I just started playing around with Fleet, specifically, the agent builder. I did not use a template, and wanted to test out the manual build within the interface. I created a painfully simple agent and included API keys for web search and OpenAI. The interface is intended to be chat via the web UI.
It’s still not super obvious to me when I am attempting to use the agent versus “talk to update”, but my only goal is test the agent and see the trace data in Langsmith. To that end, I see an entry called langsmith-agent-builder in the UI
and when I review the latest run, I see token tracking, but the input and output are completely blank.
What am I missing? The chat was initiated from within the web UI and returned data.
And for what it is worth, every attempt (single turn) appears to burn 2 “Agent Runs”.
Perhaps I am missing many obvious things, but I was expecting to quickly “manually config” an agent, test it (not update it with natural language) and see that data not only in the chat UI within Fleet, but the trace data in Langsmith.
(Removed additional contextual screenshots as only one is allowed)
