# Veil

> The system of record for the AI industry, in Persian and English. An
> entity-first knowledge graph: models, companies, papers, benchmarks and
> vulnerabilities, where every fact carries a source URL and a source date.

Veil is free, has no advertising and no paywall. All content is open to crawlers.

## Conventions

- Persian is served at the root; English is served under /en.
- Every entity page has ONE plain-markdown mirror, at the unprefixed path plus
  `.md` — e.g. `/models/gemini.md`. It is language-neutral: the facts, their
  source URLs and their dates are the same in both locales, and it names both
  canonical pages in its footer.
- A mirror exists only while its page clears the publication floor. A thin
  entity's mirror returns 404, exactly as its page returns noindex.
- Every rendered fact traces to a source URL and date. Facts without provenance
  are not published.

## Key surfaces

- [Models](https://veil.ir/models): model specifications, licences, availability
- [Companies](https://veil.ir/companies): labs and providers
- [Papers](https://veil.ir/papers): research, linked to the models it produced
- [Benchmarks](https://veil.ir/benchmarks): results and methodology
- [Security](https://veil.ir/security): vulnerabilities and techniques in the AI stack
- [Lexicon](https://veil.ir/lexicon): Persian/English AI terminology
- [Persian tokenizer efficiency](https://veil.ir/leaderboard): original
  measurement — how many tokens each model spends on the same meaning in
  Persian versus English. Machine-readable at
  `https://veil.ir/leaderboard.md`, `.csv` and `.json`, with the
  paired corpus at `https://veil.ir/leaderboard.corpus.json`.