AgentSea.Evaluate.Metric.LLMJudge (agentsea_evaluate v0.1.0)

Copy Markdown

Uses an LLM to score an output against a rubric — "LLM-as-judge". Runs over any AgentSea.Provider (so it can go through the gateway).

Options:

  • :provider{module, opts} (required)
  • :model — model id (or in the provider opts)
  • :rubric — grading instructions (default: relevance/correctness)
  • :threshold — pass cutoff in [0,1] (default 0.5)