Agent Visualizer
The Agent Visualizer Snap provides a comprehensive dual-view interface combining an interactive Diagram and a detailed step-by-step Log. This dual visualization approach enables users to trace and analyze agent behavior precisely, offering enhanced transparency into complex AI agent operations and making it easier to understand how agents process and respond to user requests.
Summary of Agent Visualizer functionality
- Parses both single and multilayer agents: Agent Visualizer Snap can parse not only the single-layer agents but also agents with nested agents to generate corresponding diagrams and logs.
- Supports multiple LLM providers: Agent Visualizer Snap can visualize agents built by either OpenAI or AWS Bedrock, or a mixture of both providers.
- Visualizes agent performing parallel tool calling: Agent Visualizer Snap can identify and display when agents are calling multiple tools at the same step.
For example, if an Agent calls two tools simultaneously in Step 1, then the step numbers of the two tools are Step 1-1 and Step 1-2. The sub-steps to the two tools would be Step 1-1.1 and Step 1-2.1, respectively.
- Supports multiple requests: AgentVisualizer can accept multiple prompts given to the Agent and record the responses to each request. You can do this by adding a JSON Generator at the front of your driver agent pipeline and then by adding the multiple prompts inside the JSON Generator. This facilitates the development of assistants and chatbot agents because multiple prompts can be tested and validated in one action.