Problem
I'm consistently seeing "✔️ Other..." for tool use in the chat log, with some exceptions. When I open the .chat file I can that these correspond to tool calls. Often (always?) the raw_output of the tool call is written as stringified JSON.
Proposed Solution
That tool call metadata that's hidden in the .chat file's JSON would be nice to display in a friendly way. Especially for tool calls that correspond to Jupyter commands, I'd love to see rich metadata: What command was called always visible as the "title" of the tool call, and the rest in a collapsible section: what args were passed in, what was the input/output? For example, jupyter-geoagents (cc @cboettig) can submit SQL queries against a geospatial dataset and inspecting those SQL queries is a critical element of a reasonable user experience for the LLM user.
Even if we don't have a way to nicely display info about the tool call, I would rather be able to click a little 🔺 to expand it in the chat log and see the raw data.
Additional context
❤️ you
Problem
I'm consistently seeing "✔️ Other..." for tool use in the chat log, with some exceptions. When I open the
.chatfile I can that these correspond to tool calls. Often (always?) theraw_outputof the tool call is written as stringified JSON.Proposed Solution
That tool call metadata that's hidden in the
.chatfile's JSON would be nice to display in a friendly way. Especially for tool calls that correspond to Jupyter commands, I'd love to see rich metadata: What command was called always visible as the "title" of the tool call, and the rest in a collapsible section: what args were passed in, what was the input/output? For example, jupyter-geoagents (cc @cboettig) can submit SQL queries against a geospatial dataset and inspecting those SQL queries is a critical element of a reasonable user experience for the LLM user.Even if we don't have a way to nicely display info about the tool call, I would rather be able to click a little 🔺 to expand it in the chat log and see the raw data.
Additional context
❤️ you