Usage

Pre-LLM

For immediately before the prompt is sent to the model

️✨ Spotlight
Emphasize important spans within your prompts to steer attention

Pre-tool

For immediately before a tool is invoked by an agent

️💡 Refraction
Validate and repair tool call sequences using classical AI planning ️⚡️ SPARC
Evaluate the static and semantic elements tool calls before execution

Post-tool

For immediately after a tool response is received

📚 JSON Processor
Dynamically extract data from large and complex JSON objects 📄🛠️ RAG Repair
Use domain documentation to repair tool call errors ️💬 Silent Review
Useful for correcting tool calls where errors are not explicit in the response

Pre-response

For immediately before the agent returns a response to the user

🛡️ Policy Guard
Guard and enforce policies in LLM responses

Build-Time

During building of tools and agents

📚 Test Case Generation
Generates the test case values and NL Utterances ️💡 Tool Enrichment
Python tool enrichment using the metadata information for improved tool calling and tool input formation ✅ Tool Validation
Validates Python tools by running test cases with a ReAct agent bound to the required tools.