SAGE Tool Use

Tool retrieval and ranking algorithms for LLM agents

🎯 Overview

sage-tooluse provides a comprehensive suite of tool selection and ranking algorithms for LLM agents:

• Keyword Selector: Fast matching based on keyword overlap
• Embedding Selector: Semantic similarity using embeddings
• Hybrid Selector: Combines keyword and embedding approaches
• DFS-DT Selector: Decision tree-based tool selection
• Gorilla Adapter: Gorilla-style tool retrieval

📦 Installation

# Basic installation
pip install isage-tooluse

# With embedding support
pip install isage-tooluse[embedding]

# Development installation
pip install isage-tooluse[dev]

🚀 Quick Start

Keyword-based Tool Selection

from sage_libs.sage_tooluse import KeywordToolSelector

# Create selector
selector = KeywordToolSelector(tools=available_tools)

# Select tools for a query
selected = selector.select(
    query="Get current weather in New York",
    top_k=5
)

for tool in selected:
    print(f"Tool: {tool.name}, Score: {tool.score}")

Embedding-based Tool Selection

from sage_libs.sage_tooluse import EmbeddingToolSelector

# Create selector with embedding model
selector = EmbeddingToolSelector(
    tools=available_tools,
    model_name="sentence-transformers/all-MiniLM-L6-v2"
)

# Select tools based on semantic similarity
selected = selector.select(
    query="What's the weather like?",
    top_k=5
)

Hybrid Tool Selection

from sage_libs.sage_tooluse import HybridToolSelector

# Combine keyword and embedding approaches
selector = HybridToolSelector(
    tools=available_tools,
    keyword_weight=0.3,
    embedding_weight=0.7
)

selected = selector.select(query="...", top_k=5)

📚 Key Components

Selectors

• KeywordToolSelector: Fast keyword-based matching
• EmbeddingToolSelector: Semantic similarity using embeddings
• HybridToolSelector: Weighted combination of multiple selectors
• DFSDTToolSelector: Decision tree-based selection
• GorillaAdapter: Gorilla-style API-centric retrieval

Base Classes

• BaseToolSelector: Abstract base for all selectors
• ToolRegistry: Central registry for selector implementations

Schemas

• Tool: Tool representation with metadata
• ToolSelection: Selection result with scores
• SelectionContext: Context for tool selection

🏗️ Architecture

sage_libs.sage_tooluse/
├── __init__.py              # Public API exports
├── base.py                  # Base selector interface
├── keyword_selector.py      # Keyword-based selection
├── embedding_selector.py    # Embedding-based selection
├── hybrid_selector.py       # Hybrid selection strategy
├── dfsdt_selector.py        # Decision tree selector
├── gorilla_selector.py      # Gorilla-style retrieval
├── registry.py              # Selector registry
├── schemas.py               # Data schemas
└── retriever/               # Retrieval utilities

🎓 Use Cases

1. Agent Tool Selection: Help agents choose the right tools
2. API Discovery: Find relevant APIs for a task
3. Function Calling: Select appropriate functions for LLMs
4. Tool Recommendation: Recommend tools to users
5. Multi-step Planning: Select tool sequences for complex tasks

🔗 Integration with SAGE

This package is part of the SAGE ecosystem and can be used with SAGE agents:

# Standalone usage
from sage_libs.sage_tooluse import HybridToolSelector

# With SAGE (when available, through interface layer)
from sage.libs.tooluse import create_selector

selector = create_selector("hybrid", tools=available_tools)

📖 Documentation

• Repository: https://github.com/intellistream/sage-tooluse
• SAGE Documentation: https://intellistream.github.io/SAGE-Pub/
• Issues: https://github.com/intellistream/sage-tooluse/issues

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

MIT License - see LICENSE file for details.

🙏 Acknowledgments

Originally part of the sage-agentic package, now maintained as an independent repository for focused development and research.

📧 Contact

• Team: IntelliStream Team
• Email: shuhao_zhang@hust.edu.cn
• GitHub: https://github.com/intellistream

Part of the SAGE ecosystem - Stream Analytics for Generative AI Engines