RagTools

RagTools is an Agentic Tools Manager designed to streamline the discovery, execution, and evaluation of various tools. It provides a structured way to manage tools via schemas, enabling efficient searching, ranking, and invocation of tools in both local and remote environments.

Features

• Tool Discovery: Store and search tool schemas with ranked relevance to queries.
• Tool Execution: Invoke tools with specific arguments and return responses.
• Flexible Implementations: Support for HTTP calls, local function calls, and streaming inputs/outputs.
• Hierarchical Relationships: Tools can have parent-child dependencies and execution order constraints.
• Performance Tracking: Measure tool effectiveness via test sets and real-world evaluations.
• Dynamic Enabling/Disabling: Tools can be launched, unlaunched, or disabled dynamically.
• Authentication Handling: Seamless integration with OAuth-protected or paywalled content.
• Intelligent Tool Matching: Tools specify trigger conditions to improve discovery.
• Multi-language SDKs: Enables interaction with the tool manager across different programming languages.
• Parallel Execution: Multiple tools can run simultaneously for efficient processing.
• Auto-Evaluation & Ranking: Tools are ranked based on usage quality, reliability, and success rates.
• Versioning & Attributes: Tools and their implementations are versioned with machine-accessible attributes.

How It Works

1. Define Tools: Tools are described using YAML manifests with metadata, permissions, and capabilities.
2. Discover Tools: A fast /search API classifies and ranks tools based on query relevance.
3. Execute Tools: Tools can be invoked with arguments, including support for streaming responses.
4. Evaluate & Improve: Automatic and human-in-the-loop evaluations enhance tool ranking over time.
5. Manage Implementations: A tool can have multiple local or remote implementations, chosen dynamically.

Use Cases

• AI-powered agents executing external tools
• Workflow automation with dynamically invoked functions
• Search and ranking of APIs and services for query-based execution
• Evaluating and managing tool effectiveness in production environments

Roadmap

• RPC-based communication for optimized tool execution
• Expanded SDK support across multiple languages
• Enhanced session management for stateful tool interactions
• Improved AI-driven tool schema generation

Contributing

We welcome contributions! Please submit issues, feature requests, or pull requests to help improve RagTools.

License

RagTools is licensed under MIT License.