Open Recipe Archive

Building the largest open historical recipe corpora on GitHub

What This Repo Is

This repo is a public collection of over 54,000 historical recipes from more than thirty countries, built from public domain cookbooks and other copyright-safe sources

These recipes are available in Markdown and JSONL for humans and agents.

Every recipe includes:

• title
• collection slug
• culture tag
• ingredients and numbered directions in Markdown
• source author
• source title
• source year
• source URL
• license

Current Corpus

• 54,843 recipes
• 31 historical collections
• Markdown per recipe
• JSONL per collection

Current collections include:

• Argentina
• Australia
• Austria
• Brazil
• Canada
• Chile
• China
• Czechia
• Denmark
• Finland
• France
• Germany
• Greece
• Guatemala
• Hungary
• India
• Indonesia
• Italy
• Japan
• Mexico
• Netherlands
• Norway
• Philippines
• Poland
• Portugal
• Spain
• Sweden
• Turkey
• United Kingdom
• United States
• Jewish diaspora
• Levant
• Louisiana Creole
• West Indies

The project goal is to span every country and diaspora in the world with recipes that have attribution, rather than asking LLMs to infer cultural cooking from a muddied corpus. Some countries have deep pre-1931 cookbook shelves; others have sparse, fragmentary sources.

Repo Layout

collections/
  <collection-slug>/
    manifest.json
    recipes.jsonl
    recipes/
      <recipe-slug>.md

index/
  collections.json
  countries.json
  cultures.json
  summary.json

scripts/
  export_sqlite.py

Quick Start

Clone the repository:

git clone https://github.com/adambouhmad/open-recipe-archive.git
cd open-recipe-archive

Read a recipe in GitHub or locally:

open "collections/ye-old-american/recipes/tomato-soup.md"

Load one collection as JSONL:

python3 -c "import json; from pathlib import Path; p=Path('collections/ye-old-american/recipes.jsonl'); print(sum(1 for _ in p.open()))"

Inspect the global index:

python3 -c "import json; print(json.load(open('index/summary.json')))"

For Humans

Each recipe is a normal Markdown file with YAML frontmatter and a provenance section.

Example:

collections/ye-old-american/recipes/tomato-soup.md

For Scripts And Scrapers

Each collection has a recipes.jsonl file with one JSON object per recipe.

Example fields:

{
  "title": "Tomato Soup",
  "slug": "tomato-soup",
  "collection": "ye-old-american",
  "collection_name": "Ye Old American Recipe Book",
  "culture": "american-historical",
  "author": "Fannie Merritt Farmer",
  "body": "## Ingredients\n\n- 1 can tomatoes...",
  "source_title": "The Boston Cooking-School Cook Book",
  "source_url": "https://www.gutenberg.org/ebooks/65061",
  "source_year": "1896",
  "license": "public-domain",
  "tags": ["american-historical", "ye-old-american"],
  "export_date": "1896-01-01"
}

Read a collection in Python:

import json
from pathlib import Path

path = Path("collections/ye-old-american/recipes.jsonl")
recipes = [json.loads(line) for line in path.open(encoding="utf-8")]
print(len(recipes), recipes[0]["title"])

For Agents And RAG

Recommended ingestion paths:

• use recipes.jsonl for structured bulk loading
• use recipes/*.md when you want readable source documents
• use index/collections.json to discover available shelves first

Suggested strategy:

1. load index/collections.json
2. pick collections relevant to a region, culture, or period
3. ingest recipes.jsonl into your vector store or search index
4. retain source_title, source_url, source_year, and license in metadata

Build A Search Site

Treat this repo as content, not infrastructure.

Minimal approach:

1. load all recipes.jsonl files
2. index title, body, collection_name, culture, and tags
3. render the corresponding Markdown file or body text in your UI
4. show provenance fields with every result

Useful stack options:

• static site generator plus prebuilt JSON index
• SQLite FTS
• Meilisearch
• Elasticsearch or OpenSearch
• vector search plus metadata filters

Provenance And Licensing

All recipes are sourced from public-domain materials or clearly labeled permissive supplements. Every recipe carries source and license metadata.

Default license status for the exported historical corpus is public-domain unless a recipe is explicitly labeled otherwise.

Contributing

Good contributions include:

• provenance fixes
• source URL fixes
• license corrections
• collection metadata improvements
• export tooling improvements
• search/demo apps built on top of the corpus

License

See LICENSE.md.