Download the catalog

Everything on this site is rendered from emitted JSON artifacts in the repository's dist/ directory. They are generated code, not hand-maintained data — regenerate them at any time from the public TOML sources.

Direct downloads

catalog.json — the full denormalized join: canonical models and every provider offering in a single document. The complete registry in one fetch.
providers.jsonandmodels.json — the normalized layers: providers, auth, endpoints, and plans in one; lab-owned canonical model facts in the other.

Per-entity files are served at stable paths underartifacts/providers/ and artifacts/models/ — for example/inference-providers/artifacts/providers/anthropic.json or/inference-providers/artifacts/models/openai-gpt-5-6.json.

Artifacts

File	Contents
`catalog.json`	The full join: canonical models and every provider offering in a single document. The complete registry in one fetch.
`providers.json`	Normalized provider layer — providers, auth, endpoints, plans.
`models.json`	Normalized model layer — lab-owned canonical model facts.
`providers/<id>.json`	One file per provider, at a stable path.
`models/<id>.json`	One file per model, at a stable path.

Provenance metadata

Every artifact — the catalog, the normalized layers, and each per-entity file — embeds two top-level fields: generated_at (when it was emitted) and source_commit (the git commit of the TOML data it was built from). Any fact can be traced back to the exact data revision.

For agents and crawlers

Machine discovery surfaces are published alongside the artifacts:robots.txt(open crawl policy + Content Signals),sitemap.xml, a RFC 9727 API catalog at.well-known/api-catalog, an ARD manifest at.well-known/ai-catalog.json, an agent-skills index at.well-known/agent-skills/index.jsonpointing at a ready-made SKILL.md for querying this registry, and consumer auth instructions at auth.md.

Generating them yourself

The artifacts are produced by pnpm emit, which reads the TOML data in data/models/ and data/providers/, validates it against the gating rules, and writes the typed artifacts todist/. No credentials or environment variables are involved — the output is fully reproducible:

git clone git@github.com:gregnazario/inference-providers.git && cd inference-providers
pnpm install
pnpm emit

The result is byte-for-byte the same data served by the links above and rendered by this site (timestamps aside). This site lives athttps://gregnazario.github.io/inference-providers/, rebuilt on every push to main by the GitHub Pages deploy workflow.