{
  "id": "protegi",
  "code": "PTL-0151",
  "term": "ProTeGi",
  "aliases": [
    "Prompt Optimization with Textual Gradients",
    "textual gradients"
  ],
  "category": "optimization",
  "definition": "ProTeGi (Prompt Optimization with Textual Gradients) improves a prompt by having a language model describe, in natural language, what is wrong with it based on the errors it produces, then edit the prompt to fix those problems, keeping the best candidates with beam search.",
  "description": "The natural-language critique plays the role of a gradient, and editing the prompt plays the role of a step in the opposite direction. A bandit-style selection procedure limits how many evaluations are needed to choose among candidates.",
  "example": null,
  "broader": [
    "prompt-optimization"
  ],
  "narrower": [],
  "related": [
    "textgrad",
    "automatic-prompt-engineer",
    "opro"
  ],
  "introduced": 2023,
  "added": "2026-10-10",
  "sources": [
    {
      "title": "Automatic Prompt Optimization with \"Gradient Descent\" and Beam Search",
      "authors": "Pryzant et al.",
      "year": 2023,
      "url": "https://arxiv.org/abs/2305.03495"
    }
  ],
  "url": "https://protologue.com/t/protegi/",
  "citation": "Protologue. (2026). ProTeGi. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0151). https://protologue.com/t/protegi/"
}