How to Trace and Evaluate a Hermes Agent with NVIDIA NeMo Relay and Phoenix

NVIDIA NeMo Relay captures the execution path of a Hermes Agent, from the initial request and file-reading tool calls to the final response. This tutorial shows developers how to inspect the tools an AI agent used, the data it accessed, the tokens it consumed, and the cost of the run. Follow a controlled Hermes Agent workflow as it reads a README and Python file, evaluates the code, and returns an answer. The video then opens the trace in Phoenix to review the session, model and tool activity, and redacted data. NVIDIA NeMo Relay records this activity while a sanitizer removes personally identifiable information before the trace leaves the machine. Use the companion tutorial to configure NVIDIA NeMo Relay and start tracing your own agent workflows for debugging, evaluation, and reliability improvements. GitHub: https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/tools/hermes-relay-tracing/README.md Tech blog: https://developer.nvidia.com/blog/tracing-agent-harness-behavior-with-nvidia-nemo-relay GitHub: https://github.com/NVIDIA/NeMo-Relay Build: https://build.nvidia.com/