Skip to main content
Mirrors builds a runnable digital twin of the backends an AI agent talks to, from the agent’s own traces. Your agent calls the twin instead of production; recorded conversations replay against it to measure how faithfully it answers; the numbers say which tool to fix and where.

​
Start here

  • Quickstart: traces in, a twin out, your agent calling it.
  • How it works: the words the product uses and what each measures.
  • Trace exports: what a good export contains, checked before you upload.

​
Ways in

​
Reference

A plain-text index for language-model crawlers is at llms.txt.