Skip to content

Inspection UI

Christopher David edited this page Dec 4, 2023 · 7 revisions

We want full transparency into what agents do - and why.

This should be in an easily accessible web UI allowing for public inspection and ease of developer iteration.

Types of agent tasks

  • Auditing a GitHub repo
  • Making code commits / PRs

Types of agent steps

  • Make an LLM call
  • Query a vector database
  • Make an API call (GitHub)

Every time an agent attempts a task step, we want an easy way to see...

  • Description
  • Input
  • Output
  • What model was used (if LLM)
  • What compute was used
  • How many tokens / how much did it cost
  • How long did it take
  • What APIs were used
  • What agent/userid

UX design

inspectionui

Clone this wiki locally