SAGE Agentic Framework

Independent package for agentic AI capabilities: tool selection, planning, workflows, and agent coordination

🎯 Overview

sage-agentic provides a comprehensive framework for building agentic AI systems with:

• Tool Selection: Multiple strategies (keyword, embedding, hybrid, DFS-DT, Gorilla)
• Planning Algorithms: ReAct, Tree of Thoughts (ToT), hierarchical planning
• Workflow Management: Workflow orchestration and optimization
• Agent Coordination: Multi-agent collaboration and registry
• SIAS: Sample-Importance-Aware Selection for tool/trajectory curation
• Reasoning: Advanced reasoning capabilities and timing decisions

📦 Installation

# Basic installation
pip install isage-agentic

# With LLM support
pip install isage-agentic[llm]

# Development installation
pip install isage-agentic[dev]

�� Quick Start

Tool Selection

from sage_agentic.agents.action.tool_selection import HybridToolSelector

# Create selector
selector = HybridToolSelector(embedder=your_embedder)

# Select tools
tools = selector.select(
    query="search for research papers",
    available_tools=all_tools,
    k=3
)

Planning

from sage_agentic.agents.planning import ReActPlanner

# Create planner
planner = ReActPlanner(llm=your_llm_client)

# Generate plan
plan = planner.plan(
    task="Analyze this document and summarize key findings",
    context={"document": doc_content}
)

Workflow Management

from sage_agentic.workflow import WorkflowEngine

# Create workflow
workflow = WorkflowEngine()

# Register and execute workflows
workflow.register("data_pipeline", pipeline_config)
result = workflow.execute("data_pipeline", inputs=data)

📚 Key Components

1. Planning (agents/planning/)

Planning algorithms and strategies:

• ToT (Tree of Thoughts): Multi-path reasoning with backtracking
• ReAct: Reasoning + Acting interleaved execution
• Hierarchical Planner: Hierarchical task decomposition
• Dependency Graph: Task dependency management
• Timing Decider: Execution timing optimization

2. Tool Selection (agents/action/tool_selection/)

Tool selection strategies:

• Keyword Selector: Rule-based keyword matching
• Embedding Selector: Semantic similarity-based selection
• Hybrid Selector: Combined keyword + embedding approach
• DFS-DT Selector: Decision tree-based selection
• Gorilla Adapter: Gorilla-style tool retrieval

3. SIAS (sias/)

Sample-Importance-Aware Selection for:

• Tool selection optimization
• Trajectory curation
• Continual learning with core-set selection

4. Evaluation (eval/)

Agent evaluation capabilities:

• Metrics tracking
• Determinism testing
• Telemetry and monitoring

5. Interfaces & Registry (interface/, registry/)

Unified interfaces and registration system for:

• Planners
• Tool selectors
• Workflows
• Agents

🔧 Architecture

sage_agentic/
├── agents/                 # Agent implementations
│   ├── action/            # Action and tool selection
│   ├── planning/          # Planning algorithms
│   └── intent/            # Intent detection
├── workflow/              # Workflow orchestration
├── sias/                  # Sample-Importance-Aware Selection
├── reasoning/             # Reasoning capabilities
├── eval/                  # Evaluation tools
├── interface/             # Protocol definitions
├── interface/             # Protocols, registries, schemas
└── registry/              # Component registry

🎓 Use Cases

1. Multi-Agent Systems: Build coordinated multi-agent workflows
2. Tool-Augmented LLMs: Select and use external tools intelligently
3. Hierarchical Planning: Decompose complex tasks into subtasks
4. Adaptive Systems: Use SIAS for intelligent sample selection
5. Research: Experiment with different planning and selection strategies

🔗 Integration with SAGE

This package is part of the SAGE ecosystem but can be used independently:

# Standalone usage
from sage_agentic import ReActPlanner, HybridToolSelector

# With SAGE (if installed)
from sage.libs.agentic import ReActPlanner  # Compatibility layer

📖 Documentation

• Repository: https://github.com/intellistream/sage-agentic
• SAGE Documentation: https://intellistream.github.io/SAGE-Pub/
• Issues: https://github.com/intellistream/sage-agentic/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 framework, now maintained as an independent package for broader community use.

📧 Contact

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