# FAISS Search Tool

The FAISS Search Tool enables efficient vector similarity search using Facebook AI Similarity Search (FAISS).

## Usage

```python
from typing import List, Dict, Any
from crewai import Agent
from crewai.tools import FAISSSearchTool

# Initialize tool
search_tool = FAISSSearchTool(
    index_type: str = "L2",  # or "IP" for inner product
    dimension: int = 384,    # Match your embedder's dimension
    embedder_config: Dict[str, Any] = {
        "provider": "fastembed",
        "model": "BAAI/bge-small-en-v1.5"
    }
)

# Add documents (with error handling)
try:
    search_tool.add_texts([
        "Document 1 content",
        "Document 2 content",
        # ...
    ])
except ValueError as e:
    print(f"Failed to add documents: {e}")

# Add large document sets efficiently
try:
    search_tool.add_texts_batch(
        texts=["Doc 1", "Doc 2", ...],  # Large list of documents
        batch_size=1000  # Process in batches to manage memory
    )
except ValueError as e:
    print(f"Failed to add documents in batch: {e}")

# Search with error handling
try:
    results = search_tool.run(
        query="search query",
        k=3,  # Number of results
        score_threshold=0.6  # Minimum similarity score
    )
    for result in results:
        print(f"Text: {result['text']}, Score: {result['score']}")
except ValueError as e:
    print(f"Search failed: {e}")

# Create agent with tool
agent = Agent(
    role="researcher",
    goal="Find relevant information",
    tools=[search_tool]
)
```

## Configuration

| Parameter | Type | Description |
|-----------|------|-------------|
| index_type | str | FAISS index type ("L2" or "IP") |
| dimension | int | Embedding dimension |
| embedder_config | dict | Embedder configuration |

## Parameters

### index_type
- `"L2"`: Euclidean distance (default)
- `"IP"`: Inner product similarity

### dimension
Default is 384, which matches the BAAI/bge-small-en-v1.5 model. Adjust this to match your chosen embedder model's output dimension.

### embedder_config
Configuration for the embedding model. Supports all CrewAI embedder providers:
- fastembed (default)
- openai
- google
- ollama

## Error Handling

The tool includes comprehensive error handling:

```python
# Invalid index type
try:
    tool = FAISSSearchTool(index_type="INVALID")
except ValueError as e:
    print(f"Invalid index type: {e}")

# Empty query
try:
    results = tool.run(query="")
except ValueError as e:
    print(f"Invalid query: {e}")  # "Query cannot be empty"

# Invalid k value
try:
    results = tool.run(query="test", k=0)
except ValueError as e:
    print(f"Invalid k: {e}")  # "k must be positive"

# Invalid score threshold
try:
    results = tool.run(query="test", score_threshold=1.5)
except ValueError as e:
    print(f"Invalid threshold: {e}")  # "score_threshold must be between 0 and 1"
```

## Performance Considerations

### Memory Management
For large document sets, use batch processing to manage memory efficiently:
```python
# Process documents in batches
tool.add_texts_batch(texts=large_document_list, batch_size=1000)
```

### Index Management
Monitor and manage index size:
```python
# Check index size
print(f"Current index size: {tool.index_size}")

# Check if index is empty
if tool.is_empty:
    print("Index is empty")

# Clear index if needed
tool.clear_index()
```

### Performance Metrics
The tool is optimized for performance:
- Search operations typically complete within 1 second for indices up to 1000 documents
- Batch processing helps manage memory for large document sets
- Input sanitization ensures query safety without significant overhead