https://promptfoo.dev/ logo
Storing custom provider "chain" as "prompt" under ...
# ❓|questions
d
I have a custom provider that creates a chained response with some logic around the LLM self-correcting its initial response if it detects an error. This seems to be working well, but I'm struggling logging the raw conversation that's taking place in the "prompt" section under "details" in the web viewer. Any suggestions or help implementing this?
i
From the custom provider I think if you return a
metadata
key, it will appear in the details view:
Copy code
return {
  "output": "...",
  "metadata": {
    "foo: "bar"
  }
}
that could be a quick way to plumb through the raw conversation
d
Cool, will try this.