---
license: cc-by-4.0
language:
- en
pretty_name: Protologue
size_categories:
- n<1K
task_categories:
- text-retrieval
- question-answering
tags:
- prompt-engineering
- llm
- glossary
- taxonomy
- definitions
- ai-agents
- retrieval-augmented-generation
configs:
- config_name: terms
  data_files: data/terms.jsonl
  default: true
- config_name: categories
  data_files: data/categories.jsonl
---

# Protologue: a taxonomy of prompting and LLM techniques

[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.23304106.svg)](https://doi.org/10.5281/zenodo.23304106)

[Protologue](https://protologue.com) is an openly licensed reference for the vocabulary of prompting and large language model techniques. Each entry has a permanent identifier, a standalone one-to-two-sentence definition, synonyms, broader, narrower and related links, and the primary source that introduced the technique.

This dataset mirrors version 1.2.2 (2026-10-14): **215 terms** in **11 branches**, including a Benchmarks branch, citing **214 primary sources**. Every arXiv citation was checked against the paper for exact title and first author, and every documentation link was checked.

## Configurations

- `terms` (default): one row per term.
- `categories`: the 11 branches and the terms in each.

## Fields (`terms`)

| Field | Description |
|---|---|
| `id` | Permanent identifier, e.g. `PTL-0024`. Never reused. |
| `slug` | URL slug; links in `broader`, `narrower` and `related` use slugs. |
| `term` | Preferred name. |
| `aliases` | Synonyms and alternate names. |
| `category` | Branch of the taxonomy. |
| `definition` | Standalone definition, quotable on its own. |
| `description` | Context, limitations and relations. |
| `example` | Illustrative prompt, where useful. |
| `broader`, `narrower`, `related` | Links to other entries (SKOS semantics). |
| `introduced` | Year the technique was introduced. |
| `sources` | Primary sources: title, authors, year, url. The first is the type source. |
| `url` | Canonical page on protologue.com. |
| `citation` | Ready-made citation for the entry. |

## Usage

```python
from datasets import load_dataset
terms = load_dataset("franklinpolymathic/protologue", "terms", split="train")
print(terms[0]["term"], "-", terms[0]["definition"])
```

The same data is available live as JSON, CSV and SKOS at <https://protologue.com/data/>, and through a free MCP server for AI agents at `https://mcp.protologue.com/mcp` (see <https://protologue.com/mcp/>). It is also mirrored on [Kaggle](https://www.kaggle.com/datasets/protologue/protologue).

## License and citation

Text and data are licensed under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). When you use a definition, cite the entry's `citation` and link to its `url`. To cite the dataset:

```
Protologue. (2026). Protologue: A Taxonomy of Prompting and LLM Techniques (v1.2.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.23304106
```

The DOI is archived on [Zenodo](https://doi.org/10.5281/zenodo.23304106) and always resolves to the latest release.

Corrections and suggestions: open an issue at <https://github.com/protologue-org/protologue>.