DatafetchPro

AI Metrics and Evaluation

three.dev

Finds failures in live AI agent traffic and ranks them by severity for fixing

Votes
117
Followers
195
Comments
14
Best daily rank
#10
three.dev website preview

About three.dev

What three.dev does

  • Routes live LLM traffic through a proxy so responses reach the app and an evaluator at the same time.
  • Scores requests and groups failures into named failure modes ranked by severity.
  • Works at the HTTP level, so teams keep their existing OpenAI, Anthropic, or Google SDKs and frameworks such as LangChain.

Setup and fixing

  • Setup starts with one prompt pasted into Claude Code, Codex, or Cursor.
  • A setup skill finds LLM calls, groups changes by feature, generates use cases, and wires up the proxy.
  • The coding agent receives what it needs to write the fix, and turning the proxy off means reverting the base URL and header changes.

Also offered

  • Early access experiments test a change on recorded traffic, then on a share of live sessions.
  • Early access specialized language models are trained on a team's own traffic for one task.

Launch history1

  1. three.dev

    Featured

    Find out how your AI agent fails. Before your users do.

    See the whole day — 12 November 2026
    117votes