185 datasets · 27 categories · 3 formats · CC0
Give your agents the facts. Skip the tokens.
185 reference datasets for the facts your agents keep looking up. Read them as tables, or fetch JSON, YAML, or TOON from one URL — free, CC0, and no API keys.
Just want the API? Read the reference →
- Datasets
- 185
- Fewer bytes in TOON
- −40%
- Signup / keys
- Zero
Loading live response…The index
27 shelves, 185 datasets.
Pick a shelf — every dataset opens in JSON, YAML, or TOON. Type / to search.
The shape of the catalog
185 datasets, one cube each.
Every cube is one dataset. Every stack is one category, and its height is how many datasets that category holds — so the silhouette below is the catalog itself. Drag to turn it, hover to identify a cube, select it to open that dataset's page.
Loading the catalog… Open page →
Fig. 1 — Cube height is the number of datasets in that category. Colour identifies the category and carries no other meaning.
Every category
- Programming Languages10
- HTTP Status Codes10
- Data Structures10
- Algorithms7
- AI & ML Terms10
- Web Technologies9
- Unix Commands9
- Design Patterns7
- Database Types9
- Operating Systems5
- Networking7
- Cybersecurity6
- Cloud Computing7
- DevOps6
- Mathematics7
- Physics6
- Chemistry5
- Biology6
- Astronomy5
- History6
- Geography5
- Economics5
- Philosophy6
- Literature6
- Linguistics5
- Psychology6
- Law5
Prefer a list you can read straight through? Open the full catalog — all 185 datasets, one page each →
TOON Studio
See the saving on your own data.
Paste JSON and this page encodes it in your browser using the same encoder the API uses — the two are held to byte-identical output. Every number below is measured on the bytes in front of you, not on a benchmark.
- —
- —
- —
- —
Token counts are estimates from a character heuristic,
not a tokenizer, so treat them as a guide rather than a
billing figure. Byte counts are exact. The encoder in
js/toon.js is held to the server's output by
scripts/tests/toon-parity.php, which compares
all 185 datasets on every run.
What people build
Four jobs this catalog does well.
Reference data is not a search problem. These are the jobs where an exact lookup beats a guess.
Give an agent a stable facts table
Put the twelve rows an agent actually needs into the system prompt, in TOON, and it stops inventing versions and field names.
Read the template → CostCut the bytes you send every request
Switch the encoding, keep the data. We measure the same payload in both formats instead of quoting a marketing percentage.
See the measurement → CorrectnessCheck an answer instead of trusting it
Compare the model's output against a known table and fail loudly when a value does not exist.
Read the method → ArchitectureSkip the index when you need one answer
For questions with a single correct answer, an exact lookup is cheaper and more reliable than a nearest-neighbour search.
Read the argument →Choose a format
One dataset, three encodings.
The same bytes live on the server once. What changes is what they cost in a prompt and how they read in an editor.
| Property | JSON | YAML | TOON |
|---|---|---|---|
| Readable by a human | Yes, with punctuation noise | The easiest to read | Yes — uniform rows become a table |
| Parser required | Built into every language | A YAML library | A line parser — this site publishes a 50-line Python one |
| Nested data | Native | Native | Native, plus a tabular form for uniform arrays |
| Size for the whole index | 57 KB | 39 KB | 35 KB (−40%) |
| Cost in a prompt | Highest | Lower than JSON | Lowest measured here |
| Reach for it when | Services exchange data and every language must parse it | A human edits the file | The same data goes into a prompt on every request |
Every dataset page shows its own sizes in all three formats. Pick one and compare →
About
Reference data should be boring in the best way: stable, cited, and cheap to put into a prompt. YJTOON keeps 185 public-domain datasets that way — stored once as JSON, served as JSON, YAML, or TOON, no authentication, no keys.
The book
Open Data for AGI
“Open datasets are not a concession to idealism. They are a practical requirement for building systems that are robust, fair, and trustworthy.”
- A field guide to the major open datasets
- The legal and economic forces shaping data policy
- Recommendations engineers, researchers, and policymakers can act on
Start free: Open data and attribution · What CC0 actually means · TOON token savings, measured
Notes
Latest writing →
Questions
Answers, briefly.
Is YJTOON free to use?
Yes. All 185 datasets are released under CC0, which places them in the public domain. There is no signup, no API key, and no paid tier. The only limit is a rate limit of 120 requests per minute per IP.
What is TOON?
TOON is a compact text encoding for structured data, aimed at prompts rather than at storage. A uniform array of objects becomes a tabular block with a single header row, and scalars are quoted only when they have to be. Across all 185 datasets here it produces about 40% fewer bytes than pretty-printed JSON for the same data.
Do I need an API key?
No. Every endpoint is anonymous. Requests are rate limited by a hash of the IP address, and raw IP addresses are never stored.
Can I use the data in a commercial product?
Yes. CC0 waives copyright entirely, so there is no attribution requirement and no restriction on commercial use. Attribution is welcome, but it is never required.
How is this different from a vector database?
A vector index returns passages that look similar to a query. Reference data answers questions that have one correct answer. For those, an exact lookup against a known dataset is cheaper and more reliable than a nearest-neighbour search.
How should I cite a dataset?
Every dataset page includes a suggested citation and a permanent URL. Attribution is not required under CC0, but it helps other people find the source.