sqlite_master WHERE type='table';")
conn.close()
logger.info("Database validation successful")
except Exception as e:
logger.warning(f"Database validation failed: {e}. Will create new database.")
# If validation fails, remove the corrupted file
try:
path.unlink()
except Exception as unlink_error:
logger.error(f"Failed to remove corrupted database file: {unlink_error}")
raise ValueError(f"Cannot remove corrupted database file: {unlink_error}")
else:
logger.info(f"Database not found at: {db_path}. Will create new database.")
return str(path)
def _create_default_storage_config(db_path: Optional[str] = None) -> StoreConfig:
"""
Create a default storage configuration with proper path handling.
Args:
db_path (str, optional): Custom database path
Returns:
StoreConfig: Configured storage configuration
"""
from ..storages.storages_config import StoreConfig, DBConfig, VectorStoreConfig
# Set default database path if not provided
if db_path is None:
db_path = "./faiss_db.sqlite"
# Ensure the database path is properly set up
validated_db_path = _ensure_database_path(db_path)
logger.info(f"Using validated database path: {validated_db_path}")
# Create index cache path (ensure it's absolute)
index_cache_path = str(Path(validated_db_path).parent.resolve() / "index_cache")
# Create storage configuration
storage_config = StoreConfig(
dbConfig=DBConfig(
db_name="sqlite",
path=validated_db_path
),
vectorConfig=VectorStoreConfig(
vector_name="faiss",
dimensions=1536,
index_type="flat_l2"
),
path=index_cache_path
)
# Ensure the index cache directory exists
Path(index_cache_path).mkdir(parents=True, exist_ok=True)
return storage_config
def _create_default_rag_config() -> RAGConfig:
"""
Create a default RAG configuration.
Returns:
RAGConfig: Configured RAG configuration
"""
from ..rag.rag_config import RAGConfig, EmbeddingConfig, ChunkerConfig
return RAGConfig(
embedding=EmbeddingConfig(
provider="openai",
model_name="text-embedding-ada-002"
),
chunker=ChunkerConfig(
chunk_size=500,
chunk_overlap=50
)
)
class FaissDatabase(BaseModule):
"""
A high-level interface for FAISS vector database operations.
This class wraps the RAGEngine and StorageHandler to provide a unified interface
for vector database operations including document ingestion, semantic search,
and corpus management.
Attributes:
rag_engine (RAGEngine): The RAG engine for document processing and retrieval
storage_handler (StorageHandler): The storage handler for persistence
default_corpus_id (str): Default corpus ID for operations
default_index_type (str): Default index type for vector operations
"""
def __init__(
self,
storage_config: StoreConfig,
rag_config: RAGConfig,
default_corpus_id: str = "default",
default_index_type: str = "vector",
storage_handler: StorageHandler = None,
file_handler: FileStorageHandler = None,
**kwargs
):
"""
Initialize the FAISS database.
Args:
storage_config (StoreConfig): Configuration for storage backends
rag_config (RAGConfig): Configuration for RAG pipeline
default_corpus_id (str): Default corpus ID for operations
default_index_type (str): Default index type for vector operations
storage_handler (StorageHandler, optional): Storage handler for file operations
**kwargs: Additional arguments for BaseModule
"""
super().__init__(**kwargs)
# Initialize storage handler for database operations
self.storage_handler = StorageHandler(storageConfig=storage_config)
# Initialize RAG engine
self.rag_engine = RAGEngine(config=rag_config, storage_handler=self.storage_handler)
# Initialize file storage handler for external file operations
# Use LocalStorageHandler as default if none provided
if storage_handler is None:
storage_handler = LocalStorageHandler(base_path="./workplace/storage")
self.file_storage_handler = storage_handler
# Set defaults
self.default_corpus_id = default_corpus_id
self.default_index_type = default_index_type
logger.info(f"Initialized FAISS database with corpus_id: {default_corpus_id}")
def query(
self,
query: str,
corpus_id: Optional[str] = None,
top_k: int = 5,
similarity_threshold: float = 0.0,
metadata_filters: Optional[Dict[str, Any]] = None
) -> Dict[str, Any]:
"""
Query the vector database with semantic search.
Args:
query (str): The query string to search for
corpus_id (str, optional): Corpus ID to search in
top_k (int): Number of top results to return
similarity_threshold (float): Minimum similarity threshold
metadata_filters (Dict[str, Any], optional): Metadata filters for search
Returns:
Dict[str, Any]: Search results with chunks and scores
"""
try:
# Check if we're already in an event loop
try:
asyncio.get_running_loop()
# We're in an event loop, use thread executor to avoid asyncio.run() conflict
logger.info("Detected running event loop, using thread executor for query")
with concurrent.futures.ThreadPoolExecutor() as executor:
future = executor.submit(self._query_sync, query, corpus_id, top_k, similarity_threshold, metadata_filters)
return future.result()
except RuntimeError:
# No event loop running, safe to use asyncio.run() in the RAG engine
logger.info("No event loop detected, using direct query processing")
return self._query_sync(query, corpus_id, top_k, similarity_threshold, metadata_filters)
except Exception as e:
logger.error(f"Query failed: {str(e)}")
return {"success": False, "error": str(e)}
def _query_sync(
self,
query: str,
corpus_id: Optional[str] = None,
top_k: int = 5,
similarity_threshold: float = 0.0,
metadata_filters: Optional[Dict[str, Any]] = None
) -> Dict[str, Any]:
"""
Synchronous version of query that can be safely called from a thread.
Args:
query (str): The query string to search for
corpus_id (str, optional): Corpus ID to search in
top_k (int): Number of top results to return
similarity_threshold (float): Minimum similarity threshold
metadata_filters (Dict[str, Any], optional): Metadata filters for search
Returns:
Dict[str, Any]: Search results with chunks and scores
"""
try:
corpus_id = corpus_id or self.default_corpus_id
# Check if corpus exists
if corpus_id not in self.rag_engine.indices:
logger.warning(f"Corpus {corpus_id} not found. Returning empty results.")
return {"success": True, "data": {
"query": query,
"corpus_id": corpus_id,
"total_results": 0,
"results": []
}}
# Create query object
query_obj = Query(
query_str=query,
top_k=top_k,
similarity_cutoff=similarity_threshold,
metadata_filters=metadata_filters
)
# Execute query
results = self.rag_engine.query(query_obj, corpus_id=corpus_id)
# Handle case where results or corpus might be None
if not results or not results.corpus:
logger.warning(f"Query returned no results for corpus {corpus_id}")
return {"success": True, "data": {
"query": query,
"corpus_id": corpus_id,
"total_results": 0,
"results": []
}}
# Format results
chunks = results.corpus.chunks if results.corpus.chunks else []
formatted_results = {
"query": query,
"corpus_id": corpus_id,
"total_results": len(chunks),
"results": []
}
for i, chunk in enumerate(chunks):
score = results.scores[i] if results.scores and i < len(results.scores) else 0.0
formatted_results["results"].append({
"chunk_id": chunk.chunk_id,
"content": chunk.text,
"score": score,
"metadata": chunk.metadata.model_dump() if chunk.metadata else {},
"doc_id": chunk.metadata.doc_id if chunk.metadata else None
})
logger.info(f"Query executed successfully. Found {len(formatted_results['results'])} results.")
return {"success": True, "data": formatted_results}
except Exception as e:
logger.error(f"Query failed: {str(e)}")
return {"success": False, "error": str(e)}
def _is_file_path(self, text: str) -> bool:
"""
Check if a string appears to be a file path.
Args:
text (str): The string to check
Returns:
bool: True if the string looks like a file path
"""
# Check if it contains path separators or file extensions
path_indicators = ['/', '\\', '.txt', '.pdf', '.md', '.doc', '.docx', '.csv', '.json', '.xml', '.html', '.htm']
return any(indicator in text for indicator in path_indicators) and os.path.exists(text)
def _process_file_path(self, file_path: str, doc_index: int, metadata: Optional[Dict[str, Any]] = None) -> List[Document]:
"""
Process a file path and return Document objects.
Args:
file_path (str): Path to the file
doc_index (int): Index of the document in the batch
metadata (Dict[str, Any], optional): Additional metadata
Returns:
List[Document]: List of Document objects created from the file
"""
try:
# Check if we're already in an event loop
try:
asyncio.get_running_loop()
# We're in an event loop, use thread executor to avoid asyncio.run() conflict
logger.info(f"Detected running event loop, using thread executor for {file_path}")
with concurrent.futures.ThreadPoolExecutor() as executor:
future = executor.submit(self._process_file_path_sync, file_path, doc_index, metadata)
return future.result()
except RuntimeError:
# No event loop running, safe to use asyncio.run() in the RAG engine
logger.info(f"No event loop detected, using direct processing for {file_path}")
return self._process_file_path_sync(file_path, doc_index, metadata)
except Exception as e:
logger.error(f"Failed to process file {file_path}: {str(e)}")
# Return a single document with error information
doc_metadata = metadata.copy() if metadata else {}
doc_metadata.update({
"doc_index": doc_index,
"insertion_time": datetime.now().isoformat(),
"source_file": file_path,
"error": str(e)
})
document_metadata = DocumentMetadata(**doc_metadata)
return [Document(
text=f"Error reading file {file_path}: {str(e)}",
metadata=document_metadata,
doc_id=str(uuid4())
)]
def _process_file_path_sync(self, file_path: str, doc_index: int, metadata: Optional[Dict[str, Any]] = None) -> List[Document]:
"""
Synchronous version of file processing that can be safely called from a thread.
Args:
file_path (str): Path to the file
doc_index (int): Index of the document in the batch
metadata (Dict[str, Any], optional): Additional metadata
Returns:
List[Document]: List of Document objects created from the file
"""
try:
# Use StorageHandler to read the file if available
if self.file_storage_handler:
result = self.file_storage_handler.read(file_path)
if result["success"]:
file_content = result["content"]
else:
raise Exception(f"Failed to read file: {result.get('error', 'Unknown error')}")
else:
# Fallback to direct file reading
with open(file_path, 'r', encoding='utf-8') as f:
file_content = f.read()
# Use RAG engine to process the content
temp_corpus_id = f"temp_file_{uuid4().hex[:8]}"
# Create a temporary document for processing
temp_doc = Document(
text=file_content,
metadata=DocumentMetadata(
source_file=file_path,
doc_index=doc_index,
insertion_time=datetime.now().isoformat()
),
doc_id=str(uuid4())
)
# Process the document through RAG engine
corpus = self.rag_engine.process_documents([temp_doc], corpus_id=temp_corpus_id)
# Convert chunks back to documents with proper metadata
documents = []
for chunk in corpus.chunks:
doc_metadata = metadata.copy() if metadata else {}
doc_metadata.update({
"doc_index": doc_index,
"insertion_time": datetime.now().isoformat(),
"source_file": file_path,
"original_chunk_id": chunk.chunk_id
})
# Create DocumentMetadata object
document_metadata = DocumentMetadata(**doc_metadata)
# Create Document object
documents.append(Document(
text=chunk.text,
metadata=document_metadata,
doc_id=chunk.chunk_id
))
# Clean up temporary corpus
self.rag_engine.clear(corpus_id=temp_corpus_id)
logger.info(f"Processed file {file_path} into {len(documents)} chunks")
return documents
except Exception as e:
logger.error(f"Failed to process file {file_path} in sync mode: {str(e)}")
# Return a single document with error information
doc_metadata = metadata.copy() if metadata else {}
doc_metadata.update({
"doc_index": doc_index,
"insertion_time": datetime.now().isoformat(),
"source_file": file_path,
"error": str(e)
})
document_metadata = DocumentMetadata(**doc_metadata)
return [Document(
text=f"Error reading file {file_path}: {str(e)}",
metadata=document_metadata,
doc_id=str(uuid4())
)]
def insert(
self,
documents: list,
corpus_id: Optional[str] = None,
metadata: Optional[Dict[str, Any]] = None,
batch_size: int = 100
) -> Dict[str, Any]:
"""
Insert documents into the vector database.
Args:
documents (Union[List[str], List[Dict[str, Any]]]): Documents to insert.
Strings can be either text content or file paths (if they look like paths and exist)
corpus_id (str, optional): Corpus ID to insert into
metadata (Dict[str, Any], optional): Additional metadata for all documents
batch_size (int): Batch size for processing
Returns:
Dict[str, Any]: Insertion results
"""
try:
corpus_id = corpus_id or self.default_corpus_id
# Process documents and create proper Document objects
processed_docs = []
file_paths_processed = []
for i, doc in enumerate(documents):
if isinstance(doc, str):
# Check if this string looks like a file path
if self._is_file_path(doc):
logger.info(f"Detected file path: {doc}")
file_docs = self._process_file_path(doc, i, metadata)
processed_docs.extend(file_docs)
file_paths_processed.append(doc)
else:
# Treat as regular text content
doc_metadata = metadata.copy() if metadata else {}
doc_metadata.update({
"doc_index": i,
"insertion_time": datetime.now().isoformat()
})
# Create DocumentMetadata object
document_metadata = DocumentMetadata(**doc_metadata)
# Create Document object
processed_docs.append(Document(
text=doc,
metadata=document_metadata,
doc_id=str(uuid4())
))
elif isinstance(doc, dict):
doc_metadata = metadata.copy() if metadata else {}
doc_metadata.update(doc.get("metadata", {}))
doc_metadata.update({
"doc_index": i,
"insertion_time": datetime.now().isoformat()
})
# Create DocumentMetadata object
document_metadata = DocumentMetadata(**doc_metadata)
# Create Document object
processed_docs.append(Document(
text=doc.get("text", ""),
metadata=document_metadata,
doc_id=doc.get("doc_id", str(uuid4()))
))
# Create corpus
corpus = Corpus(corpus_id=corpus_id)
# Process in batches
total_processed = 0
for i in range(0, len(processed_docs), batch_size):
batch = processed_docs[i:i+batch_size]
# Chunk the documents
batch_corpus = self.rag_engine.chunker.chunk(batch)
batch_corpus.corpus_id = corpus_id
# Add to index
self.rag_engine.add(self.default_index_type, batch_corpus, corpus_id=corpus_id)
# Add chunks to corpus
corpus.chunks.extend(batch_corpus.chunks)
total_processed += len(batch)
logger.info(f"Processed batch {i//batch_size + 1}, total processed: {total_processed}")
# Save index
self.rag_engine.save(corpus_id=corpus_id, index_type=self.default_index_type)
result = {
"corpus_id": corpus_id,
"documents_inserted": len(documents),
"chunks_created": len(corpus.chunks),
"total_processed": total_processed,
"file_paths_processed": file_paths_processed
}
logger.info(f"Successfully inserted {len(documents)} documents into corpus {corpus_id}")
if file_paths_processed:
logger.info(f"Processed {len(file_paths_processed)} file paths: {file_paths_processed}")
return {"success": True, "data": result}
except Exception as e:
logger.error(f"Insert failed: {str(e)}")
return {"success": False, "error": str(e)}
def delete(
self,
corpus_id: Optional[str] = None,
doc_ids: Optional[List[str]] = None,
metadata_filters: Optional[Dict[str, Any]] = None,
clear_all: bool = False
) -> Dict[str, Any]:
"""
Delete documents or chunks from the vector database.
Args:
corpus_id (str, optional): Corpus ID to delete from
doc_ids (List[str], optional): Document IDs to delete
metadata_filters (Dict[str, Any], optional): Metadata filters for deletion
clear_all (bool): Whether to clear the entire corpus
Returns:
Dict[str, Any]: Deletion results
"""
try:
corpus_id = corpus_id or self.default_corpus_id
if clear_all:
# Clear entire corpus
self.rag_engine.clear(corpus_id=corpus_id)
logger.info(f"Cleared entire corpus: {corpus_id}")
return {"success": True, "data": {"operation": "clear_all", "corpus_id": corpus_id}}
# Check if corpus exists before attempting deletion
if corpus_id not in self.rag_engine.indices:
logger.warning(f"Corpus {corpus_id} not found. Nothing to delete.")
return {"success": True, "data": {"operation": "selective_delete", "corpus_id": corpus_id, "message": "Corpus not found, nothing to delete"}}
# Only attempt deletion if there are specific criteria
if doc_ids or metadata_filters:
# Delete specific documents or by filters
self.rag_engine.delete(
corpus_id=corpus_id,
index_type=self.default_index_type,
node_ids=doc_ids,
metadata_filters=metadata_filters
)
result = {
"corpus_id": corpus_id,
"operation": "selective_delete",
"doc_ids": doc_ids,
"metadata_filters": metadata_filters
}
logger.info(f"Successfully deleted from corpus {corpus_id}")
return {"success": True, "data": result}
else:
# No deletion criteria provided
logger.warning(f"No deletion criteria provided for corpus {corpus_id}")
return {"success": True, "data": {"operation": "selective_delete", "corpus_id": corpus_id, "message": "No deletion criteria provided"}}
except Exception as e:
logger.error(f"Delete failed: {str(e)}")
return {"success": False, "error": str(e)}
def list_corpora(self) -> Dict[str, Any]:
"""
List all available corpora and their metadata.
Returns:
Dict[str, Any]: List of corpora with metadata
"""
try:
corpora = []
# Get corpus information from indices
for corpus_id, indices in self.rag_engine.indices.items():
corpus_info = {
"corpus_id": corpus_id,
"index_types": list(indices.keys()),
"retrievers": list(self.rag_engine.retrievers.get(corpus_id, {}).keys())
}
corpora.append(corpus_info)
return {"success": True, "data": {"corpora": corpora, "total": len(corpora)}}
except Exception as e:
logger.error(f"List corpora failed: {str(e)}")
return {"success": False, "error": str(e)}
def get_stats(self, corpus_id: Optional[str] = None) -> Dict[str, Any]:
"""
Get statistics about the database or a specific corpus.
Args:
corpus_id (str, optional): Corpus ID to get stats for
Returns:
Dict[str, Any]: Database statistics
"""
try:
if corpus_id:
# Stats for specific corpus
corpus_id = corpus_id or self.default_corpus_id
stats = {
"corpus_id": corpus_id,
"exists": corpus_id in self.rag_engine.indices,
"index_types": list(self.rag_engine.indices.get(corpus_id, {}).keys()),
"retrievers": list(self.rag_engine.retrievers.get(corpus_id, {}).keys())
}
# Try to get vector store stats if available
if corpus_id in self.rag_engine.indices:
vector_index = self.rag_engine.indices[corpus_id].get(self.default_index_type)
if vector_index and hasattr(vector_index, 'get_index'):
try:
index = vector_index.get_index()
if hasattr(index, 'vector_store'):
vector_store = index.vector_store
if hasattr(vector_store, 'faiss_index'):
stats["vector_count"] = vector_store.faiss_index.ntotal
stats["dimensions"] = vector_store.faiss_index.d
except Exception:
pass
return {"success": True, "data": stats}
else:
# Global stats
stats = {
"total_corpora": len(self.rag_engine.indices),
"corpora": list(self.rag_engine.indices.keys()),
"embedding_model": self.rag_engine.config.embedding.model_name,
"vector_store_type": self.rag_engine.storage_handler.storageConfig.vectorConfig.vector_name if self.rag_engine.storage_handler.storageConfig.vectorConfig else None
}
return {"success": True, "data": stats}
except Exception as e:
logger.error(f"Get stats failed: {str(e)}")
return {"success": False, "error": str(e)}
class FaissQueryTool(Tool):
"""Tool for querying the FAISS vector database with semantic search."""
name: str = "faiss_query"
description: str = "Query the FAISS vector database using semantic search to find relevant documents and chunks"
inputs: Dict[str, Dict[str, Any]] = {
"query": {
"type": "string",
"description": "The search query text to find semantically similar content"
},
"corpus_id": {
"type": "string",
"description": "Optional corpus ID to search in. If not provided, uses default corpus"
},
"top_k": {
"type": "integer",
"description": "Number of top results to return (default: 5)",
"default": 5
},
"similarity_threshold": {
"type": "number",
"description": "Minimum similarity threshold for results (default: 0.0)",
"default": 0.0
},
"metadata_filters": {
"type": "object",
"description": "Optional metadata filters to apply to search results (e.g., {'source': 'file1.txt'})"
}
}
required: Optional[List[str]] = ["query"]
def __init__(self, faiss_database: FaissDatabase = None):
super().__init__()
self.faiss_database = faiss_database
def __call__(
self,
query: str,
corpus_id: str = None,
top_k: int = 5,
similarity_threshold: float = 0.0,
metadata_filters: dict = None
) -> Dict[str, Any]:
"""Execute the query operation."""
return self.faiss_database.query(
query=query,
corpus_id=corpus_id,
top_k=top_k,
similarity_threshold=similarity_threshold,
metadata_filters=metadata_filters
)
class FaissInsertTool(Tool):
"""Tool for inserting documents into the FAISS vector database."""
name: str = "faiss_insert"
description: str = "Insert documents into the FAISS vector database with automatic chunking and embedding. Supports both text content and file paths - if a string looks like a file path and exists, it will automatically read and process the file content."
inputs: Dict[str, Dict[str, Any]] = {
"documents": {
"type": "array",
"description": "Array of documents to insert. Can be strings (text content or file paths), or objects with 'text', 'metadata', and 'doc_id' fields. If a string contains path separators or file extensions and the file exists, it will be treated as a file path and its content will be read and processed."
},
"corpus_id": {
"type": "string",
"description": "Optional corpus ID to insert into. If not provided, uses default corpus"
},
"metadata": {
"type": "object",
"description": "Optional metadata to add to all documents (e.g., {'source': 'file1.txt', 'category': 'research'})"
},
"batch_size": {
"type": "integer",
"description": "Batch size for processing documents (default: 100)",
"default": 100
}
}
required: Optional[List[str]] = ["documents"]
def __init__(self, faiss_database: FaissDatabase = None):
super().__init__()
self.faiss_database = faiss_database
def __call__(
self,
documents: list,
corpus_id: str = None,
metadata: dict = None,
batch_size: int = 100
) -> Dict[str, Any]:
"""Execute the insert operation."""
return self.faiss_database.insert(
documents=documents,
corpus_id=corpus_id,
metadata=metadata,
batch_size=batch_size
)
class FaissDeleteTool(Tool):
"""Tool for deleting documents from the FAISS vector database."""
name: str = "faiss_delete"
description: str = "Delete documents or chunks from the FAISS vector database. You can delete specific documents by ID, filter by metadata, or clear the entire corpus."
inputs: Dict[str, Dict[str, Any]] = {
"corpus_id": {
"type": "string",
"description": "Optional corpus ID to delete from. If not provided, uses default corpus"
},
"doc_ids": {
"type": "array",
"description": "Optional list of document IDs to delete. Use this to delete specific documents",
"items": {"type": "string"}
},
"metadata_filters": {
"type": "object",
"description": "Optional metadata filters to select documents for deletion (e.g., {'source': 'file1.txt'})"
},
"clear_all": {
"type": "boolean",
"description": "Set to true to clear the entire corpus. WARNING: This will delete all documents in the corpus",
"default": False
}
}
required: Optional[List[str]] = []
def __init__(self, faiss_database: FaissDatabase = None):
super().__init__()
self.faiss_database = faiss_database
def __call__(
self,
corpus_id: str = None,
doc_ids: list = None,
metadata_filters: dict = None,
clear_all: bool = False
) -> Dict[str, Any]:
"""Execute the delete operation."""
return self.faiss_database.delete(
corpus_id=corpus_id,
doc_ids=doc_ids,
metadata_filters=metadata_filters,
clear_all=clear_all
)
class FaissListTool(Tool):
"""Tool for listing available corpora in the FAISS vector database."""
name: str = "faiss_list"
description: str = "List all available corpora and their metadata in the FAISS vector database. This tool takes no parameters."
inputs: Dict[str, Dict[str, Any]] = {}
required: Optional[List[str]] = []
def __init__(self, faiss_database: FaissDatabase = None):
super().__init__()
self.faiss_database = faiss_database
def __call__(self) -> Dict[str, Any]:
"""Execute the list operation."""
return self.faiss_database.list_corpora()
class FaissStatsTool(Tool):
"""Tool for getting statistics about the FAISS vector database."""
name: str = "faiss_stats"
description: str = "Get statistics about the FAISS vector database or a specific corpus. Optionally provide a corpus_id to get stats for a specific corpus."
inputs: Dict[str, Dict[str, Any]] = {
"corpus_id": {
"type": "string",
"description": "Optional corpus ID to get statistics for. If not provided, returns global statistics"
}
}
required: Optional[List[str]] = []
def __init__(self, faiss_database: FaissDatabase = None):
super().__init__()
self.faiss_database = faiss_database
def __call__(self, corpus_id: str = None) -> Dict[str, Any]:
"""Execute the stats operation."""
return self.faiss_database.get_stats(corpus_id=corpus_id)
class FaissToolkit(Toolkit):
"""
Toolkit for FAISS vector database operations.
This toolkit provides a comprehensive set of tools for interacting with FAISS vector databases,
including semantic search, document insertion, deletion, and database management operations.
The toolkit integrates with the existing RAG engine and storage infrastructure to provide
a unified interface for vector database operations that can be easily used by agents.
"""
def __init__(
self,
name: str = "FaissToolkit",
storage_config: Optional[StoreConfig] = None,
rag_config: Optional[RAGConfig] = None,
default_corpus_id: str = "default",
default_index_type: str = "vector",
db_path: Optional[str] = None,
storage_handler: StorageHandler = None,
file_handler: FileStorageHandler = None,
**kwargs
):
"""
Initialize the FAISS toolkit.
Args:
name (str): Name of the toolkit
storage_config (StoreConfig, optional): Configuration for storage backends
rag_config (RAGConfig, optional): Configuration for RAG pipeline
default_corpus_id (str): Default corpus ID for operations
default_index_type (str): Default index type for vector operations
db_path (str, optional): Custom database path. If provided, will check for existing database or create new one
storage_handler (StorageHandler, optional): Storage handler for file operations
file_handler (FileStorageHandler, optional): File handler for file operations
**kwargs: Additional arguments
"""
# Use default configurations if not provided
if storage_config is None:
storage_config = _create_default_storage_config(db_path)
if rag_config is None:
rag_config = _create_default_rag_config()
# Create the shared FAISS database instance
faiss_database = FaissDatabase(
storage_config=storage_config,
rag_config=rag_config,
default_corpus_id=default_corpus_id,
default_index_type=default_index_type,
storage_handler=storage_handler,
file_handler=file_handler
)
# Create tools with the shared database instance
tools = [
FaissQueryTool(faiss_database),
FaissInsertTool(faiss_database),
FaissDeleteTool(faiss_database),
FaissListTool(faiss_database),
FaissStatsTool(faiss_database)
]
super().__init__(name=name, tools=tools, **kwargs)
# Set instance variables after super().__init__()
self.faiss_database = faiss_database
logger.info(f"Initialized {name} with {len(tools)} tools")
def get_database(self) -> FaissDatabase:
"""
Get the underlying FAISS database instance.
Returns:
FaissDatabase: The FAISS database instance
"""
return self.faiss_database